centre international d’Études supÉrieures en sciences ... · chacón, ing. josé zúñiga, msc....
TRANSCRIPT
CENTRE INTERNATIONAL D’ÉTUDES SUPÉRIEURES
EN SCIENCES AGRONOMIQUES ! MONTPELLIER SUPAGRO
ÉCOLE DOCTORALE
Systèmes Intégrés en Biologie, Agronomie, Géosciences,
Hydrosciences, Environnement (SIBAGHE)
Doctorat
Eaux Continentales et Société
Federico GÓMEZ DELGADO
!"#$%&&'& ()*"#+#,-.'%&/ 0$#!()&-#+#,-.'%& %1 *% 1"23&4%"1 *% &0*-5%31& *23& '3 62&&-3 2,"#4#"%&1-%" $240-%"
$#56-3231 *%& 2!!"#$(%& %7!0"-5%312+%& %1 *% 5#*0+-&21-#3 !#'" +%& 6%&#-3& *%& &%"8-$%& ()*"#+#,-.'%& %38-"#33%5%312'7
()*"#+#,-$2+/ %$#!()&-#+#,-$2+ 23* &%*-5%31 !"#$%&&%& -3 2 $#44%% 2,"#4#"%&1") 62&-39
$#56-3-3, %7!%"-5%312+ 23* 5#*%++-3, 5%1(#*& 1# 2&&%&& ()*"#+#,-$2+ %38-"#35%312+ &%"8-$%&
Thèse dirigée par Roger MOUSSA et Olivier ROUPSARD
&:;<=>;= ?= @A BCD=EFG= HI@I
Membres du jury :
8JKL=> 23*"02&&-23 -$!%4 $=EJMG=N/ 2><:>O "JPP:G<=;G
4GJ>D=QD ,2++2"1 $R=GDR=;G -*2%2/ 6JGD=?:>=/ %QPJM>= %SJET>J<=;G
+JGQ ,#11&$(2+U !G:N=QQ=;G '>TV=GQT<C BW#Q?:/ 3:GVXM= "JPP:G<=;G
":M=G 5#'&&2 *TG=D<=;G B= "=DR=GDR= -3"2/ 5:><P=??T=G *TG=D<=;G B= <RXQ=
#?TVT=G "#'!&2"* $R=GDR=;G $TGJB/ 1;GGTJ?FJ/ $:Q<J "TDJ $:YBTG=D<=;G B= <RXQ=
5JGD 8#+1Z *TG=D<=;G B= "=DR=GDR= -3"2/ 5:><P=??T=G !GCQTB=><
iii
Résumé
La production hydroélectrique est fortement impactée par la sédimentation dans les retenues
de barrages, qui sont influencés par l’occupation du sol, l'infiltration et les interactions entre
les eaux de surface et les aquifères. Pour le paiement des Services Hydrologiques
Environnementaux au Costa Rica, une quantification de l'impact des utilisations des terres
cultivées en café sur le fonctionnement des bassins est nécessaire. Cette thèse vise à : 1)
étudier les composantes du bilan hydrologique dans un bassin agroforestier nouvellement
équipé 2) quantifier les flux d’eau et de sédiments à différentes échelles : de la parcelle au
bassin et de l’événement de crue à l'échelle annuelle, et 3) simuler le bilan hydrique et la
production de sédiments en tenant compte du ruissellement sur les versants et les routes. Les
processus hydrologiques, écophysiologiques et de transfert de sédiments ont été suivis sur le
bassin (pluie, débit, évapotranspiration, humidité du sol, niveau de l'aquifère, turbidité) et sur
des parcelles (ruissellement de surface et érosion). Un nouveau modèle éco-hydrologique a
été développé pour simuler les termes du bilan hydrologique et de sédiments. Des
améliorations sont en cours pour tenir compte de l'effet des routes sur la genèse du
ruissellement de surface et de l’érosion. Le faible ruissellement de surface, la faible érosion
des parcelles et la faible production de sédiments, compte-tenu des conditions biophysiques et
des pratiques de gestion (arbres d’ombrage, désherbage), offrent clairement des Services
Ecosystémiques en réduisant les transferts superficiels de pesticides et de sédiments et en
permettant une bonne régulation du débit de rivière.
Mots clés : hydrologie, écophysiologie, sédiments, bassin versant, modélisation, café, bilan
hydrologique, agroforesterie, processus hydrologiques, débit, ruissellement de surface,
aquifère, turbidité, parcelle, effets d'échelle, érosion
v
Abstract
The profitability of hydropower is affected by soil erosion and sedimentation in dam
reservoirs, which are influenced by land use, infiltration and aquifer interactions with surface
water. In order to promote the payment of Hydrological Environmental Services (HES) in
Costa Rica, a quantitative assessment of the impact of land uses (such as coffee farming) on
the functioning of drainage basins is required. This thesis seeks: 1) to study the water balance
partitioning in a newly equipped coffee agroforestry basin 2) to estimate the water and
sediment yield at various spatio-temporal scales: from plot to basin and from event to annual
scale; and 3) to simulate the water and sediment yields, at both annual and peakflow scale, by
including the surface runoff from hillslopes and roads. The main hydrological,
ecophysiological and sediment processes were monitored during one year at the basin
(rainfall, streamflow, evapotranspiration, soil humidity, aquifer level, turbidity) and at two
plots (surface runoff and erosion). A new eco-hydrological model was developed to close the
water balance, and the annual sediment yield was also quantified. Improvements are in
progress to take into account the effect of roads in surface runoff generation. The low surface
runoff, low plot erosion and low basin sediment yield observed under the current biophysical
conditions (andisol) and management practices (no tillage, planted trees, bare soil kept by
weeding), offer potential HES by reducing the superficial displacement capacity for fertilizers
and pesticides, yielding low sediment loads and regulating streamflow variability through
highly efficient mechanisms of aquifer recharge-discharge.
Keywords: hydrology, ecophysiology, sediment, drainage basin, modelling, coffee, water
balance, agroforestry, hydrological processes, streamflow, surface runoff, aquifer, turbidity,
plot, scale effects, erosion
vi
1RXQ= PGCPJGC= J; Q=T> B= ?W'5" %D:[&:?Q 0D:?:MT= 4:>D<T:>>=??= [ 6T:MC:DRTET= B=Q
&:?Q [ 2MG:CD:QOQ<XE=Q \5:><P=??T=G &;P2MG:Y$-"2*Y-3"2Y-"*]/ H P?JD= !T=GG= 8TJ?J
6^< @H/ _`IAI 5:><P=??T=G $=B=S @ a 4GJ>D= =< B= ?W'5" +JF:GJ<:TG= BbC<;B= B=Q
-><=GJD<T:>Q &:?Y2MG:QOQ<XE=Y(OBG:QOQ<XE= +-&2( \-3"2Y-"*Y5:><P=??T=G &;P2MG:]/ H
P?JD= !T=GG= 8TJ?J 6^< H`/ _`IAI 5:><P=??T=G $=B=S @ a 4GJ>D=
vii
Remerciements
I would like to express my gratitude to my co-supervisor Dr Olivier Roupsard for supporting
me over the last three years, wishing but also demanding only the best from my work. His
scientific interest and ambition were essential for the proper completion of this thesis. Dr
Roger Moussa played an outstanding role in the PhD, supervising my work and generously
sharing his knowledge, experience and kindness. It was an honour but also a pleasure to study
hydrology under Roger’s guide. I would also like to extend my thanks to Dr Lars Gottschalk
and Dr Vasken Andréassian for acting as rapporteurs of this thesis, and to Dr Marc Voltz and
Dr Francesc Gallart for participating in the jury as président and examinateur, respectively.
Thanks to Dr Guerric le Maire, Dr Pierre Chevallier and Dr Marcel van Oijen for their
constructive advice and guidance during the thesis committee meetings.
Acknowledgements are due to the Centre de Coopération Internationale en Recherche
Agronomique pour le Développement (Cirad) for providing financial support to this PhD, and
to the Laboratoire d'étude des Interactions Sol - Agrosystème - Hydrosystème (LISAH) for
welcoming me during my work periods in France. Sincere thanks to Dr Jean-Michel Harmand
and Dr Marc Voltz for backing me as the directors of UPR-80 (Cirad) and LISAH,
respectively.
I extend my thanks to the following persons: Dr Robert Habib, Mrs Hélène Guillemain, Mrs
Géraldine Porri, Mrs Elizabeth Moreau, MSc. Simon Taugourdeau and Dr Jacques Avelino
(Cirad); Mrs Cécile Fontana and Mr Arnaud Dubreuil (LISAH); Mrs Martine Barraud
(SupAgro); Ing. Javier Orozco, Ing. Gilberto de la Cruz, Lic. Sadí Laporte, Lic. Rafael
Chacón, Ing. José Zúñiga, MSc. Gustavo Calvo, MSc. Luis Meléndez and Sra Margot Ávila
(ICE) and Dr Nils Roar (University of Oslo).
I would also like to thank the CAFNET project and Dr Philippe Vaast for providing financial
support, as well as to the PCP Agroforestry Systems with Perennial Crops and Dr Bruno
Rapidel for logistic and financial assistance.
Hacienda Aquiares and Mr Alfonso Robelo have offered fundamental support to the PhD by
allowing the installation of experimental equipment in the Aquiares farm and providing
viii
research funds, with the kind support of MSc. Guillermo Ramirez, Mr Rafael Acuña Vargas,
Mr Manuel Jara and Mr Alonso Barquero. I am deeply indebted to Mr Álvaro Barquero and
his family and Mr Alexis Pérez for their careful field work.
CATIE (Centro Agronómico Tropical de Investigación y Enseñanza) offered valuable
collaboration and infrastructure. Thanks to Dr John Beer, Dr Glenn Galloway, Dr Francisco
Jiménez, Dr Jeffrey Jones, MSc. Elias de Melo, Mrs Patricia Hernandez, MBA Margarita
Alvarado and Lic. Noily Navarro.
I want to thank my friends and fellows: Mauricio Galleguillos, Dennis Hallema, Armand
Crabit, Jesús Fernández, Simon Lalauze, Jérôme Ghesquière, Lionel Bouvet, Rossano
Ciampalini, Michaël Rabotin and Florent Levavasseur (LISAH); Louise Meylan and Fabien
Charbonnier (Cirad) and Raffaele Vignola, Rintaro Kinoshita, Carlos Cerdán and Pablo
Imbach (Catie).
Special thanks to Irina Gottschalk, Muriel Navarro, Etienne Leblois, Alejandra Apablaza and
Albane le Maire for their friendship, encouragement and many memorable dinners.
I dedicate this thesis to my wife Raquel in grateful thanks for her love, inexhaustible patience
and trust in me, to my father Manuel, who for more than two decades has suggested and
encouraged me to pursue a PhD, and to my mother Elizabeth for her caring advice and loving
support.
ix
Résumé substantiel
Cette étude se propose premièrement de présenter le contexte général et l'état actuel des
Services Environnementaux Hydrologiques (SEH) au Costa Rica et les hypothèses sur
lesquelles ils ont été développés et mis en œuvre, en insistant sur le rôle de producteurs de
café comme pourvoyeurs de SEH et celui des producteurs d'hydroélectricité comme
utilisateurs. En second, l'étude offre une perspective sur les processus écophysiologiques,
hydrologiques et sédimentologiques et présente les modèles. Un examen bibliographique est
centré sur les forêts tropicales et les systèmes agroforestiers à base café. Le bassin
expérimental d’Aquiares a été spécialement équipé pour fournir les données nécessaires à
cette thèse, il est décrit en détail. Viennent ensuite trois articles (trois chapitres). Le premier
traite de l'élaboration d'un nouveau modèle appelé « Hydro-SVAT » pour étudier le
comportement éco-hydrologique du bassin expérimental et boucler son bilan hydrique. Le
deuxième décrit les flux hydriques de surface et la production de sédiments dans le bassin, en
passant de l’échelle parcelle à celle du bassin, soit pendant les épisodes de fortes pluies, soit
pendant les récessions. Le troisième (en construction) vise l'amélioration de la simulation des
débits de pointe, nécessaire à une simulation correcte des transferts de sédiments, cherche
donc à améliorer le modèle Hydro-SVAT en introduisant une routine de ruissellement sur
routes, avec différentes variantes de modèles évaluées soit pendant les épisodes de fortes
pluies, soit sur l’année entière.
Services Environnementaux Hydrologiques, hydro!électricité et café
Le chapitre introductif traite de l'importance croissante et récente des SEH au Costa Rica,
soutenue par une tradition particulière sur le développement durable, dans un contexte où
l'énergie hydroélectrique et la production agricole (notamment le café) sont des activités
économiques prépondérantes. Notons que les SEH peuvent être fournis soit par les
écosystèmes actuels (tels que des forêts vierges), soit au-travers de Programmes de Gestion de
Bassin (PGB). Dans les deux cas, une quantification objective est requise. Il est soutenu ici
que les méthodes d'observation / expérimentation / modélisation constituent le moyen le plus
fiable pour obtenir ces évaluations, et peuvent également fournir une base utile pour les
x
programmes de suivi, ou pour l'évaluation périodique des objectifs des PGB ou autres projets
à vocation SEH. L’ignorance de ces approches quantitatives pourrait conduire à des
évaluations erronées, basées sur des mythes et / ou à des accords socio-économiques peu
solides pour le paiement des SEH. A titre d'exemple, l'utilisation abusive de la conception
populaire selon laquelle les arbres sont « des usines à eau » (alors qu'ils devraient être
considérés aussi comme des pompes à eau puisqu’ils puisent l’eau du sol pour assurer leurs
processus vitaux) peut conduire à de grands programmes de replantation en concurrence
directe avec la production d’eau dans un bassin versant hydro-électrique. La figure 1 montre
les résultats de dix expériences bassins jumelés reprises par Scott et al. (2004). Ces auteurs
ont étudié les effets du reboisement en utilisant trois différentes espèces d'arbres (Pinus
radiata, P. patula, et Eucalyptus grandis) dans les bassins d’Afrique du Sud. Dans tous les
cas une réduction significative du débit a été observée.
Figure 1. Réduction des débits mesurée dans cinq expériences de reboisement de bassins en Afrique du Sud. Les courbes sont ajustées pour 100% de plantation dans le bassin et lissées pour le débit annuel moyen avant la plantation (d'après Scott et al., 2004)
Bien que le reboisement ou les pratiques d'agroforesterie se soient avérées efficaces pour
réduire l'érosion des parcelles, la production de sédiments à l'échelle du bassin peut montrer
des réponses très variées à ces pratiques et être fortement affectée par des facteurs autres que
celui de l'utilisation des terres (par exemple climat, sol / lithologie, topographie, gestion,
xi
infrastructures comme les routes et aussi processus d'érosion liés à l’échelle). Par exemple,
une compilation de 60 études faisant état de la production annuelle de sédiment dans des
bassins généralement inférieurs à 100 km2 a été publiée par Bruijnzeel (2004), voir figure 2,
qui nous propose une relation probable entre différentes combinaisons de substrat géologique
et d'utilisation des terres. Ainsi, les bassins à substrat de granit sous forêt n'ont jamais présenté
de production de sédiments supérieure à 1 t ha-1 an-1. Au contraire, les bassins sur des sols
volcaniques / marnes affichant une utilisation mixte des terres ont montré des valeurs
supérieures à 100 t ha-1 an-1.
Figure 2. Gammes de production de sédiments par des bassins en Asie du Sud-Est en fonction du substrat géologique et de l'utilisation des terres. Catégories: I, forêts, granit, II, forêts, grès / schistes; III, forêts, roches volcaniques; IV, forêts, marnes; V, forêts exploitées (RIL: exploitation à faible impact); VI, forêts coupées, roches sédimentaires (barre inférieure: micro-bassins); VII, forêts coupées, roches volcaniques; VIII, forêts coupées, marnes, IX, bassins moyens à grands, utilisation mixte des terres, granit; X, idem, roches volcaniques; XI, idem, roches volcaniques et marnes; XII, urbanisé (barre inférieure) , exploitation minière et la construction de routes (barre supérieure).(d'après Bruijnzeel, 2004)
xii
Processus et modèles: ecophysiologie, hydrologie et sedimentologie
Ce chapitre (sur les processus et les modèles) donne des bases pour comprendre les liens
biophysiques entre les approches écophysiologiques, hydrologiques et sédimentologiques,
liens qui sont indispensables pour évaluer les SEH liés à l’eau et à la production de sédiments
par des bassins agroforestiers caféiers. Du point de vue du pourvoyeur de SEH, la plupart des
pratiques de gestion qu’il s’agit de mettre en œuvre au champ interférent avec
l'écophysiologie de la plantation, en affectant des processus majeurs, à la fois d’hydrologie et
de sédimentologie (par exemple en affectant l'évapotranspiration, l’infiltration et l'érosion).
Mais les variables d'intérêt pour la plupart des bénéficiaires des SEH seront perçues et donc
doivent être étudiées via des approches typiquement hydrologiques et sédimentologiques ainsi
qu’à la bonne échelle, c’est-à-dire à l'échelle du bassin. Ensuite, un équilibre doit être atteint
soit au niveau des mesures (par exemple les variables à mesurer, résolution spatio-temporelle
des observations) soit au niveau des étapes de modélisation (par exemple une représentation
équilibrée des processus dans les différents domaines scientifiques, conceptualisation
adéquate de leurs liens).
En outre, un élément important qui est souvent source de confusion entre les chercheurs
travaillant dans différents domaines, est la définition de l'érosion et celle des productions
sédimentaires. Alors que le premier terme est couramment employé pour identifier les
processus de perte de sol à l'échelle de la parcelle (qui peut être considéré à plus grande
échelle comme un "potentiel" de perte laminaire de sol), le second terme est davantage
reconnu pour des évaluations à l'échelle du bassin (compte tenu de la perte réelle de sol par le
système après avoir examiné les processus de dépôt et de pertes de masse). Bien que certaines
procédures de changement d'échelle spatiale aient été proposées afin de lier l'érosion des
parcelles à la production de sédiments par les bassins, leur intérêt devrait rester purement
indicatif, sachant que beaucoup d'autres facteurs demandent la plus grande attention. Pour
preuve que les estimations de production de sédiments par un bassins en fonction de sa
surface peut être très entachées d’incertitude, de Vente et al. (2007) ont fourni deux
ensembles de telles relations (obtenues par synthèse de plusieurs études sur des bassins à
l’échelle régionale) affichant des comportements opposés (Fig. 3). De cette figure, il est
évident qu'il n'y a pas de relation univoque à établir entre la production de sédiments par un
bassin et sa surface.
xiii
Figure 3. Exemples de production de sédiments en suspension (SSY) en fonction de la superficie du bassin (A). (A) Une relation négative est observée entre A et SSY pour plusieurs bassins à travers le monde, groupés en 18 régions (chaque courbe représente une région). (B) Exemple de relations positives entre A et SSY pour plusieurs bassins groupés en 11 régions (d'après de Vente et al., 2007).
Revue bibliographique des études écophysiologiques, hydrologiques et
sedimentologiques, avec l’accent mis sur les tropiques et les systèmes à
base café
A partir de notre étude bibliographique sur les études écophysiologiques et hydrologiques, on
note que ces champs disciplinaires renvoient deux approches principales pour mesurer,
modéliser et la fermer le bilan hydrique : l'ingénierie (hydrologie) approche visant à
comprendre les flux de surface et de restitution à l'échelle du bassin, et l'agronomie
(écophysiologie) approche plus intéressée par les bilans hydriques à échelle de la parcelle et
par les effets de la micro-climatologie locale sur les réponses physiologiques des plantes.
Toutefois, lorsque les deux domaines sont d'un intérêt équivalent pour une recherche, un
travail interdisciplinaire multi-échelle est souhaitable, en respectant l'importance du système
biophysique bien visible, mais aussi les composants cachés du sous-sol. Par conséquent,
certaines boîtes noires de modélisation que l’on trouve généralement à l’interface de ces deux
disciplines peuvent être efficacement élucidées lors d’une approche combinée.
Notre approche bibliographique sur l'érosion et la sédimentologie souligne le problème du
changement d’échelle de la parcelle au bassin. Un exemple de différences importantes qui
peuvent être trouvées dans la pratique est montré par Verbist et al. (2010) voir figure 4.
L'érosion de la parcelle (indiquée sur l'axe horizontal comme « plot ») a été systématiquement
xiv
et considérablement plus faible que les estimations de production de sédiments par le bassin
(indiqué comme « bas ») dans trois bassins différents (notés « WR », « WT » et « WP ») en
Indonésie. Ainsi, la production de sédiments est de 3 (pour WR) à 10 fois (pour WP et WT)
plus forte que l'érosion des parcelles mesurées.
Figure 4. Comparaison de la gamme mesurée de production de sédiments (t ha-1 an-1) au niveau de la parcelle (0,004 à 0,012 ha) et le niveau du bassin (bas), pour les bassins Way Ringkih (WR, 738 ha), Way Tebu (WT, 771 ha) et petai Way (WP, 1391 ha) dans Sumberjaya, en Indonésie (d'après Verbist et al., 2010)
En ce qui concerne l'érosion, les résultats antérieurs de nombreuses recherches dans les
parcelles de café montrent une grande variabilité, et nous insistons sur le fait que ces résultats
sont très dépendants du site et que les estimations universelles doivent être considérées avec
prudence. L’unique étude disponible sur la production de sédiments dans un bassin café
révèle un grand manque de connaissances dans ce domaine particulier. Une revue des études
sédimentologiques dans d'autres types de bassins indique que la technique la plus précise et la
plus récente pour quantifier les flux de sédiments celle de la mesure en continu de la turbidité,
qui se comporte correctement à la fois pendant les fortes pluies ou pendant les épisodes de
xv
récession, mais exige un étalonnage précis in situ. Néanmoins, nous avons noté de grandes
lacunes dans le traitement des incertitudes sur les estimations de la concentration et sur la
production totale de sédiments. Enfin, il est largement démontré dans cette bibliographie qu'il
est conceptuellement erroné de changer d’échelle directement à partir de résultats de parcelle
ou d’intégrer directement des résultats particuliers avec l'intention de fournir des estimations à
échelle du bassin.
Une méthodologie pour l'évaluation des SEH est donc proposée, en se basant sur les
approches éco-hydrologiques et sédimentologiques et en s'appuyant sur la combinaison
d'approches expérimentales et de modélisation. Les variables cibles doivent être clairement
définies, soit dans cette étude la recharge et les productions d’eau et de sédiments (tous trois à
l'échelle du bassin). Les deux premiers éléments demandent une fermeture adéquate du bilan
hydrique (qui exige des travaux de modélisation afin de déterminer les éléments non
mesurables), tandis que le troisième élément est abordé au-travers de la mesure continue du
flux de sédiments, compte-tenu de toutes ses incertitudes méthodologiques et techniques.
Site d’étude, dispositif experimental et données
Afin de suivre la proposition méthodologique présentée dans la section précédente, un
dispositif expérimental entièrement nouveau a été construit, en sélectionnant et en équipant un
bassin agroforestier à base café (1 km2) sur les pentes du volcan Turrialba, caractérisé par des
andosols perméables (Fig. 5). Le travail de terrain s'est avéré très efficace, offrant une
occasion unique d'observer directement et simultanément plusieurs variables éco-
hydrologiques et sédimentologiques, voies d'écoulement, compartiments de stockage de l’eau
dans cet environnement tropical (par exemple, les précipitations, le climat, les débits, la
teneur en eau du sol, l'évapotranspiration, l’indice foliaire, le niveau de l’aquifère, la turbidité
et l'érosion des parcelles). En outre, l'observation de terrain a permis la compréhension de
processus insoupçonnés qui ont aidé à expliquer les propriétés importantes et des tendances
dans les données sur ce bassin (par exemple très fortes infiltrations, faibles niveaux d’érosion,
fort ruissellement de surface sur les routes), et donc d'améliorer la paramétrisation du modèle
et même sa conceptualisation.
xvi
Figure 5. a) Emplacement du bassin du fleuve Reventazón au Costa Rica, en Amérique Centrale. b) Position du bassin expérimental à l'intérieur du bassin du Reventazón. c) Le bassin expérimental du projet « Coffee-Flux » dans la ferme d’Aquiares destiné à mesurer les composantes du bilan hydrique.
Modélisation du comportement hydrologique d’un basin agroforestier à
base café au Costa Rica
Ce chapitre sur article (Gómez-Delgado et al., 2011, accepté dans « Hydrology and Earth
System Sciences ») présente la fermeture du bilan hydrique et sa partition dans le site
expérimental, combinant des approches de modélisation et expérimentales. Le comportement
général de ce bassin agroforestier caféier durant plus d'un an d'observations peut être résumé
par le fait que 91% des précipitations (R) ont été infiltrées par cet andosol très perméable,
64% de R a été retrouvé dans le débit de rivière, 25% de R a été mesuré comme
évapotranspiration, il n’y a pas eu pas de grandes variations saisonnières de stock d'eau du sol
et on a détecté une contribution importante de la nappe au débit de base de la rivière.
Un nouveau modèle global éco-hydrologique appelé « Hydro-SVAT » a été élaboré et est
présenté dans la figure 6. Ce modèle a nécessité un couplage de routines hydrologiques
(Hydro) et écophysiologique (SVAT) et a été calibré en utilisant le débit à la sortie du bassin.
xvii
Il a été validé par des mesures indépendantes et directe de l'évapotranspiration, de l’humidité
du sol et des niveaux de la nappe phréatique.
Figure 6. “Hydro-SVAT”, le modèle éco-hydrologique conceptuel proposé pour le bassin expérimental, repésenté comme une entité unique.
Cet exercice de modélisation permet l'estimation d'un bilan hydrique complet (voir figure 7),
y compris les valeurs de certains composants « cachés », tels que la percolation profonde, qui,
dans ce bassin expérimental s'élevait à 14% de R.
La précision du modèle a été jugée assez juste selon notre analyse d'incertitude, alors qu’une
nouvelle approche d’analyse de sensibilité a permis de rechercher les origines de cette
incertitude. Le premier paramétrage du modèle a été considéré satisfaisant, bien qu’une
xviii
simplification du modèle ait été tentée, en fixant les deux ou trois paramètres les moins
sensibles.
Le modèle conceptuel éco-hydrologique "Hydro-SVAT" permet d'évaluer des conditions
différentes de climat, de couvert végétal, des sols et d'hydrogéologie, requérant seulement les
précipitations et le débit. Une validation dans un nouveau bassin agroforestier café sera tentée
(doctorat en cours de Mario Villatoro dans le bassin de Llano Bonito à Tarrazu, Costa Rica).
Les valeurs des paramètres déjà calibrés pourront être utilisées comme valeurs de départ pour
effectuer de nouveaux étalonnages dans des sites aux conditions similaires, ceci afin de
produire des estimations réalistes de l'évapotranspiration réelle, de teneur en eau du sol et de
recharge des aquifères, qui sont des variables rarement mesurées dans les bassins des pays en
développement.
Figure 7. Partition du bilan hydrique simulée par « Hydro-SVAT » dans le bassin expérimental d’Aquiares
Le bassin expérimental agroforestier à base café sur andosols s'est montré très conservatif en
termes d'infiltration, de recharge des aquifères, de régulation du débit de rivière et de
réduction du transport des contaminants en surface. La présente évaluation est une bonne base
pour la négociation du paiement des services environnementaux aux utilisateurs finaux de
cette eau « bleue », à opposer à l'eau « verte » transpiré par les plantes, ou encore à l’eau
« brune» qui ruisselle et érode en surface. Cependant, on doit rester conscient de certaines
vulnérabilités potentielles inhérentes (par exemple la capacité de transport de contaminants
vers les aquifères) pour lesquelles aucune donnée n'est encore disponible.
xix
Production d’eau et de sédiments dans un système agroforestier à base café
pour différentes échelles spatio!temporelles : de la parcelle au bassin et de
l’événement pluvieux au total annuel
Ce chapitre est basé sur un article qui sera soumis après révision complète en Novembre 2010
(Gómez-Delgado et al., 2011, en préparation). S'appuyant sur les données expérimentales, une
approche par changement d’échelle spatio-temporelle, de la parcelle au bassin et de
l'événement pluvieux au total annuel, permet de tracer les principales sources de l'érosion des
sols dans le bassin expérimental et de proposer des hypothèses sur les principaux processus de
production de sédiments. Sur un an, un très faible ruissellement de surface et des taux
d'érosion très faibles ont été mesurés sur les parcelles et sur le bassin. Des écarts ont été
observés dans les productions totales entre ces deux échelles, le bassin montrant un
ruissellement de surface et un transport de sédiments plus élevé que les parcelles extrapolées,
ce que nous avons interprété comme étant l'effet des routes sur le ruissellement de surface, le
contact routes-parcelles sur l'érosion, et d'autres sources de production de sédiments telles que
les berges et le lit de la rivière. Les périodes de récession représentaient 66% du rendement
annuel de sédiments, ce qui contredit de nombreuses études qui ont mesuré ou estimé que
l’essentiel de la production annuelle était le fait des événements de forte pluie. Par
conséquent, seul un tiers des sédiments mesurés à la sortie du bassin a été générés lors des
fortes pluies, dont seulement 16% auraient pu être générés par les parcelles, toujours selon les
mesures. Les parcelles n’ont donc généré qu’environ 5% de la production annuelle de
sédiments par le bassin, alors qu’elles en représentaient 95% de la surface) Même si les effets
bénéfiques de l'agroforesterie ont été confirmés ici (parcelles de café ombragées produisant la
moitié de l'érosion par rapport aux parcelles de café en monoculture), les routes jouent un rôle
beaucoup plus important sur le ruissellement de surface et la production de sédiments durant
les fortes pluies. En outre, la majorité de la production des sédiments n'a pas été générée
pendant les événements de pluie, mais en période de récession. Par conséquent, l'attention
devrait être concentrée sur l'observation des sédiments en suspension pendant les périodes de
basses eaux.
Le capteur optique à rétrodiffusion fournit une estimation du rendement annuel de sédiments
par le bassin avec une incertitude acceptable (voir figure 8), calculée via une approche
incertitude-standard qui a été pour la première fois appliquée aux calculs de la charge
sédimentaire. Ces résultats confirment que le bassin expérimental de café agroforestier est très
xx
conservateur en termes de rendement annuel de sédiments, n’ayant pas dépassé 1 t ha-1 an-1.
L'évaluation présentée ici semble être une bonne base pour la négociation du paiement des
services environnementaux. Il pourrait sembler superflu d'essayer de réduire encore ces
apports de sédiments déjà très faibles. Toutefois, si l'on décidait de le faire, il serait
souhaitable en priorité de stabiliser les berges et le lit de rivière, et de protéger le contact entre
les routes et les parcelles, plutôt que de simplement planter plus d'arbres par défaut.
Figure 8 Débit de rivière Q et concentration de sédiments en suspension S (lignes noires) dans le bassin expérimental pour une période de mesure d'un an. Un intervalle de confiance 95% (zone grise) est affiché autour des valeurs estimées. t = 30 min.
xxi
Il serait intéressant de comparer les rendements de sédiments et les origines du ruissellement
de surface ainsi que l’érosion à l’aide d’approches similaires mais dans d’autres bassins à
café, et en particulier, sur d’autres types de sols. Il est également envisagé de construire un
modèle couplé écophysiologie-hydrologie-érosion à partir de ces résultats expérimentaux
(chapitre suivant).
Amélioration des simulations de débits et de transport de sédiments de
pointe grâce à la conception d'un modèle semi!distribué éco!hydrologique, incluant le ruissellement sur les chemins : stratégies de validation sur des
événements de pluies ou en continu
Le modèle-bloc écohydrologique « Hydro-SVAT » présenté ci-dessus qui représente le bassin
entier comme une seule entité donne de bons résultats pour reproduire le débit sur une base
annuelle, mais nous avons souligné que les débits de pointe durant les événements de forte
pluie (ruissellement de surface pour l’essentiel) restaient sous-estimés, ce qui aura
certainement une incidence néfaste sur la simulation des productions des sédiments tant à
l’échelle annuelle que pour les événements pluvieux.
Nous avons déduit que les routes pourraient contribuer de manière significative au
ruissellement de surface durant les événements de fortes pluies à partir des résultats de
ruissellement de surface et d’érosion comparés aux échelles parcelle et bassin. La nécessité de
les conceptualiser comme une entité hydrologique différente dans le bassin, en plus des
parcelles, s’est imposée. Ceci nous a conduits à une version révisée et semi-distribuée du
premier modèle, en distinguant deux types d'unités de surface, les parcelles et les routes. A ce
jour, le simple ajout d'une routine d'interception par les routes (dans le modèle dit M4) n’a pas
amélioré significativement les simulations de l'écoulement de surface, par rapport à la version
simplifiée d'Hydro-SVAT (M3). Ceci est visible dans la figure 9, où le débit simulé par les
quatre modèles testés est comparé pour deux épisodes de fortes pluies, ceux du 24/06/2009
(Fig. 9c, d pour M3 et M4) et ceux du 02/11/2009 (Fig. 9g, h pour M3 et M4). Deux pistes
peuvent être poursuivies en vue d'atteindre un modèle satisfaisant : ajouter une routine de
ruissellement sur routes directement au modèle original à dix paramètres (afin de profiter de
sa représentation adéquate de la dynamique des aquifères), et / ou rendre variable la
proportion des surfaces imperméables autres que les routes (comme les sentiers intra-
parcelles) dont les dimensions sont inconnues en réalité.
xxii
0.0
0.1
0.2
0.3
0.4
16:00 18:30 21:00 23:30 02:00
Q (
m3 s
-1)
Q mea
Baseflow
QBF+SR M1
QBF M1
0.0
0.1
0.2
0.3
0.4
16:00 18:30 21:00 23:30 02:00
Q (
m3 s
-1)
Q mea
Baseflow
QBF+SR M2
QBF M2
0.0
0.1
0.2
0.3
0.4
16:00 18:30 21:00 23:30 02:00
Q (
m3 s
-1)
Q mea
Baseflow
QBF+SR M3
QBF M3
0.0
0.1
0.2
0.3
0.4
16:00 18:30 21:00 23:30 02:00
Q (
m3 s
-1)
Q mea
Baseflow
QBF+SR M4
QBF M4
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
11:00 14:30 18:00 21:30 01:00
Q (
m3 s
-1)
Q mea
Baseflow
QBF+SR M1
QBF M1
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
11:00 14:30 18:00 21:30 01:00
Q (
m3 s
-1)
Q mea
Baseflow
QBF+SR M2
QBF M2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
11:00 14:30 18:00 21:30 01:00
Q (
m3 s
-1)
Q mea
Baseflow
QBF+SR M3
QBF M3
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
11:00 14:30 18:00 21:30 01:00
Q (
m3 s
-1)
Q mea
Baseflow
QBF+SR M4
QBF M4
Figure 9. Simulations de la partition des débits de rivière au cours de deux événements de fortes pluies. Evénement du 24/06/2009: (a) M1: modèle simple, (b) M2: modèle Hydro-SVAT 10 paramètres, (c) M3: modèle Hydro-SVAT 7 paramètres, et (d) M4: Hydro-SVAT avec ruissellement sur routes. Evénement du 02/11/2009: (e) M1, (f) M2, (g) M3, (h) M4. Qmea: débit mesuré ; débit de base: séparation manuelle du débit de base ; QBF: débit de base modélisé, QSR: ruissellement de surface modélisé, QBF + QSR: total des débits modélisés.
a)
c)
e)
b)
d)
f)
g) h)
NS = 0.80 Pr = 0.25 Vr = 0.12
NS = 0.93 Pr = 0.13 Vr = 0.02
NS = 0.71 Pr = 0.24 Vr = 0.10
NS = 0.90 Pr = 0.09 Vr = 0.11
NS = 0.61 Pr = 0.53 Vr = 0.08
NS = 0.73 Pr = 0.44 Vr = 0.03
NS = 0.51 Pr = 0.51 Vr = 0.06
NS = 0.75 Pr = 0.44 Vr = 0.00
xxiii
Le travail est toujours en cours, et le résultat souhaité (un modèle semi-distribué qui puisse
simuler avec précision à la fois l'hydrographe annuel et débits de pointe tout en gardant un
niveau raisonnablement bas de complexité) appelle de re-conceptualisation en profondeur du
modèle actuel.
Deuxièmement, étant donné que Hydro-SVAT présente déjà dix paramètres dans sa version
initiale, il apparaissait nécessaire d'explorer les possibilités de proposer une version simplifiée
qui a été évaluée en termes de flexibilité et d'efficacité. La diminution du nombre de
paramètres basée sur l’analyse de sensibilité a produit un nouveau modèle qui s'est bien
comporté sur les mêmes données utilisées pour l'analyse de sensibilité, mais qui a montré des
lacunes lorsqu'il a été utilisé sur un nouvel ensemble de données. Toutefois, une réduction
significative des capacités d’un modèle peut se produire lorsque l'analyse de sensibilité est
utilisée pour réduire la quantité de paramètres, en se basant sur un ensemble de données
unique (comparer les performances des modèles M2 et M3 dans la Fig. 9, parcelle b vs c, et
parcelle f vs g). De meilleurs résultats peuvent être obtenus si cette procédure est appliquée en
utilisant une approche par échantillon fractionné.
La méthode dite par " point de départ dirigé " pour une calibration mono-objectif (basée sur le
coefficient de Nash-Sutcliffe: NS) semblait fonctionner correctement, maintenant
effectivement la calibration de tous les modèles à l’intérieur d’espaces paramétriques
réalistes. En supplément, une utilisation d’algorithmes à démarrage multiple pourrait être
testée afin de produire des résultats plus généraux.
Les techniques de validation en continu ou bien basées sur les événements sont ensuite
comparées afin de fournir une rétroaction à l'étape de paramétrage du modèle. L'efficience de
forme de l’hydrographe C2M (le coefficient NS borné) utilisé dans l'étape de validation a été
un critère beaucoup plus stable que le NS classique, offrant des statistiques résumées plus
fiables (par exemple, la moyenne d'un groupe d’épisodes de fortes pluies). Le manque
d'indépendance de certains critères étudiés ici (NS, C2M, erreurs globales et relatives sur le
volume, et les erreurs de pics globales et relatives) représente un grave problème pour leur
utilisation en tant que fonctions objectives dans des procédures multi-objectifs de calibration
du modèle qui s'appuient sur des routines de pondération. L'utilisation de l'analyse en
composantes principales pourrait être utile pour mieux approcher les conditions théoriques
exigées par ces techniques.
xxiv
Finalement, la comparaison des quatre modèles (certains comprenant l'effet des routes sur la
génération du ruissellement de surface, d'autres simplifiant le nombre des réservoirs ou de
paramètres du modèle) nous amène à une conclusion inattendue : les meilleures performances
sur le débit et sa partition restent celles du modèle original, celui à dix-paramètre Hydro-
SVAT. Dans ce modèle la routine qui permettrait d’affiner la simulation du ruissellement par
les routes n'a pas encore été testée.
Ces résultats très préliminaires confirment que les modèles simples ne sont pas toujours
suffisamment précis, surtout si plusieurs propriétés de systèmes complexes doivent être
prédites (pas seulement la forme des hydrographes des débits, mais les pics nécessaires à un
futur modèle de production de sédiments). Les évaluations standardisées, compte-tenu de la
relation entre efficacité du modèle et nombre de paramètres, peuvent fournir des informations
utiles dans le processus de sélection de modèles.
Perspectives
La méthodologie proposée dans cette thèse pour évaluer les SEH générés par des bassins
agroforestiers à base de café au Costa Rica, est censée apporter une nouvelle méthode de
quantification. La méthode a été choisie à partir de l’approche bibliographique présentée dans
le premier chapitre de ce document. Il est proposé que cette méthodologie puisse être utilisée
comme base durant la conception, la négociation ou la mise en œuvre pour le paiement des
SEH, ou encore pour la post-évaluation des projets (Programmes de Gestion de Bassin).
Toutefois, des travaux de terrain supplémentaires pourraient être effectués afin d'améliorer la
précision des méthodes biophysiques. Par exemple, des études supplémentaires sur les
versants et les eaux de ruissellement routier pourraient aider à renforcer les connaissances sur
l'influence de ces éléments dans la réponse hydrologique pendant les épisodes de fortes pluies.
Il faudrait également obtenir des renseignements sur les gains et pertes d'eau par les routes qui
traversent les limites du bassin. Bien que les eaux de ruissellement et l'érosion mesurées sur
nos parcelles expérimentales se soient avérées nettement moins importantes que ce que nous
avions supposé au départ, des répétitions de placettes arborées ou non devraient être ajoutées
à la configuration expérimentale actuelle (en particulier au pied des pentes dans des
conditions où le sol est plus proche de la saturation), pour permettre une généralisation, ou au
contraire réviser nos premiers résultats à l’échelle parcelle. L’effet des arbres sur l’infiltration
xxv
doit être approfondi (doctorat en cours de Laura Benegas sur ces sujets). De nouvelles
données sur la composition isotopique de l’eau de la rivière, couplée à la composition des
autres compartiments du bassin apporterait une meilleure compréhension de la dynamique des
eaux souterraines dans le bassin (doctorat en cours de Laura Benegas). D’autre part, il faudrait
étudier les sources des sédiments en suspension, soit pendant les épisodes de fortes pluies,
soit, plus important encore, pendant les périodes de récession, qui ont transporté les deux tiers
de la charge sédimentaire annuelle. Par conséquent, une analyse sur les origines de sédiments
dans la rivière est d'un intérêt majeur, en évaluant la quantité qui provient des berges ou du lit
de la rivière, mais encore des routes et des contacts entre les routes et les parcelles.
En ce qui concerne le modèle Hydro-SVAT, de nouvelles observations pour déterminer les
variations saisonnières des propriétés écophysiologiques amenées par les pratiques agricoles
(par exemple, l'effet de la taille du café sur l’interception de la pluie, ou l’effet du désherbage
sur l'infiltration), pourraient aider à affiner les hypothèses du modèle, ainsi que ses capacités
d'émulation.
La mise en œuvre de tests d'infiltration et de la capacité au champ sur les sols du bassin
(doctorat en cours de Laura Benegas, études antérieures par MSc Rintaro Kinoshita) donnerait
des valeurs réelles pour la conductivité hydraulique à saturation, la saturation des sols et la
capacité au champ, ce qui pourrait améliorer sensiblement le réalisme conceptuel du modèle,
et réduire en même temps, le nombre de paramètres. L'exécution répétée de ces tests dans le
bassin permettrait l'amélioration de l'actuel modèle Hydro-SVAT sous la forme d'un modèle
semi ou entièrement distribué.
Enfin, puisque les forts événements pluvieux produisent un tiers de la production annuelle de
sédiments dans le bassin expérimental, l’objectif d'amélioration de la simulation des débits de
pointe est justifié. Bien que les routines de ruissellement sur route aient été essayées pour ces
périodes, le comportement du modèle reste imprécis et d'autres hypothèses doivent être
explorées. Par exemple, la variation temporelle des propriétés du sol pendant les tempêtes
pourrait être étudiée afin de tester à quel point l'inclusion de cet effet peut améliorer les
simulations du modèle. D'autres variantes de modèle semi-distribué présenté dans le dernier
chapitre de cette thèse technique peuvent être également examinées.
Le développement module de simulation de production de sédiments couplable à Hydro-
SVAT fait encore défaut, mais cela exigera de résoudre d'abord le problème de la simulation
xxvi
des débits de pointe. En se basant sur le comportement observé des concentrations de
sédiments dans le temps, un modèle stochastique pourrait être utile pour représenter la
production de sédiments pendant les périodes de récession, alors que les approches
conceptuelles ou empiriques permettraient de reproduire le comportement hystérétique de la
courbe débit - concentration en sédiments, ce qui pourrait fournir un cadre approprié pour la
modélisation des pics de sédiments durant les fortes pluies.
Enfin, il très souhaitable de poursuivre la validation du modèle Hydro-SVAT dans d’autres
bassins café (doctorat en cours par Mario Villatoro à Llano Bonito, Tarrazzu). Le bilan
hydrique généré par ce modèle pourra aussi être comparé à celui du modèle général
d’agroforsterie café (van Oijen et al., 2010b), soit pour le valider, soit pour permettre de
raffiner certaines de ses hypothèses sur partition infiltration-ruissellement de surface, évapo-
transpiration, ruissellement de subsurface, drainage profond, percolation, etc. (doctorat en
cours de Louise Meylan).
Considérant enfin qu’Hydro-SVAT est fondamentalement un modèle générique pour la
partition du bilan hydrique à l'échelle parcelle et bassin, il est proposé de développer aussi des
applications sous différents bassins forestiers (par exemple, les plantations d'eucalyptus au
Brésil: projet Eucflux, les forêts secondaires au Congo, UE-Climafrica projet).
xxvii
Contents
RESUME ................................................................................................................................................................................................. III
ABSTRACT .............................................................................................................................................................................................. V
REMERCIEMENTS .............................................................................................................................................................................. VII
RESUME SUBSTANTIEL......................................................................................................................................................................IX
CHAPTER 1 INTRODUCTION............................................................................................................................................................. 1
@c@ ()*"#+#,-$2+ %38-"#35%312+ &%"8-$%& \(%&]/ ()*"#!#d%" 23* &%*-5%31 )-%+* -3 $#&12 "-$2 cccccccccccccccccccccccccccccccccccccccccc @
1.1.1 Costa Rican programme for the payment of HES .................................................................................................. 1
1.1.2 Hypothesis on the role of forests, agroforests and trees on the water balance partitioning ............. 3
1.1.3 Hypothesis on the role of forests, agroforests and trees on the reduction of sediment yield ............. 6
1.1.4 On the importance of verifying the working hypotheses, monitoring the environmental targets
and evaluating the actual hydrological services.................................................................................................................. 7 @cH (%&/ $#44%% 23* 2,"#4#"%&1") -3 $#&12 "-$2 cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc@I
1.2.1 The coffee sector in the world and in Costa Rica..................................................................................................10
1.2.2 Environmental services provided by coffee systems ...........................................................................................10 @c_ +-3U-3, $#44%% (%& d-1( ()*"#!#d%" "%.'-%"%5%31& -3 $#&12 "-$2cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc@@ @c` $#3$+'*-3, "%52"U& cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc@H @ce 1(%&-& ()!#1(%&%& cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc@_ @cA 1(%&-& #6f%$1-8%&ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc@_ @cg 1(%&-& #8%"8-%d cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc@_
CHAPTER 2 PROCESSES AND MODELS: ECOPHYSIOLOGY, HYDROLOGY AND SEDIMENTS........................................17
Hc@ %$#!()&-#+#,) ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc@h
2.1.1 Definitions of terms and equations of processes ..................................................................................................18
2.1.2 Models......................................................................................................................................................................................20
HcH ()*"#+#,-$2+ !"#$%&&%& 23* 5#*%+& ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccHA
2.2.1 Hydrological processes ....................................................................................................................................................26
2.2.2 Models......................................................................................................................................................................................28
Hc_ %"#&-#3 23* &%*-5%31#+#,) ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccHi
2.3.1 Plot interrill (sheet) erosion ..........................................................................................................................................29
2.3.2 Plot rill erosion ....................................................................................................................................................................32
2.3.3 Gully erosion .........................................................................................................................................................................34
2.3.4 The Universal Soil Loss Equation: an empirical approach ..............................................................................35
2.3.5 Sediment yield at basin scale ........................................................................................................................................37 Hc` $#3$+'*-3, "%52"U& cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc_i
xxviii
CHAPTER 3 BIBLIOGRAPHIC REVIEW ON ECOPHYSIOLOGICAL, HYDROLOGICAL AND SEDIMENT STUDIES,
WITH A FOCUS ON TROPICAL AND COFFEE!BASED SYSTEMS........................................................................................41
_c@ d21%" 62+23$% 4"#5 !+#1 1# 62&-39 5%2&'"%5%31& 23* '!&$2+-3, cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc`@
3.1.1 Water balance at plot scale ...........................................................................................................................................42
3.1.2 Water balance at basin scale ........................................................................................................................................44
3.1.3 Water balance at both, plot and basin scales ........................................................................................................48
3.1.4 Water balance at regional scale..................................................................................................................................49 _cH !+#1 %"#&-#3 23* 62&-3 &%*-5%31 )-%+*9 5%2&'"%5%31& 23* '!&$2+-3, ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccce@
3.2.1 Erosion at plot scale ..........................................................................................................................................................51
3.2.2 Sediment yield at basin scale ........................................................................................................................................52
3.2.3 Upscaling, from plot erosion to basin sediment yield.........................................................................................58 _c_ $#3$+'*-3, "%52"U&9 !"#!#&-3, 5%2&'"%5%31 23* 5#*%++-3, 5%1(#*& 4#" $#56-3%* %$#!()&-#+#,-$2+/ ()*"#+#,-$2+
23* &%*-5%31#+#,-$2+ 2&&%&&5%31& -3 $#44%% 2,"#4#"%&1") 62&-3& cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccA@
3.3.1 Water balance: ecophysiological and hydrological approaches...................................................................61
3.3.2 The problem of upscaling plot erosion to basin sediment yield ....................................................................62
3.3.3 Proposing a methodology for the assessment of HES based on ecophysiological, hydrological and
sedimentologic approaches .........................................................................................................................................................62
CHAPTER 4 STUDY SITE, EXPERIMENTAL SETUP AND DATA..............................................................................................65
`c@ &1'*) &-1%9 +#$21-#3/ $+-521% 23* &#-+ cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccAe `cH %7!%"-5%312+ &%1'! cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccAh
4.2.1 Rainfall (R) and climate ..................................................................................................................................................68
4.2.2 Streamflow (Q) ....................................................................................................................................................................69
4.2.3 Soil water content (!).......................................................................................................................................................69
4.2.4 Evapotranspiration (ET).................................................................................................................................................70
4.2.5 Leaf Area Index (LAI)........................................................................................................................................................71
4.2.6 Water table level (z)..........................................................................................................................................................72
4.2.7 Turbidity (")..........................................................................................................................................................................73
4.2.8 Plot surface runoff (SRplot) and plot erosion...........................................................................................................74 `c_ *2129 5%2&'"%5%31 !%"-#*/ 5-&&-3, 82+'%& 23* ,2!Y4-++-3, cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccg`
CHAPTER 5 MODELLING THE HYDROLOGICAL BEHAVIOUR OF A COFFEE AGROFORESTRY BASIN IN COSTA
RICA...................................................................................................................................................................................................77
ec@ -31"#*'$1-#3 cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccgi ecH 1(% &1'*) &-1% cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccchH
5.2.1 Location, soil and climate...............................................................................................................................................82
5.2.2 Experimental setup............................................................................................................................................................83 ec_ ()*"#Y&821 +'5!%* 5#*%+ cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccchA
5.3.1 Model structure ...................................................................................................................................................................87
xxix
5.3.2 Model parameterization, calibration and validation ........................................................................................95
5.3.3 Uncertainty and sensitivity analysis ..........................................................................................................................98 ec` "%&'+1& ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccii
5.4.1 Hydrological behaviour of the basin..........................................................................................................................99
5.4.2 Ecophysiological behaviour of the basin............................................................................................................... 101
5.4.3 Water balance partitioning and closure............................................................................................................... 102 ece *-&$'&&-#3 cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @I`
5.5.1 Validation, uncertainty and sensitivity analysis for the Hydro#SVAT model........................................104
5.5.2 Model simplification....................................................................................................................................................... 109
5.5.3 Hydrological processes in the experimental basin ........................................................................................... 111
5.5.4 The coffee agroforestry basin and hydrological services .............................................................................. 115 ecA $#3$+'&-#3&ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @@A
CHAPTER 6 WATER AND SEDIMENT YIELD IN A COFFEE AGROFORESTRY SYSTEM AT VARIOUS SPATIO!
TEMPORAL SCALES: FROM PLOT TO BASIN AND FROM EVENT TO ANNUAL SCALE............................................ 129
Ac@ -31"#*'$1-#3 cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @_I
6.1.1 Hydrological Environmental Services.................................................................................................................... 130
6.1.2 Spatial scales from the plot to the basin ............................................................................................................... 131
6.1.3 Temporal scales from the event to the whole year........................................................................................... 131
6.1.4 Uncertainty in sediment estimations......................................................................................................................132
6.1.5 Objectives and methodology.......................................................................................................................................133 AcH &1'*) &-1%9 +#$21-#3/ &#-+ 23* $+-521% cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @_` Ac_ 521%"-2+& 23* 5%1(#*&ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @_e
6.3.1 Experimental setup......................................................................................................................................................... 135
6.3.2 Measurement period, missing values and gap#filling...................................................................................... 137
6.3.3 Estimation of basin sediment yield and its uncertainty................................................................................. 137
6.3.4 Definition of storm events, baseflow separation and estimation of surface runoff from roads...140 Ac` "%&'+1& ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @`H
6.4.1 Water and sediment yields during storm events............................................................................................... 142
6.4.2 Annual water and sediment yield.............................................................................................................................148
6.4.3 Upscaling from storm to annual scale and from plot to basin scale ........................................................ 152 Ace *-&$'&&-#3 cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @e_
6.5.1 Water yield of the experimental basin: upscaling in time............................................................................. 154
6.5.2 Runoff on the experimental plots: upscaling in time....................................................................................... 155
6.5.3 Sediment yield of the experimental basin and erosion in the experimental plots: upscaling in time
156
6.5.4 Where is the sediment coming from and when? A spatio#temporal upscaling ...................................157
6.5.5 Uncertainty in sediment estimations......................................................................................................................159 AcA $#3$+'&-#3&ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @AI
xxx
CHAPTER 7 IMPROVING WATER AND SEDIMENT PEAKFLOW SIMULATIONS FROM AN ECO!HYDROLOGICAL
MODEL, BY INCLUDING THE EFFECT OF DIRECT ROAD!RUNOFF: CONTINUOUS VERSUS EVENT!BASED
VALIDATION STRATEGIES ...................................................................................................................................................... 175
gc@ -31"#*'$1-#3 cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @gA gcH &1'*) &-1%9 +#$21-#3/ &#-+ 23* $+-521% cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @gi gc_ 521%"-2+& 23* 5%1(#*& cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @h@
7.3.1 The experimental setup ................................................................................................................................................ 181
7.3.2 Data set and hydrological models............................................................................................................................182
7.3.3 The road#interception routine...................................................................................................................................182
7.3.4 Model calibration procedure......................................................................................................................................185
7.3.5 Model evaluation procedure.......................................................................................................................................186 gc` "%&'+1& 23* *-&$'&&-#3 cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc @hi
7.4.1 Model performance on continuous time ............................................................................................................... 189
7.4.2 Model performance on storm events ...................................................................................................................... 193
7.4.3 Streamflow partition .....................................................................................................................................................198 gce $#3$+'&-#3&ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc HII
7.5.1 On the effect of roads in model performance...................................................................................................... 200
7.5.2 On the model structure and the reduction of parameters ............................................................................ 200
7.5.3 On the “directed” model calibration .......................................................................................................................201
7.5.4 On the two time scale (continuous#time, storm event) multi#criteria validation...............................201
CHAPTER 8 GENERAL CONCLUSIONS AND PERSPECTIVES................................................................................................ 207
hc@ #3 ()*"#+#,-$2+ %38-"#35%312+ &%"8-$%&/ ()*"#!#d%" 23* $#44%% cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc HIg hcH #3 !"#$%&&%& 23* 5#*%+&9 %$#!()&-#+#,)/ ()*"#+#,) 23* &%*-5%31#+#,) ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc HIh hc_ #3 1(% 6-6+-#,"2!(-$ "%8-%d #3 %$#!()&-#+#,-$2+/ ()*"#+#,-$2+ 23* &%*-5%31#+#,-$2+ &1'*-%&/ d-1( 4#$'& #3
1"#!-$2+ 23* $#44%%Y62&%* &)&1%5& ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc HIi hc` #3 1(% &1'*) &-1%/ %7!%"-5%312+ &%1'! 23* *212ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc H@I hce #3 5#*%++-3, 1(% ()*"#+#,-$2+ 6%(28-#'" #4 2 $#44%% 2,"#4#"%&1") 62&-3 -3 $#&12 "-$2 cccccccccccccccccccccccccccccccccccccccccccccc H@I hcA #3 d21%" 23* &%*-5%31 )-%+* -3 2 $#44%% 2,"#4#"%&1") &)&1%5 21 82"-#'& &!21-#Y1%5!#"2+ &$2+%&9 4"#5 !+#1 1# 62&-3
23* 4"#5 %8%31 1# 233'2+ &$2+%ccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc H@@ hcg #3 -5!"#8-3, d21%" 23* &%*-5%31 !%2U4+#d &-5'+21-#3& 1("#',( 1(% *%&-,3-3, 2 &%5-Y*-&1"-6'1%* %$#Y
()*"#+#,-$2+ 5#*%+ -3$+'*-3, 1(% %44%$1 #4 *-"%$1 "#2*Y"'3#449 $#31-3'#'& 8%"&'& %8%31Y62&%* 82+-*21-#3
&1"21%,-%& cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc H@_ hch !%"&!%$1-8%&cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc H@e
REFERENCES ..................................................................................................................................................................................... 217
LIST OF TABLES ............................................................................................................................................................................... 247
LIST OF FIGURES.............................................................................................................................................................................. 249
APPENDIX: PAPER ACCEPTED: MODELLING THE HYDROLOGICAL BEHAVIOUR OF A COFFEE
AGROFORESTRY BASIN IN COSTA RICA.............................................................................................................................. 255
xxxi
$:NN== JMG:N:G=Q<GO FJQT> T> 2j;TJG=Q NJGEc !R:<: FO 3T?Q ":JG &k?<R;>c
1
CHAPTER 1
Introduction
1.1 Hydrological Environmental Services (HES), hydropower and sediment
yield in Costa Rica
1.1.1 Costa Rican programme for the payment of HES
The provision of Environmental Services (ES) is a widely discussed topic in Costa Rica, as a
pioneer country in the financial recognition of the external benefits of natural forest (Chomitz
et al., 1999) and agroforestry systems (Zbinden and Lee, 2005), through a scheme of Payment
for Environmental Services (PES). The Costa Rican forestry law no. 7575 defines the ES in
terms of (1) mitigation of greenhouse gas emissions (fixation, reduction, sequestration,
storage and absorption), (2) water protection for urban, rural or hydroelectric exploitation, (3)
protection of biodiversity for its conservation and sustainable use (scientific and
pharmaceutical employment, genetic research and improvement), and (4) protection of
ecosystems, life forms, and natural scenic beauty for touristic and scientific purposes.
The PES program has made substantial progress in charging water users for the ES they
receive, and more limited progress in charging biodiversity and carbon sequestration users
(Pagiola, 2008). The participation of hydropower producers seems determinant in the relative
success of either HES payments or other environmental investments, like those under basin
management programs (Jaubert, 2001), given the high dependence of the country in
hydropower (Chomitz et al., 1999), which generated 78% of the total electricity consumption
in Costa Rica during 2008 (ICE, 2009).
Small hydropower producers, operating run-of-the river plants, have limited water storage
capacity, and thus depend on the basin to provide adequate seasonal flows to sustain their
electricity production (Postel and Thompson, 2005). In addition to the maintaining of a stable
streamflow during the dry season, small produces target to reduce peakflows when the project
can not use the additional water because of its small storage capacity (Rojas and Aylward,
2002). On the other hand, producers with large water reservoirs mostly experience problems
2
with sedimentation processes. The Angostura reservoir (Fig. 10), located in the Caribean
region of Costa Rica, clearly exemplifies the effects of high sediment inflow rates.
Figure 10. Angostura reservoir suffers an extended sedimentation problem, entailing large cleaning costs. (Courtesy of ICE)
Operational costs linked to cleaning the useful storage volume of dam-reservoirs are directly
proportional to suspended sediment concentrations in reservoir tributaries. In both cases, HES
payments are based on the assumption that natural forests (or reforestation schemes like
agroforestry) provide the ES of capturing and retaining water, and preventing excessive soil
erosion in areas with steep slopes. However, there is still much uncertainty in these
assumptions, as the links between land use and hydrology in Costa Rica have not been
thoroughly investigated by quantitative research (Anderson et al., 2006).
3
1.1.2 Hypothesis on the role of forests, agroforests and trees on the water
balance partitioning
Concerning the aims of small hydropower producers at paying HES, it is highly probable that
peakflows are smaller under forest covered headwater basins, or that they might be reduced
by replanting trees (Guillemette et al., 2005). However, decades of hydrological studies
caution against uncritically accepting the conventional wisdom that forests are like sponges,
soaking up water in the wet season and slowly releasing it in the dry season (Chomitz et al.,
1999).
According to Saint-André et al. (2007), water yield is altered through changes in
transpiration, interception, and evaporation, the first two terms tending to increase when
grasslands or shrublands are replaced with trees. Planting trees is often assumed to increase
infiltration (e.g. Ilstedt et al., 2007). Transpiration rates are influenced by changes in rooting
characteristics, leaf area, stomatal response, plant surface albedo, and turbulence (Brooks et
al., 1997; Vertessy, 1999; Hoffmann and Jackson, 2000; Jackson et al., 2001). Although
transpiration is traditionally considered the most important component of forest
evapotranspiration (ET), interception and subsequent evaporation from the canopy can also
increase substantially, particularly with conifers (Pearce and Rowe, 1979; Cannell, 1999).
Evaporation of intercepted precipitation is generally low in grasslands, but can account for
10-20% of rainfall for broadleaf trees and 20-40% for conifers (Le Maitre et al., 1999). In
tropical plantations of broadleaf trees where rainfall events of large intensity are dominant,
evaporation of intercepted precipitation is usually lower than 10% (Whitehead and Beadle,
2004). The sum of the changes in evaporation and transpiration in plantation basins leads to
an increase in ET (Holmes and Sinclair, 1986); for example, ET from a basin planted with
eucalyptus could be 40-250 mm higher than from a grassland basin (Zhang et al., 1999). The
impact of afforestation on water resources is likely to vary with the climate (rainfall of the
zone), the soil water reserve, the type of plantation and its age. Considering potential climate
feedbacks associated with reforestation, Jackson et al. (2005) simulated for eastern USA that
plantations would increase evapotranspiration and decrease air temperature, hence likely
decreasing precipitations, as compared to crops or pastures (the simulation is presented for
temperate zones, the conclusion can be different under the tropics). Dolman et al. (2004)
stressed that in general, ETR would be larger above forests, forests having a lower albedo than
4
crops or grasslands (10-15% for forests but 20-25% for grasslands), hence more available
energy, more sensible heat flux, creating a “forest-breeze” above forests. Combined with
other phenomena, this would generally lead to increased precipitations over forests.
Few systematic global analyses of the effects of afforestation on water yield have been
proposed. Farley et al. (2005) analyzed 26 basin datasets, including annual streamflow and
lowflow, concluding that reductions in streamflow can be expected following afforestation of
grasslands and shrublands, and may be most severe in drier regions. These authors assessed
the direction, range, and extent of changes in total annual streamflow and low flow associated
with afforestation, examined the interactions with original vegetation type (grassland or
shrubland), tree species planted, plantation age, and climate, and provided a predictive
framework for modelling the effects of afforestation on water yield for carbon sequestration
scenarios. According to Farley et al. (2005), streamflow decreased consistently and
substantially with afforestation across their entire meta-analysis. More than one-fifth of the
basins experienced reductions of 75% or more during at least one year and 13% of the basins
experienced 100% reductions in streamflow for at least one year. Bosch and Hewlett (1982)
compared 94 paired basin experiments, verifying the increment in overall water yield
following reductions in the vegetation cover. Bruijnzeel (1990; 2004) summarizes some
experiments in which basin water yield increased after tropical forest removal, as long as the
infiltration capacity of the surface is not impaired (see Fig. 11).
Figure 11. Relationship between reduction in forest cover and increase in basin water yield. (Figure extracted from Bruijnzeel, 2004)
5
The same phenomenon acting in the opposite direction was demonstrated by following the
effects of planting practices. Figure 12 shows the results of ten paired basin experiments
reported by Scott et al. (2004). They studied the effects of afforestation using three different
tree species (Pinus radiata, P. patula, and Eucalyptus grandis) in basins of South Africa. In
all cases a significant reduction in streamflow was observed.
Figure 12. Reductions in streamflow as measured in five basin afforestation experiments in South Africa. The curves are scaled for 100% planting of the basin and smoothed to the mean annual streamflow prior to planting. (Figure extracted from Scott et al., 2004)
These increases can be substantially explained by the considerable reduction of soil water
extraction through evapotranspiration processes (which may seriously affect low flows),
though many other factors can also interfere equilibrating the hydrological system in a hardly
predictable fashion. Based on these and other studies with similar results (see Van Lill et al.,
1980; Smith and Scott, 1992; Sahin and Hall, 1996; Scott and Smith, 1997; Sharda et al.,
1998; Samra et al., 2001; Wilk et al., 2001; Andréassian, 2004), no obvious conclusions can
be drawn about the hydrological effects (in terms of a reduction or increase of water yield, or
its seasonal patters) of either reforesting or forest preserving schemes, without considering a
complex set of influencing factors of different types: ecophysiological, climatological,
geomorphological, geological, pedological and of subsequent land management (Chomitz and
Kumari, 1998; Bruijnzeel, 2004; Scott et al., 2004; Pagiola, 2008).
6
1.1.3 Hypothesis on the role of forests, agroforests and trees on the reduction of sediment yield
Regarding the reduction in sediment yield, to which large hydropower producers bet when
they pay for HES, scientific facts seems to confirm the popular view of trees (particularly
when planted in dense stands) as checkers of soil erosion at plot scale (Bruijnzeel, 1990;
Wunder et al., 2008). However, at the basin scale that relationship is much less clear, as
factors like geological substrate (Bruijnzeel, 2004; Verbist et al., 2010), local infrastructure
like roads (Ziegler et al., 2000; Ziegler et al., 2004; Rijsdijk, 2005; Sidle et al., 2006; Rijsdijk
et al., 2007b; Vanacker et al., 2007), or the presence of mass wasting and land slides
(Brasington and Richards, 2000; Beylich and Sandberg, 2005; Rijsdijk et al., 2007a; Schwab
et al., 2008), can sometimes have equal or greater importance than that of land cover or
management practices. For instance, a compilation of 60 studies reporting annual sediment
yields in basins generally below 100 km2 was summarized by Bruijnzeel (2004) as presented
in Fig. 13, in order to propose a likely relationship between different combinations of
geological substrate and land use. Hence, a granite substrate basin under forest never
presented sediment yields above 1 t ha-1 y-1. On the contrary, basins on volcanic/marls soils
displaying a mixed land use got to yield values above 100 t ha-1 y-1.
According to Bruijnzeel (1990) it is rather hazardous to compare water yields for basins with
different land uses and simply ascribe differences in yield to the contrast in vegetation, but
this is even more true in the case of basin sediment yields, given that single extreme events
may change the entire sediment picture of an area by releasing enormous amounts of sediment
at once (White, 1990; Douglas et al., 1999; Gallart et al., 1999; Lane et al., 2006; García-Ruiz
et al., 2008). This may promote various types of temporary storage influencing sediment
yields for quite some time. But other factors like either the basin delivery-ratio problem
(Walling, 1983; Lu et al., 2006; de Vente et al., 2007), the sediment sampling intensity
(Gilroy et al., 1990; Lewis, 1996; Wass et al., 1997; Brasington and Richards, 2000;
Schoellhamer and Wright, 2003; Lane et al., 2006), the computation methods (Walling, 1977;
Lewis, 1996; Sun et al., 2001) or the measurement errors and/or uncertainty propagation
(Wass and Leeks, 1999; Jansson and Erlingsson, 2000) can strongly modify sediment yield
figures over time and space, thus making unrealistic any simple causal relationships between
land use management and basin sediment yield.
7
Figure 13. Ranges in reported basin sediment yields in southeast Asia as a function of geological substrate and land use. Categories: I, forest, granite; II, forest, sandstones/shales; III, forest, volcanics; IV, forest, marls; V, logged (RIL: reduced impact logging); VI, cleared, sedimentary rocks (lower bar: micro-basins); VII, cleared, volcanics; VIII, cleared, marls; IX, medium-large basins, mixed land use, granite; X, idem, volcanics; XI, idem, volcanics plus marls; XII, urbanised (lower bar), mining and road building (upper bar). (Figure extracted from Bruijnzeel, 2004)
1.1.4 On the importance of verifying the working hypotheses, monitoring the
environmental targets and evaluating the actual hydrological services
As it is discussed in the next sections, besides the HES being provided by an agroforestry
basin in terms of water quantity regulation (which is studied by closing of the WB), the
current state (and possible improvement) of water quality by management practices could be
also relevant HES to provide by a drainage basin. In the specific case of sediment yield, the
link between agricultural activities (exacerbating natural erosion) and hydropower production
(as a major water use) is the base to consider possible payments for HES intending to reduce
8
reservoir sedimentation. But any management practice being promoted throughout such
payments, or in the frame of basin management programs, should have been properly studied
in terms of the expected biophysical and financial benefits arising from its implementation.
This might be a factor allowing simpler negotiations for the price of the provision of HES,
while assuring sustainable payment mechanisms. Monitoring programs should be also put into
operation to verify that the effect of the practices is completely satisfactory.
The lack of knowledge about a particular HES, or on the true results of implemented practices
dues sometimes to the lack of systematic observations of the target phenomenon, but also to
the wrong conceptualization of existent techniques and/or models, which are often taken out
of its range of application in the search of generalized results. For instance, when quantifying
erosion and sediment yield at plot and basin scales, the methodological limits of the Universal
Soil Loss Equation (USLE, Wischmeier and Smith, 1978) and posterior updates have often
been pointed out, taking into account that this equation was developed for agricultural
estimations over plot scale data. Even though the model creators clearly state that the USLE
was designed to compute longtime average soil losses from sheet and rill erosion, this
equation has been used rather commonly to calculate total erosion (including mass erosion) at
greater scales (like the basin), ignoring facts such as that it does not predict deposition and
does not compute sediment yield from gully, streambank, and streambed erosion. The
tendency to use this equation to assess HES or possible effects of management practices at
basin scale, ignoring the additional complications brought about by the basin structure and
associated processes, responds to its simplicity and generality. However, it is also
incentivated by the frequent interest of large scale water users (like hydropower producers) in
basin scale meters, while maintaining a methodological link with the promotion and
implementation of management practices at plot scale.
Under a coherent conceptual framework, respecting the biophysical hypothesis and the
methodological working assumptions, both water and sediment HES could be effective
provided in order to mitigate some ecosystem threats induced by hydropower generation
activities. However, a clear understanding of the actual biophysical behaviour of these
systems is often lacking, as much as accurate information on the effects of land rehabilitation
on basin sediment yield (as discussed above), be it through physical soil conservation
measures, the introduction of agroforestry-based cropping systems or full-scale reforestation
9
(Bruijnzeel, 2004). As we have seen so far, the local context matters and universal rules of
thumb do not exist on these issues. In addition, monitoring programs to assess the
effectiveness of implemented practices are not well documented in Costa Rica and, like in
other tropical regions, are mainly focused on the plot scale (Verbist et al., 2010), thus
carrying to scrutiny for not delivering the desired reductions in downstream sediment rates
(Rijsdijk et al., 2007a). Referring to basin management programs in Central America,
Kaimowitz (2004) states: “most basin management projects give little priority to research and
monitoring, even though these are essential to effective basin management. It is emphasized
here that actual impacts of management projects are rarely assessed a posteriori, while only
virtual impacts are estimated a priori. Often organizations use the results from those studies
that are carried out to reaffirm their pre-existing positions, rather than to learn and adapt their
strategies. Others simply ignore them”. This is complemented by the view of Pagiola (2008),
who considers impossible to determine the extent to which the Costa Rican program of
payment for ES has successfully generated those ES. Then he adds: “Although the program of
payment for ES has established a strong system to monitor land user compliance with
payment contracts, the program remains weak in monitoring its effectiveness in generating
the desired services”. The sustainability of such programmes seems to be conditioned to the
capacity of the promoting entities to demonstrate in credible terms the real provision of ES to
the water users and payers, especially under user-financed programs. For that, appropriate
monitoring programs have to be implemented and genuinely threatened areas must be
prioritized (Southgate and Wunder, 2007; Pagiola, 2008; Wunder et al., 2008). Referring to
the specific case of hydropower producers paying for HES in Costa Rica, Rojas (2002)
indicated that they even do not know if in return for their payment they are getting more or
less water, or water of a different quality, while Fallas (2002) suggests that there is a high
degree of uncertainty in these assessments, which is complicated by the scarce data sources
available in the country. According to Rojas and Aylward (2003) these results raise doubts
about the assumption that forests and timber plantations are, by default, the optimal land use
for water users, and that forests do provide the benefits claimed by the proponents of ES
payments. The real services that are (or are not) provided, depend on what the water is used
for, and on the specific conditions of each basin.
10
1.2 HES, coffee and agroforestry in Costa Rica
1.2.1 The coffee sector in the world and in Costa Rica
Coffee is one of the most traded agricultural commodities in the world employing 100 million
people (Vega and Rosenquist, 2001), employing many small and medium farmers in 51
producing countries (Castro et al., 2004). Coffee plantations have enormous economic
importance for developing countries in the humid tropics (Beer et al., 1998). In Costa Rica,
coffee accounted for 15% of the agricultural exports in 2008 and covered 2% of the territory
(SEPSA, 2009). According to Payán et al. (2009) depressed international coffee prices from
1999 to 2003 lead to a search for coffee niche markets, offering greater economic premiums.
1.2.2 Environmental services provided by coffee systems
Coffee production in Costa Rica shows a very large diversity of conditions (soils, climate,
altitude, slopes, cultivars), of management practices (monoculture, agroforestry, level of
intensification, from highly conventional to low conventional or to organic), of quality and of
marketing and prices (certified or not, premium or not), which provides interesting conditions
to study the provision of ES and HES. Considering this diversity, modelling and scaling-up
tools are much needed, such as SVAT1 multipurpose models (e.g., van Oijen et al., 2010a, b)
or else hydrological models.
Many Costa Rican farmers adopted agroforestry systems, which over the last 30 years was
mainly used to improve micro-climatic conditions, soil fertility and sustainability of the
coffee plantations in lower lands (altitude < 700 m.a.s.l.), poorly suited for coffee
monoculture (Beer et al., 1998; van Kanten and Vaast, 2006). In addition to better prices for
coffee, revenue of diversified crops and spreading of economic risk are offered by shade trees
in terms of fruit and/or timber production, as well as decreasing production costs through the
reduction of agrochemical inputs, have become important financial mechanisms for many
small-scale and organic coffee producers (Beer et al., 1998; Lyngbæk et al., 2001; Vaast et
al., 2005). But many other benefits of agroforestry systems have been identified. Siles (2007)
classifies them into two groups (1) effects on micro-environment and (2) effects on the crop
1 Soil-Vegetation-Atmosphere Transfer Model
11
and its management. The first group includes protection of biodiversity, water
quantity/quality control, soil erosion and lixiviation control, and carbon sequestration; while
the second encloses enhancements in soil fertility, influence on microclimate and on yield and
quality of coffee. The benefits in the first group coincide with the three main ES that are
under implementation in Costa Rica, through the already discussed forestry law no. 7575, i.e.,
(1) carbon sequestration, (2) water and soil conservation and (3) preservation of the
biodiversity. A recent European project, CAFNET2 has been dedicated to the study of ES for
the coffee sector.
1.3 Linking coffee HES with hydropower requierements in Costa Rica
Coffee plantations are present in four main basins used for hydropower generation in Costa
Rica (Reventazón, Grande de Tárcoles, Parrita-Pirrís and Grande de Térraba) (see Fig. 14),
but also in others with minor hydroelectric facilities. This makes evident the possible trade-
offs of the payment for HES from hydropower producers to coffee farmers. Water and soil
conservation are important ES that can be traded under the existing legal framework, between
this two main and traditional sectors of the country. But the only guaranty of sustainable
funding for such schemes relies on a real understanding and knowledge of the cause-effect
relationships between practices and services, as well as on formal monitoring methodologies,
clearly addressed to the measurement of the critical targeted variables. In case that it can be
demonstrated by scientific means, the payment for the provision of HES by a currently
running agro-ecosystem might also be discussed, in the sense that these potential payments
could persuade the owners of these systems to maintain the existing land use and agricultural
practices, keeping the environmental impact as low as it is.
2CAFNET: (EuropAid/121998/C/G: 10/2006-09/2009) “Connecting, enhancing and sustaining environmental
services and market values of coffee agroforestry in Central America, East Africa and India.” Coordinator
Philippe Vaast.: http://www.worldagroforestry.org/eastafrica/programs/cafnet
12
Figure 14. Map of Costa Rica. Coloured areas indicate the seven coffee producing regions (from norwest to southeast: North Zone, Occidental, Central Valley, “Los Santos”, “Turrialba”, “Perez Zeledón” and “Coto Brus”). Dark brown areas denote coffee plantations. Black thick lines delimit the thirty four drainage basins of Costa Rica. Green points and red lines indicate the presence of hydroelectric plants or projects. Sources: SIGICAFE - Icafé and ICE (2009).
1.4 Concluding remarks
The Environmental Services, in general, and the Hydrological Environmental Services (HES),
in particular, seem to currently play an important role, and have outstanding perspectives in
the sustainable development of a tropical humid country like Costa Rica, where hydropower,
agricultural/coffee production and ecologically-centred activities are of main importance in
terms of human and economic development. The quantification of the HES provided by
currently existing ecosystems, as well as those supplied through the implementation of
management programs, should be objectively quantified and monitored in each particular
context, avoiding the use of ambiguous or universal hypotheses, in order to avoid inaccurate
assessments promoting unfair, myth-based and/or unsustainable socio-economical agreements
for the payments of HES.
Reventazón
Grande de Tárcoles
Pirrís-Parrita
Grande de Térraba
13
1.5 Thesis hypotheses
Three main hypotheses are proposed:
Land use (e.g. coffee) and management (e.g. agroforestry, roads) influence water balance,
erosion processes and water quality
The partition between infiltration and runoff is a key process for HES, and is affected by
land management
The response of a basin can be very different to the integration of individual plots responses
1.6 Thesis objectives
To study ecophysiological, hydrological and sediment processes in a coffee agroforestry
basin:
To measure the main water and sediment fluxes from plot to basin
To develop a generic eco-hydrological model for HES assessment
1.7 Thesis overview
A general outline of the thesis is presented below:
Chapter 1: Introduction
This chapter exposes the relevance of the theme under study and how the Hydrologic
Environmental Services can link two traditional sectors in Costa Rica, the hydropower
and the coffee producers, by targeting the reduction in sediment yield. A brief
description of the research context is also presented here.
Chapter 2: Processes and models: ecophysiology, hydrology and sediments
Elementary concepts on ecophysiology, hydrology and sediment processes are
introduced providing some equations and simple models as a theoretical background for
the following chapters, and helping to understand the links between these disciplines.
14
Chapter 3: Bibliographic review on ecophysiological, hydrological and
sediment studies, with focus on tropical and coffee!based systems
This chapter presents some experimental approaches and findings of previous studies
dealing with the main topics of the thesis. First, a literature review on water balance
studies is given for both, plot and basin scales. Second, measurement and modelling
works on plot erosion and basin sediment yield are presented, discussing the upscaling
issues between the two scales. General conclusions are finally drawn from this review.
Chapter 4: Study site, experimental setup and data
This chapter provides a description of the local conditions in the study basin, explaining
all the experimental devices that were installed to measure the water and sediment
fluxes in the atmospheric, soil and aquifer systems. General information is provided
regarding the data recorded in the experimental basin.
Chapter 5: Modelling the hydrological behaviour of a coffee agroforestry
basin in Costa Rica
The chapter presents the results of the main hydrological and ecophysiological
measurements in the basin, describes the lumped3 conceptual4 Hydro-SVAT model
designed to reproduce the eco-hydrological behaviour of the system, and provides
discussions on the application of the model, the resulting hydrological processes, the
closure of a complete water balance and the implications of these results on the
provision of Hydrological Environmental Services.
Chapter 6: Water and sediment yield in a coffee agroforestry system at
various spatio!temporal scales: from plot to basin and from event to annual scale
This chapter presents the measured water and sediment yields in the experimental basin
during both, storm events and the entire year of recording. Equivalent measurements are
3 A lumped model is one that treats the entire basin as a complete unit, producing a unique integrated response
hydrograph for the surface, non-saturated and saturated lateral flows, as well as for evapotransporation fluxes.
4 A conceptual model is an intermediate scheme between theoretical (physical) and empirical models.
Hydrologic models are here considered as conceptual if the processes linking input and output variable(s) have
some physical meaning, but they do not intend to provide a detailed description of the relation between such
variables.
15
reported for two experimental plots located inside the basin. Cross comparisons are
made in time and space scales. These comparative analyses are used to track the spatial
sources of erosion. The associated uncertainty of the sediment estimations is calculated
using standard metrological approaches, newly applied to sediment analysis.
Chapter 7: Improving water and sediment peakflow simulations from an eco!
hydrological model, by including the effect of direct road!runoff: continuous
versus event!based validation strategies
The chapter reports the results of an attempt to improve peakflow simulations from the
original Hydro-SVAT model developed in Chapter 5. A road interception routine is
added to a simplified version of the model, while an original validation approach is
adopted to evaluate the model performance on continuous-time as well as over storm
events.
Chapter 8: General conclusions and perspectives
This final chapter provides a summary of the main findings and conclusions drawn from
the thesis, and exposes the perspectives concerning the use of combined experimental
and modelling methods to asses Hydrological Environmental Services.
16
17
CHAPTER 2
Processes and models: ecophysiology, hydrology and sediments
A review of the main ecophysiological, hydrological and erosion/sediment processes is
presented in this chapter, together with the main statements of some models of common
application in the respective fields. The final section presents some concluding remarks on
this topic.
The three themes here presented are frequently studied by different disciplines, even though
they are constituent elements of a continuous natural ecosystem, here delimited by the water
divider of a drainage basin system. The hydrological cycle, in which the water moves
throughout the atmospheric, vegetative, soil and subsoil media, counts with an enormous
amount of possible flow paths and reservoirs, defining several possible processes and storage
units. The interaction of water with soil, in particular, drives soil loss fluxes that behave in a
highly complex nature. A simplification of the reality is then imperative, and several schemes
have been developed for that purpose. In the frame of this thesis, we will focus on the water
components presented in Fig. 15, on which the ecophysiological, hydrological and sediment
processes and models to be introduced in the next three sections, rely on.
Figure 15. Water balance partition, presenting the main eco-hydrological processes in a drainage basin. R: rainfall, In: interception, T: tree transpiration, Th: throughfall, St: stemflow, Eu: soil evaporation, RE: root extraction, SR: surface runoff, i: infiltration, D: drainage, IF: interflow (non-saturated flow), BF: baseflow, DP: deep percolation and Q: streamflow.
18
2.1 Ecophysiology
In this section, the water balance is first explained by its constitutive equations through the
atmospheric, vegetative and soil media. Second, a set of models is given to explain the energy
balance and the evapotranspiration process, followed by short discussion on the integration of
the different components through Soil-Vegetation-Atmosphere Transfer (SVAT) models.
2.1.1 Definitions of terms and equations of processes
A. The complete ecosystem water balance equation
The general soil water balance equation is including a total of 10 variables (Allen et al.,
1998):
DpEsTRoInSFCrIrR !"""#!$ (1)
where (all expressed in mm) !$: variation of the soil water stock, R: precipitation, Ir:
irrigation, Cr: capillary rises, !SF: difference between entering and outgoing lateral
subsurface flow, In: interception, Ro: surface run-off, T: transpiration, Es: soil evaporation
and Dp: deep percolation.
The interception term is given by;
ff STRIn # (2)
where (all expressed in mm) R: precipitation, In: interception, Tf: throughfall and Sf:
stemflow.
The definition of runoff can vary much between the papers. For pure hydrologists, it generally
means the sum of surface plus subsurface lateral drainage, and corresponds to all that is not
evaporated or deeply percolated (e.g. Farley et al., 2005), which flows out as streamfow. For
ecophysiologists, runoff rather means surface runoff.
B. Soil water content, dry bulk density, porosity
- Soil gravimetric water content ($g) refers to the mass of water per unit of dry soil (gH2O
g-1 drysoil). It can be computed as:
19
% &drysoil
drysoilwetsoil
gM
MM #$ (3)
with M, the mass of soil.
- Dry bulk density ('db: unit gsoil cm-3): it is the density of undisturbed dry soil (sampled
horizontally along the profile of a trench, using a cylinder of known volume, V):
cylinder
drysoil
dbV
M#' (4)
-Soil volumetric water content ($v) refers to the volume of water per unit of soil volume
(m3H2O m-3
soil). $g is converted into $v following:
dbgv '$$ (# (5)
C. Soil Relative Extractable Water (REW), critical REW and soil water deficit (SWD)
The extractable water (mm) is the difference between the current $ and the wilting point
value)*or else the minimum value of $)*%$m&) with*z being the soil height in (m):
% & 310zEW m$$ # (6)
The maximum extractable water (mm) is the difference between the field capacity $ and the
wilting point, or else, the minimum value of $:
% & 310zEW mfcM $$ # (7)
The relative extractable water (0 < REW < 1) is defined by the ratio between EW and EWM:
% &
mfc
m
MEW
EWREW
$$
$$
## (8)
The critical REW (REWc) can be defined as the threshold value for which the ratio between
actual transpiration and potential evapotranspiration (PET), i.e. T/PET, starts to decrease with
decreasing REW. Above REWc, T/PET is assumed to remain relatively constant with REW
(see Fig. 16). REWc has often been reported to be around 0.4 (e.g., Granier et al., 1999).
20
2.1.2 Models
According to Dufrêne et al. (2005), hydrologic processes control drought effect on
photosynthesis soil carbon and nitrogen dynamics (Parton et al., 1987), and thus some of the
forest models also couple the carbon budget with a model simulating the water cycle.
Figure 16. Ratio T/PET calculated from sap flow measurements in an oak stand as a function of relative extractable water (REW) calculated from neutron probe measurements (from Bréda and Granier, 1996). Two data sets are reported: LAI=6 (black circles) and LAI=4.5 (open triangles). The dotted line shows the critical REW (REWc). (Figure extracted from Granier et al., 1999)
The rainfall reaching the ground is shared out into soil evaporation, transpiration,
interception, infiltration or runoff. According to the application domain, the hydrology models
are more or less sophisticated. For example, the evapotranspiration can be calculated as
function of a potential evaporation (Granier et al., 1999) or estimated following Monteith
(1965) or Shuttleworth and Wallace (1985). The soil can be divided into numerous layers
(Braud et al., 1995) or parameterized into one or several buckets (Eagleson, 1978).
Nevertheless, detailed SVAT models are difficult to use for the investigation of the spatial
and temporal variability of land surface fluxes. The large number of parameters they involve
requires detailed field studies and experimentation to derive parameter estimates (Boulet et
al., 2000). Simple water balance models using simple soil and stand parameters and basic
climatic data are often sufficient to predict temporal variation in soil water content (Granier et
al., 1999).
21
A. General energy balance model
APn HGEHR """# + (9)
where Rn: net radiation, H: sensible heat flux, +: latent heat flux for water evaporation, E:
evaporation, G: soil heat storage and HAP: air and plants heat storage.
B. Potential evapotranspiration models (PET)
Physical formulas (e.g. Penman, 1948):
This formula was built initially to estimate evaporation by water surfaces, or by surfaces
saturated in H2O and without stomatal resistance.
)(
))(()(
,+
,
"!
" (!# asn eeufGR
PET (10)
where PET: reference evapotranspiration, !: slope vapour pressure curve,*Rn: net radiation at
the crop surface, G: soil heat flux density, ,: psychrometric constant, f(u): a wind function
with the form f(u) = a + b(u) and es-ea: vapour pressure deficit.
Biophysical-formulas: Penman-Monteith reference evapotranspiration (ET0) (Allen et al.,
1998). The FAO Expert Consultation on Revision accepted the following unambiguous
definition for the reference surface: “A hypothetical reference crop with an assumed crop
height of 0.12 m, a fixed surface resistance of 70 s m-1
and an albedo of 0.23”. The reference
surface closely resembles an extensive surface of green grass of uniform height, actively
growing, completely shading the ground and with adequate water.
The FAO Penman-Monteith reference PET equation is a simple representation of the physical
and physiological factors governing the evapotranspiration process. From it, one may
calculate crop coefficients Kc, i.e. Kc = ETc ET0-1
.
Daily time!step (Allen et al., 1998, p. 24)
% & % &
% &2
2
3401
273
9004080
u.
eeuT
GR.
ETasn
o""!
"
" !
#,
, (11)
22
where: ET0: reference evapotranspiration (mm d-1
), Rn: net radiation at the crop surface (MJ
m-2
d-1
), G: soil heat flux density (MJ m-2
d-1
), T: mean daily air temperature at 2 m height
(°C), u2: windspeed at 2 m height (m s-1
), es: saturation vapour pressure (kPa), ea: actual
vapour pressure (kPa), es–ea: vapour pressure deficit (kPa), !: slope vapour pressure curve
(kPa °C-1
) and ,: psychrometric constant (kPa °C-1
).
Hourly time!step (Allen et al., 1998, p. 74)
% & % &- .
% &2
2
3401
273
374080 0
u.
eeuT
GR.
ET
aThr
hr
n
o""!
"
" !
#,
,
(12)
where ET0: reference evapotranspiration (mm h-1
), Rn: net radiation at the crop surface (MJ m-
2 h
-1), G: soil heat flux density (MJ m
-2 h
-1), T: mean hourly air temperature (°C), u2: average
hourly windspeed at 2 m height (m s-1
), e0
(Thr): saturation vapour pressure at air temperature
(kPa), ea: average hourly actual vapour pressure (kPa), !: slope vapour pressure curve (kPa
°C-1
),*,: psychrometric constant (kPa °C-1
).
C. Soil water models and their integration with plant and rainfall interception
models
Soil water transfer models can be classified into three categories: (i) stochastic models
(probabilistic approach, models very specific to the local conditions and poorly transferable),
(ii) functional deterministic models (capacitive models, using two values of $v, the wilting
point and the field capacity), and (iii) deterministic mechanistic models which couple several
phenomenons (Sansoulet, 2007). Deterministic mechanistic models offer the possibility to
study interactions between phenomenons, or sensitivity to phenomenons. However, their
potentialities are restricted by the number of parameters and initial conditions required for
simulation. Their validation is also difficult. Some successful cross-comparison of functional
models FAO (Doorenbos and Pruitt, 1977) or Ritchie (1985), with a mechanistic model
(Maraux and Lafolie, 1998) are available in the literature (e.g. Maraux et al., 1998).
There is an increasing demand for large scale (region, continent) and for long term studies on
forest-site-climate interactions, as much for hydrological as for forest management purposes.
23
Such extensive applications require robust water balance models using simple soil and stand
parameters and basic climatic data, in order to run simulations over many years.
D. Soil!Vegetation!Atmosphere Transfer models (SVAT) models
Many SVATs appear rather unbalanced, in terms of modelling effort allocated rather to the
climate, the soil or the plant. In this case, it would be recommended to couple the module of
interest with complementary modules, originating from complement models.
It is of particular importance that SVAT models include a retro-action loop between soil water
availability and stomatal conductance, in order to limitate transpiration when soil dries out.
All models presented below propose a form of retroaction, or allow including one.
Reference SVATs (Multi!layer and 3D architectural models)
Soil-Vegetation-Atmosphere Transfer models (SVAT) can be classified into (i) “reference”
models, such as multi-layer (e.g. Baldocchi and Harley, 1995), discrete 2D and 3D models (de
Reffye et al., 1988; Dauzat et al., 2001; Sinoquet et al., 2001), or else (ii) “simplified
models”, such as single-layer big-leaf models (Sellers et al., 1992; Amthor, 1994; Lloyd et
al., 1995) or sun-shade models (single-layer with two leaves: Sinclair et al., 1976; Norman,
1980; de Pury and Farquhar, 1997; Leuning et al., 1998; Wang and Leuning, 1998). The
respective (dis)advantages of those models were often discussed (e.g. Raupach and Finnigan,
1988; de Pury and Farquhar, 1997; Leuning et al., 1998). It is generally assumed that
reference multi-layer and 3D models are accurate but require a number of calculations, which
becomes a drawback for their inclusion into Global Circulation Models (e.g. Sellers et al.,
1992) and Regional Transport Models. 3D models for instance, rely upon a detailed
representation of canopy architecture. They are inherently the most promising for solving
recurrent theoretical problems affecting non-ideal canopies: e.g. row structure with large gaps,
azimuthal heterogeneity, complex leaf angle distribution function (LADf), distinction between
green and non-green elements or aggregation (clumping) at different scales (Weiss et al.,
2004). However, with regard to their complexity and to the number of parameters required,
3D models were made available only for few plant species, and their applications remained
seldom so far. We argue here that 3D models do offer an outstanding reference for testing the
performances or for validating simplified models to be applied in non-ideal conditions, the
latter models being more promising for general applications.
24
Example of a balanced H2O model: BILJOU (Granier et al., 1999)
BILJOU (bilan jour) is a daily water balance model where the main aim is to quantify drought
intensity and duration in forest stands. This model considers the plot or the stand as an
undivided entity, and use lumped values of input variables and parameters. It has a simple
structure and aims basically at quantifying drought intensity and duration in forest stands. It
uses a small set of parameters and standard daily meteorological data (potential
evapotranspiration and precipitation). It can also be run intra-daily. Evapotranspiration, which
is generally the largest flux component, besides throughfall, is estimated here from
ecophysiological relationships at stand scale. These relationships are driving by maximum
leaf area index (LAI) to calculate canopy transpiration, understorey evapotranspiration and
rainfall intercepted by tree canopies. It can be used for different sites and purposes, like long
term (several years) or short term studies on the impact of drought on forest stands. This
model estimates the main terms of the hydrological cycle in forest stands: soil water content,
stand transpiration and interception, drainage. Runoff is neglected in the original version. It
allows also computing seasonal and annual integrated water stress indices to characterize
drought events affecting physiological processes and growth of trees from various species,
over long term periods. Water stress is assumed to occur when relative extractable soil water
(REW) drops below a threshold of 0.4 under which transpiration is linearly reduced due to
stomatal closure. Day-to-day estimates of soil water content during the growing season allows
to quantify duration and intensity of drought events, and to compute stress indexes.
This model is iterative and the variation in soil water content is calculated at a daily pace as
(list of symbols and variables given in Table 1):
DEuTInPW #! (13)
where !W is the change in soil water content between two successive days.
Note here that runoff is considered negligible.
The BILJOU model combines several advantages of simplicity, few number of parameters
required, and compliancy with spatialization purposes (spatializable parameters). Its
calibration requires soil water monitoring along a vertical profile, together with soil water
potential (measured or calculated). The parameter “rm”, which represents Tmax/PET in
absence of water stress has to be calibrated for the stand (using sapflow during the wet period
25
for instance), unless the proposed relationship between rm and LAI is used (Granier et al.,
1999). Similarly, the relationship between “r” (T/PET) and REW in conditions of water
shortage, the critical value of REWc, as well as an empirical relationship between In and P
have to be documented locally.
Table 1. List of symbols and variables in BILJOU (Granier et al., 1999).
Symbols Variable
R Rainfall
Th Throughfall: R-In
In Rainfall interception
Is Water stress index
T Tree transpiration
Eu Evaporation from understorey plus soil
PET Potential evapotranspiration (Penman formula)
ET Actual evapotranspiration: T+Eu+In
r Equals T PET-1
rm Equals Tm PET -1
in absence of water stress
W Available soil water
Wm Minimum soil water (i.e., lower limit of water availability)
WF Soil water content ar field capacity
EW Extractable water
EWm Maximum extractable water
REW Relative extractable water
REWc Critical REW, at which tree transpiration begins to decrease
SWD Soil water deficit
Di Drainage at the bottom of soil layer i
mici Microporosity of soil layer i
maci Macropososity of soil layer i
Fi Water flow refilling soil layer i
LAI Leaf area index
K Extinction coefficient for PAR
fTPAR Fraction of transmitted PAR
Rn Net radiation
Tcor Transpiration diminished by 20% of In
Rnu Net radiation of the understorey
AEu Available energy of the understorey
REi Root extraction in layer i
RfDi Fine root relative distribution in layer i
26
It is assumed that the fraction of water uptaken by the roots in every soil layer is proportional
to their respective fine root density. Also the evapotranspiration of the understorey is a fixed
fraction of its available energy.
Once calibrated and validated, we believe that this simple model offers good perspectives for
a wide range of situations. It deserves to be adapted for multistrata systems and to be coupled
to a hydrological model.
2.2 Hydrological processes and models
Generally, hydrological modelling is based on a coupled approach of a production function
which separates the total rainfall into evapotranspiration, runoff and infiltration, and a transfer
function which routes water on hillslopes and through the channel network (see Figure WB).
The coupled production/transfer function can be conducted under a lumped or a spatially
distributed approach. Herein, we analyse the various approaches to model both production
and transfer function, and then we analyse the structure of both lumped and distributed
models.
2.2.1 Hydrological processes
A. The production function: modelling runoff/infiltration/evapotranspiration
The production function separates the rainfall into evapotranspiration (see the
ecophysiological model above), surface runoff and infiltration. Simulating the infiltration
process can be done either using physically-based models (e.g. Green and Ampt, 1911;
Richards, 1931), conceptual models (e.g. Diskin and Nazimov, 1995), or empirical relations
(e.g. Horton, 1933; SCS, 1972). These models have been developed either directly at a global
basin scale (e.g. SCS, 1972; Diskin and Nazimov, 1995) or use infiltration models developed
at the local scale of soil columns (e.g. Green and Ampt, 1911; Richards, 1931). Few models
were specifically designed for the plot scale. In the context of the distributed hydrological
modelling of farmed basins, the plot is the largest homogeneous unit with regard to crop type
and soil surface conditions. Therefore, some authors tried to adapt existing models to this
particular scale (Chahinian et al., 2005, 2006).
27
The runoff or rainfall excess re(t) [LT-1
] and the infiltration rates i(t), at any time t, depend on
the value of the potential infiltration rate f(t), and on the value of rainfall intensity R(t) [LT-1
],
at the time considered:
% & % & % & % & % & 0andif ##/0 trtRtItftR e (14)
% & % & % & % & % & % & % &tftRtrtftItftR e ##/1 andif (15)
Many factors influence the potential infiltration rate, including the soil surface features, the
proportion of soil covered by vegetation, the current water content, the soil hydrodynamic
properties, namely, the saturated hydraulic conductivity, the hydraulic conductivity-soil
moisture relation, the soil retention curve, and the initial water content at the beginning of the
rainfall event. All infiltration models integrate this information using various parameters and
variables whose values need to be determined either by measurements or calibration.
B. The transfer function: Routing the flow on hillslopes and through the channel
network
Once infiltration and surface runoff calculated, a transfer function is used to route the excess
water on hillslopes and through the channel network to the outlet. Transfer functions can
range from “simple” unit hydrographs (Sherman, 1932) to more complex formulations such
as the kinematic or diffusive wave equations (Moussa and Bocquillon, 1996), or the complete
Saint-Venant (1871) equations.
The most popular approaches to modelling fluvial hydraulics have been one-dimensional
solutions of the full Saint-Venant (1871) equations which remain the most powerful and
accurate models largely used by many engineering packages. Often, the full Saint-Venant
description of shallow water flow is unnecessary, and modellers have therefore simplify the
full Saint-Venant equations by neglecting terms in the momentum equation whenever
justified by the physical conditions (Moussa and Bocquillon, 1996). In most of the practical
applications of flood routing in natural channels, the system is reduced to the diffusive wave
equation. In the particular case when the pressure-gradient term could also be neglected, the
diffusive wave becomes the kinematic wave model. Thus, the diffusive wave can be
considered a higher order approximation than the kinematic wave approximation. The choice
of a kinematic or diffusive wave is often made on pragmatic grounds in that full Saint-Venant
28
equations need complex numerical approaches for the resolution of the differential equations,
and would be too computationally intensive.
Therefore, the choice of the appropriate transfer function (Saint-Venant, diffusive wave,
kinematic wave, unit hydrograph, etc.) by the modeller, depends on the terms that can be
neglected in the full Saint-Venant equations, but also on the availability and the accuracy of
input data such as the topography (cross section shape, slope), the hydraulic properties of
river reaches (Manning roughness coefficient), and the input flood hydrographs. For all these
reasons, recent researches in hydraulics examine reduced complexity approaches.
2.2.2 Models
We distinguish lumped and spatially distributed approaches.
A. Lumped rainfall!runoff models
Since the 1960s, lumped, conceptual rainfall-runoff models have been used in hydrology
(Crawford and Linsley, 1966; Perrin, 2000). These models consider the basin as an undivided
entity, and use lumped values of input variables and parameters. For the most part (for a
review, see Fleming, 1975; Singh, 1995), they have a conceptual structure based on the
interaction between storage elements representing the different processes, with mathematical
functions to describe the fluxes between the stores. Generally, automatic techniques are used
to calibrate the model parameters.
B. Spatially distributed hydrological models
To predict the effects of land use change in agricultural basins, one needs distributed models.
Although it was once sufficient to model basin outflow, it is now necessary to estimate
distributed surface and groundwater flow characteristics, such as flow depth, flow velocity
and water table level. These flow characteristics are the driving mechanisms for sediment and
nutrient transport in landscapes and unless they can be predicted reasonably well, water
quality models cannot be expected to adequately simulate sediment and nutrient transport.
Some examples of models based on this structure include the Système Hydrologique
Européen model (SHE) (Abbott et al., 1986), HYDROTEL (Fortin et al., 1995), the Limburg
29
Soil Erosion Model (LISEM) (De Roo and Offermans, 1995) or MHYDAS (Moussa et al.,
2002).
When using a distributed model, it is also crucial to adopt rigorous parameterisation,
calibration and validation procedures. The parameterisation procedure aims at assessing most
of the parameter values directly or indirectly from field data and identifying a minimum
number of parameters to be adjusted through the calibration process (Refsgaard, 1997) in
order to avoid over-parameterisation (Beven, 1996). However, the use of complex distributed
models is justified only if their distributed results can be used with some confidence. Together
with a restricted use of calibration, a multi-site, multi-scale and multi-criteria validation is
probably a good way to further constrain distributed models.
2.3 Erosion and sedimentology
Plot (hillslope) erosion is commonly studied on three main elements: interrill regions, rill
regions and gullies. A simple definition of a rill is a channel that can be removed through
ploughing, while a gully has a larger dimension. A review of these types of erosion is
presented below.
2.3.1 Plot interrill (sheet) erosion
Both, soil particle detachment by rainfall and erosion transport inside the plot by rainfall or
surface runoff, are given is separate subsections.
A. Interril soil detachment by rainfall
The dominant mechanism influencing interril erosion is soil particle detachment by raindrop
impacts. The rate of detachment depends on the rainfall energy and the slope gradient and it is
inversely proportional to soil shear strength and to the protective effects of water layers on the
ground (when surface runoff occurs). Some relationships derived to estimate the rainsplash
detachment are presented below, regarding the effects of these four variables.
Rainfall kinetic energy
Concerning rainfall kinetic energy, its value is considered to be several thousands times that
of shallow interrill flow, reaching values up to 6MPa (Ghadiri and Payne, 1981). Either as a
30
function of rainfall energy or of rainfall intensity, rates of rainsplash could be obtained
according to:
b
R EaD # (16)
d
R IcD # (17)
where DR: rainsplash detachment rate (kg m-2
s-1
), a and c: soil erodability coefficients, E:
rainfall energy (J m-2
s-1
), I: rainfall intensity (mm s-1
) and b and d: exponents, the latter
generally assumed to equal two (Sharma et al., 1993). Given that the link between rainfall
energy and intensity has been questioned, Sharma et al. (1995) suggest the next equation to
calculate rainfall kinetic energy:
2#
#N
i
iii nvdkE1
23 (18)
Where k: product of all constants in the equation (dimensionless), N: number of drop diameter
classes, di: mean diameter of drops in class i (mm), vi: impact velocity of drops of diameter di
(m s-1
) and ni: number of drops in class i.
Slope gradient
A reduction of resistance to detachment with gradient, varying with soil cohesion and angle of
friction, was reported by Torri and Poesen (1992). Soils with a low angle of friction (~25º),
the detachment rate doubles with an increase in gradient from 3º to 19º. In contrast, for soils
with a high angle of friction (~55º), there is only a 10% increase in detachment over the same
range of gradient. The influence of slope on soil particle detachment is sometimes neglected
in physically-based models, working at plot and basin scales, because of the difficulty of
characterizing the ‘effective slope’, which needs to be measured over distances of several
drop diameters from the point of raindrop impact. In addition, it is dependent upon local
surface roughness and the angle of internal friction of the soil (Quansah, 1981; Torri and
Poesen, 1992). This slope is not the same as the general surface slope, which is generally
smaller.
31
Soil shear strength
In relation to the soil detachment D (g), Cruse and Larson (1977) proposed that it depends of
the soil shear strength (g cm-2
) as:
% & 422 1000040092043376 3" # 44 ...D (19)
This implies that detachment increases with the sand (Quansah, 1981) and moisture content
(Truman and Bradford, 1990).
Layer of water on ground surface
Torri et al. (1987) demonstrated an exponential decrease in the detachment rate with
increasing depth of surface runoff. This is described by the equation:
hqeDD # 0 (20)
where D: rate of detachment (m3 s
-1 m
-1), D0: rate of detachment for water depth equal to zero,
q: an empirical constant that varies with soil surface condition, and h: depth of the water layer
on the ground (m).
B. Interril erosion transport
Transport by rainfall droplets
When large infiltration rates preclude surface runoff, droplet transport may be a more
important transport mechanism. All rainfall causes some splash transport. A sediment
transport model for splash has been proposed as follows (Poesen, 1985):
% & % &- . % &- .55'
5 sin422222
50 10190sin3010 ..
dD
s eD..R
cosEq "# (21)
where qs: net downslope splash transport (m3 m
-1 s
-1), E: rainfall kinetic energy (J m
-2 s
-1), !:
slope (degrees), RD: resistance to detachment (J kg-1
), "d: bulk density (kg m-3
), and D50:
median grain size (m).
32
Transport by surface runoff
Although splash droplets transport some sediment, this mechanism has frequently much less
transport capacity than that provided by flow energy. The sediment transport capacity has
been suggested to behave according to the next equation (Everaert, 1991):
cb
s Daq 6#7 (22)
where q’s: interril flow sediment transport capacity (g cm-1
s-1
), a: constant (dimensionless),
: Bagnold’s modified stream power (g1.5
s!4.5
cm!0.67
), D: particle size ("m), and b and c
(dimensionless) are exponents that vary with particle size.
The effects of particle size, slope and discharge on transport capacity of surface runoff are
taken into account by the equation (Agarwal and Dickinson, 1991):
dcb
s DSqaq 50#7 (23)
where q’s: sediment transport capacity (kg m-1
s-1
), q: unit discharge (m-1
s-1
), S: slope (mm
mm-1
), D50: median particle size (mm).
Like the detachability, the transport distance of the particle is a function of its size. Parsons et
al. (1998) proposed the next general equation to predict travel distances of particles ranging
from 3 mm to 10 mm in diameter, under rainfall intensities up to 138 mm hr-1
falling onto
shallow flow, up to 5 mm deep:
M
EE.L
.
F
. 98103525250# (24)
where L: travel distance per time unit (cm min-1
), E: rainfall energy (J m-2
s-1
), EF: flow
energy (Jm-2
s-1
), and M: particle mass (g). Parsons and Stromberg (1998) developed a
relationship for L against particle diameter D (cm), showing that travel distance declines with
particle size:
4439970 .DL # (25)
2.3.2 Plot rill erosion
The acceleration of interril processes lead to flow detachments shaping well defined channels
or rills on the hillslope plots.
33
A. Rill shear velocity
A shear velocity of around 3-3.5 cm s-1
has been estimated as necessary for rills to develop.
Rauws and Govers (1988) proposed the next relationship between the critical grain shear
velocity and soil cohesion:
C..uc 560890 "# (26)
where uc: critical grain shear velocity (cm s-1
) and C: soil cohesion (kPa). However, as
Nearing (1991) has pointed out, although the shear stress exerted by shallow flow around
such a threshold is of the order of a few pascal, the typical shear strength of soils is several
kPa. Furthermore, Nearing (1994) conducted laboratory experiments that demonstrated that,
for similar mean shear stress, sediment entrainment by laminar flow is negligible compared to
that for turbulent flow.
B. Rill shear stress
The rill shear stress of the flow (kg m-1
s-2
) can be obtained as (Govers et al., 2007):
SRg H'4 # (27)
where ": water density (kg m-3
), g: gravitational acceleration (m s-2
), RH: hydraulic radius of
the rill (m) and S: slope gradient (m m-1
). The hydraulic radius can, in turn, be estimated as:
1 # prH WAR (28)
where Ar: wetted cross section area (m2), which equals to the flow width multiplied by the
flow depth (assuming a rectangular rill cross section); and Wp: wetted perimeter (m), equal to
the flow width plus two times the flow depth (under the same assumption).
C. Rill soil detachment
The most widely applied relationship to predict rill detachment capacity is based on the
excess of shear stress over a critical value applied by the concentrated flow (Foster et al.,
1981; Govers et al., 2007):
% &bcc aKD 44 # (29)
34
where Dc: amount of sediment detached per unit of bed surface per unit of time (kg m-2
s-1
),
K: soil erodibility factor (s m-1
), : shear stress of the flow (kg m-1
s-2
), c: critical shear stress
of soil (kg m-1
s-2
) and a, b: constants (dimensionless).
2.3.3 Gully erosion
Gully erosion originates from the increase in flow energy and consequently, of rilling
detachment. Gullies reach such a depth that is not possible to remove them by ploughing.
Foster (1986) suggested that gully erosion may be of equal importance to the sum of rill and
interrill erosion. Poesen et al. (2003) show that gullies may account for between 10% and
94% of total soil loss.
A. Critical flow shear stress
A relationship for the critical flow shear stress (Begin and Schumm, 1979) is given by:
SAc a
c ,4 # (30)
where c: critical flow shear stress (kg m-1
s-2
), c: constant (dimensionless), #: weight of water
per unit volume, A: drainage area (ha), a: exponent (dimensionless) and S: slope (m m-1
). The
properties A and S are linked, and they together define threshold conditions for gully
initiation. Locations with and without gullies may be defined in terms of a discriminant
function of the form:
bAaS # (31)
where a and b are simple regression coefficients. Such functions have been typified for
different sites and land-use conditions (Parsons, 2006).
B. Ephemeral gully width
Watson et al. (1986) developed a model to predict ephemeral gully width W (m) as:
024016038703960662 .
c
...
p SnQ.W # 4 (32)
35
where Qp: peak flow discharge (m3 s
-1), n: Manning’s roughness coefficient (dimensionless),
and c: critical flow shear stress (Pa). A simpler equation that applies to croplands is given by
(Nachtergaele et al., 2002):
4120512 .Q.W # (33)
C. Gully detachment
Poesen et al. (1998) related detachment rate in ephemeral gullies DR (kg m-2
s-1
) to
concentrated flow shear stress (N m-2
):
080750 ..DR # 4 (34)
2.3.4 The Universal Soil Loss Equation: an empirical approach
The Universal Soil Loss Equation (USLE) was developed in United States based on soil
erosion data, collected since the 1930s by the USDA Soil Conservation Service (Wischmeier
and Smith, 1960; Wischmeier and Smith, 1965; Wischmeier and Smith, 1978) and it is the
most employed method around the world (Nearing et al., 2006). The USLE was initially
based on statistical analyses of more than 10 000 plot-years of data, collected from natural
runoff plots located at 49 erosion research stations in United States. Additional data from
runoff plots and experimental rainfall simulator studies were incorporated into the final
version published in 1978 (Wischmeier and Smith, 1978), allowing to assess not only
cropland but other land uses. The model has a simple form, in which the soil loss is explained
by six factors:
PCSLKRA # (35)
where A: average annual interrill and rill soil loss over the plot area that experiences net loss
(t ha-1
y-1
), R: rainfall erosivity (MJ mm hr-1
ha-1
y-1
), K: soil erodibility (t hr MJ-1
mm-1
), L:
slope length factor (dimensionless), S: the slope steepness factor (dimensionless), C: cropping
factor (dimensionless), and P: conservation practices factor (dimensionless).
The standard plot on which the equation is based had a 9% slope, a 22.13 m length and (for
most of the early erosion plots) a 1.83 m width. Under these conditions, the LS factor equals
to one where the plot is fallow, tillage is up and down slope and no conservation practices are
applied (i.e. CP = 1). In this state: K = A/R. Wischmeier et al. (1971) presented a simpler
36
method to predict the soil erodibility K as a function of soil particle size, organic matter
content, soil structure and profile permeability. A nomograph was also provided in order to
obtain approximate K if this information is at hand. The LS factors can be determined from a
slope effect chart by knowing the length and gradient of the slope. The cropping management
(C) and conservation practices (P) factors should be determined empirically from plot data,
although some tables are already available for different crops and practices.
An important remark is that the USLE predicts soil loss but not basin sediment yield. In
addition, it predicts average annual soil loss, not soil loss for storm events or for individual
years. The USLE reduced a complex system to a quite simple one for purposes of erosion
prediction. There are many complex interactions within the erosional system which are not,
and cannot be, represented within the USLE. On the other hand, for the purposes of general
conservation planning and assessment, the USLE has been, and still can be, used with success
(Nearing et al., 2006).
The USLE was upgraded to the Revised Universal Soil Loss Equation (RUSLE1) during the
1990s (Renard et al., 1991; Renard et al., 1997) and to RUSLE1.06c in mid 2003 (USDA-
ARS-NSL, 2003). RUSLE1 is land-use independent and applies to any land use having
exposed mineral soil and Hortonian overland flow. In addition, RUSLE2 was also officially
released in mid 2003, and it is land-use independent too (Foster et al., 2001; USDA-ARS-
NSL, 2003). RUSLE1 and RUSLE2 are hybrid models that combine index and process-based
equations. RUSLE2 expands on the hybrid model structure and uses a different mathematical
integration than does the USLE and RUSLE1. Both are computer based, and have routines for
calculating timevariable soil erodibility, plant growth, residue management, residue
decomposition, and soil surface roughness as a function of physical and biological processes.
The two RUSLE models also compute deposition on concave slopes, for vegetative strips, in
terrace channels, and in sediment basins using process-based equations for transport capacity
and deposition. RUSLE2 also computes an enrichment ratio for the sediment leaving the end
of the slope. Enrichment ratio is the ratio of specific surface area of the sediment to specific
surface area of the soil subject to erosion, and is useful for assessments of chemical carrying
capacity of sediment. Both RUSLE models have new relationships for L and S factors which
include ratios of rill and interrill erosion, and additional P factors for rangelands and
subsurface drainage, among other improvements.
37
2.3.5 Sediment yield at basin scale
A. The sediment delivery ratio
The sediment delivery ratio is a relationship that was developed to facilitate the link-up
between potential erosion (sometimes obtained from plot estimations) and actual basin
sediment yield. The limitations of this practical concept were termed at the beginning of the
80’s by Walling (1983), highlighting the “blackbox” nature of the spatial and temporal
lumping processes and the lack of detailed empirical investigations, emphasizing the
characterization of pathways, sinks and time constants involved in the delivery process.
The sediment delivery ratio (SDR) (dimensionless) is defined as the fraction of total eroded
sediment within a drainage basin that finds its way to the basin outlet:
PE
SSYSDR # (36)
where SSY: suspended sediment yield at the outlet of the basin (t ha-1
y-1
) and PE: gross
potential erosion in the basin (t ha-1
y-1
).
Several relationships (usually inversely proportional) between this ratio and the basin area are
summarized in the study, but other basin properties like basin “relief”, basin length,
bifurcation ratio, slope of the main channel, curve number, annual runoff and gully density
have also been used as predictor variables. According to Walling (1983), although attractive
for its simplicity, the sediment delivery ratio concept involves numerous problems and there
is a need for a more refined approach. At that time, the developments in distributed modelling
of erosion and sediment yield seemed to represent a significant advance, but a simpler
approach embodying the improved knowledge associated with these developments was
considered an outstanding task for operational application.
B. Area!based estimations of sediment yield
de Vente et al. (2007) revisited the unsolved sediment delivery problem showing that, in the
absence of reliable spatially distributed process-based models for the prediction of sediment
transport, area-specific sediment yield has often assumed to decrease with increasing drainage
basin area, according to:
baSSY 8# (37)
38
where #: basin area (km2), and a and b are empirical parameters. In the case of an inverse
relationship between SSY and #, b adopts a negative value. This relationship, however, is not
a universal rule, and direct proportionality between SSY and # or even more complex
associations between these variables have been also reported among the study cases reviewed
by de Vente et al. (2007) (see Fig. 17). The authors remark that both erosion and sediment
deposition processes are scale dependent. Going from small (m2) to larger areas (below 1
km2) more erosion processes become active leading to a rise in sediment yield with increasing
basin area. However, for larger areas (above 1 km2) erosion rates generally decrease and
deposition in sediment sinks increases due to decreasing slope gradients, causing a decrease in
sediment yield with increasing basin area.
Land cover conditions and human impact determine if hillslope erosion is dominant over
channel erosion or vice versa. In the first case, sediment yield is expected to decrease with
increasing basin area, while in the latter case the yield will show a continuous positive
relation with the area.
Figure 17. Examples of Suspended Sediment Yield (SSY) as a function of basin area (A). (a) A
negative relationship is observed between A and SSY for several basins worldwide, grouped in s8
regions (each curve represents a region). (b) Example of positive relations between A and SSY for
several basins gruped in 11 regions. (Figure extracted from de Vente et al., 2007).
As an example, agricultural basins are frequently dominated by hillslope erosion processes.
Conversely, in basins with limited human impact and a well-developed vegetation cover,
channel erosion dominates. The role of specific erosion processes (i.e., splash, sheet, rill,
gully and mass wasting) on the relationship between area and sediment yield was also
39
explored. Only for very large areas (above ~104 km
2) a decrease in sediment yield is observed
when drainage density decreases or channel banks are stabilized. Finally, spatial patterns in
lithology, land cover, climate or topography can cause sediment yield to increase or decrease
at any basin area and can therefore result in non-linear relations with this variable. Altogether,
with increasing area often first a rise and then a decrease in sediment yield are observed.
These trends between area and yield can, however, be disturbed at any spatial scale due to
distinct spatial differences in soil erodibility, rainfall characteristics, land use or topography
within the basin, resulting in an increase or decrease in sediment yield independent of the
basin area. Furthermore, temporal variation in the susceptibility to erosion and sediment
transport may cause the area-yield relation to change over time. The authors conclude that
large regional, local and even temporal variability in the trend between area and sediment
yield implies that prediction of yield based on area alone is troublesome, and preferably
spatially distributed information on land use, climate, lithology, topography and dominant
erosion processes is required, as well as the awareness of the history of human impact and
geomorphic development in the studied basin.
2.4 Concluding remarks
This chapter provided the theoretical bases of several concepts on ecophysiology, hydrology
and erosion/sedimentology, most of them necessary to understand and model the eco-
hydrological behaviour of an agricultural basin, and to properly conceptualize and interpret
the experimental results of erosion and sedimentology field research.
On agricultural basins, man intervention can produce visible changes in the system
ecophysiology, interferring on hydrological processes, soil degradation and the final sediment
response of the basin. Then, any proposal dealing with agro-ecological practices introducing
further changes in such a system should be based on the understanding of the elemental
processes presented in this chapter. The development of new schemes for Hydrological
Environmental Services or Basin Management Programs, to be promoted at the basin scale
and intending the reduction of suspended sediment yield at the basin outlet, should be based
on these theoretical lines, addressing to a inter-disciplinary assessment of the analytical needs
in the operational context. A balanced trade-off across the different scientific approaches
should be achieved in the selection and design of several methodological aspects, like the
40
variables to measure, the spatio-temporal resolution of observational procedures, the
combination of analytical approaches and the modelling complexity.
Although the processes and analytical methods on ecophysiological and hydrological analyses
have clear and well defined bases, some confusion is often found concerning the operative
definition of erosion and sediment yield. While the first term is commonly employed to
identify (interrill and rill) soil loss processes at the plot scale, which can be seen at larger
scales like a “potential” soil loss from hillslope plots, the second term is more accepted for
basin scale assessments considering the “actual” soil loss from the system after depositional
processes). Although some general area-based upscaling procedures have been developed to
link plot erosion to basin sediment yield, many other factors (e.g., lithology, land cover, effect
of infrastructure, climate or topography) have to be seriously considered on that purpose.
41
CHAPTER 3
Bibliographic review on ecophysiological, hydrological and
sediment studies, with a focus on tropical and coffee!based
systems
As we discuss above, the evaluation and negotiation of HES demand a good conceptual
framework of the agro-ecosystem at both, the plot scale in which management practices are
usually implemented (especially the agricultural practices), and the basin scale that is the
elemental unit for the largest (and most profitable) water uses by man. The understanding of
the main water and sediment balances and processes is the first condition to assess the HES
variables targeted here. The knowledge of the environmental impacts of a given management
practice on those variables is also needed. Then, background knowledge concerning those
budgets, processes and environmental impacts can be obtained through an adequate revision
of previous research experiences. This bibliographic review is divided into two main topics,
the water balance and the sediment yield. Both topics are explored in two different spatial
scales: the agricultural plot and the drainage basin. Some studies comparing field
measurements at both scales are also reported here.
An important remark is that this review is focused on the understanding of tropical coffee
agroforestry plantations. Hence, studies carried out on these particular systems were
prioritized over those developed in other conditions. On the contrary, when no information
was found in coffee systems for a given topic (e.g. the measurement of basin sediment yield
by turbidimetric methods), bibliography dealing with other environments is presented.
3.1 Water balance from plot to basin: measurements and upscaling
The measurement and modelling of water balance (WB) has been widely studied through
multi-technique approaches and multiple scales, as presented below. We consider four main
themes: (1) water balance at plot scale, (2) water balance at basin scale, (3) water balance at
both, plot and basin scales, and (4) water balance at other scales, with different modelling
purposes.
42
3.1.1 Water balance at plot scale
At the plot scale many analyses applied to microclimatology and ecophysiology have defined
the WB as a budget in the region comprehended between the near-ground atmosphere (the
boundary layer) and the volume of soil colonized by roots. These approaches may include
components like precipitation, interception, throughfall, stemflow, surface runoff, infiltration,
soil water variation, soil evaporation, actual evapotranspiration and soil drainage.
This type of studies has different applications, for instance:
- Grier and Running (1977) obtained the WB in western Oregon (USA) in experimental
plots along transect lines crossing different climates and soils, from measurements of
growing season precipitation, open pan evaporation and estimates of soil water storage, to
demonstrate that WB drives community leaf area in coniferous forests with some
independence of species.
- Joffre and Rambal (1993) established a WB in the southwestern Iberian peninsula (Spain),
under savanna rangeland and mediterranean evergreen oak trees and mediterranean sub-
humid to humid climate. For this purpose, they based on measurements of rainfall and soil
water content and on drainage estimations, iteratively closing the WB to obtain the actual
evapotranspiration and surface runoff components (constraining the analysis by the Penman
potential evapotranspiration). The purpose of that study was to determinate whether isolated
trees modify the WB of a dehesa ecosystem.
- Granier et al. (2000) present a large amount of ecophysiological experiments in a beech
forest plot on luvisol and stagnic luvisol soils in Hesse, France. The measurements included
actual evapotranspiration (eddy covariance), leaf area index, throughfall, steamflow, soil
water content and sap flow, in order to develop a canopy conductance / transpiration model
for application at larger spatial scales.
- Granier et al. (1999) developed a WB model to quantify drought intensity and duration in
forest stands, based on both measured rainfall and calculated Penman potential
evapotranspiration, and parameters depending on leaf area index, maximum extractable
water, forest type, and optionally, soil characteristics and porosity.
43
Plot WB have been also measured and analyzed in tropical regions, in banana and coffee.
Here we present a selection of some WB studies carried out to assess tropical crops like
coffee and banana:
- Both Jiménez (1986) and Imbach et al. (1989) measured rainfall, throughfall, pan
evaporation, soil properties and water content, and made an estimations of actual
evapotranspiration and percolation, in order to quantify the WB of two coffee agroforestry
systems under different shade tree species: Cordia alliodora and Erythrina poeppigiana.
Both studies were developed in Turrialba, Costa Rica, in a humid tropical climate and under
coffee agroforestry systems on Inceptisols (Typic Dystropept).
Unlike the previous works, the next two studies taken into account the runoff formation
processes:
- Charlier et al. (2009) established the plot WB in a banana plantation in Basse Terre,
Guadeloupe, under a maritime humid tropical climate and on Umbric Andisol. The rainfall
and surface runoff were measured during storm events, and a model was developed to
assess the importance of the banana canopy interception and stemflow processes on surface
runoff response.
- Cannavo et al. (2011) measured rainfall, climatology, throughfall, stemflow, sapflow,
surface runoff and soil water content and estimated Penman-Monteith reference
evapotranspiration. They worked in a coffee agroforestry system shaded by Inga densiflora
in Barva, Costa Rica, under a tropical climate and on andisols (Dystric Haplustands). The
authors followed the ecophysiology over time and analyzed the soil properties. The model
Hydrus 1D was used to characterize the WB and to estimate drainage fluxes in shaded and
non-shaded plots, finding lower values at the former, which reduces the risk of groundwater
contamination by agrochemicals.
Finally, some WB studies provide specific comparisons of the surface runoff generation in
both, shaded and unshaded coffee plots:
- Harmand et al. (2007) installed 1×1 m2 plots in coffee agroforestry (AF) (E. deglupta
shade trees) and coffee monoculture (MC) plantations on acrisols (Ultisol), in San Isidro del
General, Costa Rica, under a tropical climate. They found slightly lower annual runoff
coefficients under shaded systems, in comparison to the unshaded ones.
44
- Siles (2007) used similar plots to measure the annual runoff coefficients in coffee
plantations in Barva, Costa Rica, featuring a tropical climate and andisols (Dystric
Haplustands. In this case surface runoff on AF plots was a half of that on MC plots.
- Verbist et al. (2010) present the results of runoff experiments in 4×10 m2 plots on
Cambisols in Sumberjaya, Indonesia, under a humid tropical climate. A ration of 0.5
between the runoff coefficients of AF and MC coffee plantations is reported.
3.1.2 Water balance at basin scale
A. In temperate climate
Water balance studies are summarized below, starting with measurements in developed
countries. Then, a similar selection is presented for tropical countries. Finally, a number of
studies focusing on the water yield of coffee basins are exposed.
- Dawes et al. (1997) studied a 109 ha cropping rotation basin in New South Wales,
Australia; calibrating a distributed ecohydrological model and validating it with field
measurements. They recorded rainfall, streamflow, climate, leaf area index, total
evaporation, and soil water content and properties. After closing the WB the streamflow
was found to be a very small component, and the authors concluded that the basin could
have been treated as a set of soil columns.
- Cook et al. (1998) closed the WB in a woodland 126 km2 basin in Darwin, Australia;
through the combination of micro-meteorological, soil physical and groundwater chemical
methods. They measured streamflow, overstory and total evapotranspiration (by using heat
pulse and eddy correlation, respectively), soil water content, matric suction and water table
variations. Groundwater dating with chlorofluorocarbons was used in this study to estimate
net groundwater recharge rates. The authors state that only by combining these independent
methods can the complete WB be determined, but they never considered the use of
modelling tools as a possible alternative to complement the observations from a simpler
measurement scheme.
- Jothityangkoon et al. (2001) studied a 2545 km2 semi-arid basin in southwestern Australia;
dominated by eucalypt forests, by measuring of rainfall, streamflow and potential
evaporation (using a class A pan). They proposed a top-down model building, starting with
45
a simple scheme to perform streamflow properties at the annual scale, and then increasing
the model complexity while decreasing time scales (monthly and daily). Here, for instance,
successive bucket models were tested, and actual evapotranspiration was assumed a fraction
of the potential one, as a function of the soil water content. The aim of this study was to
configure parsimonious models while identifying key hydrological processes at each time
scale.
- Querner (1997) proposed a physically based model that could be used under different
hydrological conditions, and able to simulate surface, non-saturated and saturated flows. Its
application in a basin at the east of the Netherlands was validated through independent
measurements of streamflow, soil water content, water table level and estimated
evapotranspiration, thus giving an overall WB.
- Ceballos and Schnabel (1998) obtained the WB of an experimental basin in Cáceres,
Spain; under a dehesa landuse, by obtaining the actual evapotranspiration as the difference
between the annual figures of measured rainfall and measured streamflow. They also
measured the soil water content and some soil properties.
- Wigmosta et al. (1994) developed a distributed hydrology-vegetation model including
canopy interception, evaporation, transpiration (two layer Penman-Monteith), and snow
accumulation and melt, as well as streamflow generation via the saturation excess
mechanism and saturated subsurface flow through a quasi three-dimensional routing
scheme. Digital elevation data are used to model topographic controls on the micro-
climatology and on downslope water movement. This model was applied in a 2900 km2
basin in Montana, USA, and calibrated using 2 years of recorded precipitation and
streamflow. The verification was made against 2 additional years of streamflow and against
high resolution radiometer based spatial snow cover data.
- Arnold and Allen (1996) applied their Soil and Water Assessment Tool (SWAT) to close
the WB at monthly and annual time steps in three experimental basins of Illinois, USA
(with areas between 122 and 246 km2). They calibrated the model using rainfall,
climatology and streamflow, and made the validation using measured data of several
hydrologic components including, surface/groundwater streamflow partition, surface and
soil evapotranspiration, groundwater evapotranspiration, groundwater recharge and the
change in the groundwater storage.
46
- Smerdon et al. (2009) worked on a 130 km2 semi-arid basin in British Columbia, Canada;
covered by pine, fir, and spruce forest lands. They measured precipitation, climatology and
streamflow, and estimated potential evapotranspiration by the Thornthwaite method. Actual
evapotranspiration, unsaturated flow and the resulting groundwater recharge rates were
modelled with the MIKE-SHE model for data-rich areas, while a simpler WB was used to
estimate the recharge in data-limited areas. These recharge rates were introduced then in a
larger-scale flow model to predict the groundwater recharge processes in the basin.
B. In tropical climate
Basin WB studies have also being developed in tropical countries. A summary of these
studies is presented below:
- Lesack (1993) obtained the WB of a 23.4 ha rain forest basin in the Central Amazon of
Basin, through the measurement of rainfall, streamflow and soil water content, and the
estimation of actual evapotranspiration as the WB residual. This author used a hydrograph
separation technique to quantify the annual storm flow during 173 storms over the entire
record, as no hillslope studies were carried out.
- Roberts and Harding (1996) closed basin WBs by measuring rainfall, streamflow and soil
moisture, this latter being used to calibrate an evapotranspiration model for different land
uses, whose outputs were then integrated on a percentage land cover basis to provide basin
scale estimates under a mixed land use. Drainage and forest canopy interception were also
estimated from soil moisture data. They applied this approach to three basins of the
Aberdare mountains in Kenya; covered by pine forest, bamboo forest and grass, with areas
from 36 to 65 ha.
- UNESCO (2007) closed a nationwide WB in Costa Rica, covering a 33 year span (1970-
2002) and using its simplest formulation (rainfall equals the sum of streamflow, actual
evapotranspiration and a measurement-error term), and based on 34 individual basin WBs.
Interpolation and expert-criteria based analyses allowed to produce spatial fields of rainfall,
streamflow and actual evapotranspiration, allowing the estimation of a discrepancy term
varying from -12% and +17% over the 34 basins, with a mean value of 2.6% at the national
level. For this, the soil water content was assumed invariant throughout the overall balance
period.
47
- Fujieda et al. (1997) studied two subtropical forest basins in Serra do Mar, Brazil (56 ha
and 37 ha); measuring rainfall, streamflow, crown interception, surface runoff and soil
water storage. They encountered a significant interception component and a very efficient
aquifer recharge, which provides sustained baseflow throughout the year in these relatively
small basins. The evapotranspiration was estimated as the residual of the WB equation.
- Kinner and Stallard (2004) used the hydrological model TOPMODEL to assess the water
balance in a tropical rainforest basin in Barro Colorado Island, Panama; from rainfall,
streamflow and potential evaporation (pan and atmometer) data. Some of the main concerns
from the experimental results of this study are linked to the uncertainty on the relative
balance between inter-basin groundwater leakage and evapotranspiration, given the low
control of this method on the true dynamics of actual evapotranspiration.
- Genereux et al. (2005) found that the inter-basin groundwater flow ranged from nearly
zero up to two thirds of the basin water input, measuring the WB components in two
adjacent rainforest basins in Sarapiquí, Costa Rica. In this analysis they measured rainfall
and streamflow, and estimated the annual actual evapotranspiration as a function of solar
radiation, basin on a previously established relationship based on eddy covariance
measurements. They also made some rough estimations of the annual variation in soil water
storage, which lead them to consider negligible this component.
C. In coffee plantations
The water yield in small coffee basins was measured in all the following reports:
- Manik (2008) studied two coffee basins of 4 and 27 ha under an AF system, with
respective slopes of 29% and 26%, as well as four coffee basins under MC with areas
varying from 3 to 20 ha and mean slopes between 20% and 46%, all located in West
Lampung, Indonesia. The runoff coefficients of individual storm events are given, showing
a much higher range for MC practices, but maximum values that can duplicate those of AF
plots.
- Baumann et al. (2008) worked in two coffee MC basins (2.8 and 1.7 ha) located in
Chiapas, Mexico; one with and the other without individual terraces, with respective mean
slopes of 46% and 33%. Here the runoff coefficients of individual events are more variable
for no terraced plots, presenting slightly higher maximum values than those of terraced
48
plots. Under a given rainfall accumulation threshold, the clay fraction of the soil seemed to
have greater impact on runoff generation than the slope gradient. Above that threshold,
however, the slope becomes the controlling factor.
- Hakkim et al. (2004) measured rainfall and streamflow in a 0.75 km2 basin of Kerala,
India; with mean slope of 24% and under coffee MC, which allowed to compare with
similar measurements in other three monoculture basins planted with cashew and tea, and
another basin under a dense forest.
3.1.3 Water balance at both, plot and basin scales
Some fewer reports tried to compare WB techniques at both the plot and the basin scales,
aiming at verifying the reliability of the upscaling (or downscaling) exercise. Some examples
of these investigations are given below (no studies were found in coffee systems).
- Wilson et al. (2001) aimed to study the evapotranspiration processes within a deciduous
forest in Tennessee, USA; by comparing soil-plot (interception model, soil eddy covariance
and sapflow), multiplot (eddy covariance above tree canopy) and basin (rainfall minus
streamflow) evapotranspiration data, across different time scales. One of their main
conclusions was that eddy covariance and basin WB methods provide acceptably reliable
estimations of annual evapotranspiration, without demanding elaborate scaling procedures
to represent spatial areas that are orders of magnitude greater than sap flow or soil water
budget methods. Another interest finding was the soil water budget seems limited for the
estimation of annual evapotranspiration, either because of excessive missing data during
periods with rainfall inputs and rapid water movement within the soil profile or because of
possible deep tree uptake, beneath the depth of measurement probes reach (specially during
drought times).
- Moussa et al. (2002) established a simple WB at the storm event scale (hydrograph
partition) by measuring rainfall, streamflow, water table levels and surface runoff in two
plots (1200 m2 and 3240 m2), in a 0.91 km2 experimental basin in Southern France. A
distributed hydrological model allowed assessing the role of tillage practices and of a
existing ditch network on peakflow properties, but also testing some hypotheses about the
runoff generation processes, the interactions between groundwater and ditch flows and the
water routing to basin outlet. A consistent link was established between the plot and the
49
basin scales, modelling the hydrological behaviour at both scales according to the observed
data.
- Maeda et al. (2006) made several measurements of WB components and fluxes in an
intensively observed hillslope plot (0.33 ha) and the small experimental basin (1.5 ha) that
contains this plot, under a suburban model (secondary) forest in the Aichi Prefecture, Japan.
At this site they measured gross precipitation, throughfall, stemflow, soil water content, and
surface and subsurface flow, as well as streamflow at the two scales (plot and basin). They
also measured with special emphasis the soil physical characteristics and structure in the
plot. Based on these observations, a spatially distributed model was built to test different
schematizations of soil thickness and its physical properties. All the parameters were
measured or extracted from literature, and the measured fluxes (streamflow, subsurface
flow) and water storage in soil were used to test model accuracy. The aim was to clarify
what factors affect long term spatiotemporal fluctuations in soil moisture and streamflow,
and conclude on some sort of environmental services: headwater conservation and recharge
function, and a gentle drainage function moderating flood discharge and drought.
3.1.4 Water balance at regional scale
The WB has been studied under the plot approach (which excludes surface, subsurface and
groundwater lateral flows, and aquifer recharge processes) but at much larger scales. No
specific applications were found in coffee regions, so examples in other land uses are
presented instead:
- Stephenson (1990) explored the relationship between the physiognomy of vegetation and
the WB at the sub-continental scale in North America, basing on calculated as rainfall,
Penman potential evapotranspiration and soil water content to obtain the actual
evapotranspiration and the water deficit.
- Neilson (1995) developed the Mapped Atmosphere-Plant-Soil System (MAPSS)
biogeography model and applied it in the United States. This model predicts potential
vegetation, plant functional types and leaf area index (according to an equilibrium
dependent on soil and climate characteristics), and is based on mechanistic calculations of
site WB, incorporating competition between grass and woody life-forms for light and water.
The interconnections between vegetation and hydrologic processes provide a set of
50
constraints on the model structure. The model seems robust in the ecophysiological
simulation at yearly time-step of the potential vegetation type and leaf area that could be
supported at a site, through the incorporation of thermal, humidity and productivity
constraints on vegetation life-forms, at regional, continental or global scales. However, it is
not designed for monthly or daily time-steps and assumes steady water content in soil and
aquifers.
- Imbach et al. (2010) applied MAPPS at the scale of Mesoamerica, allowing to simulate
leaf area index, potential vegetation and the WB balance partitioning at 1 km2 resolution
and for an annual time step.
- Arnold et al. (1999) applied the SWAT model to close the daily WB in the United States,
basing on what they called four control volumes: snow pack, soil profile, shallow and deep
aquifers. All the parameter values were fixed from available databases, so that no
calibration was made. The model accounts for the timing of lateral flows from runoff
(occurring from hours to days), non-saturated runoff (days to weeks), shallow aquifer
(months to years) and deep aquifer (years to decades: not simulated in this study). It also
includes specific modules to take into account different land uses/management and climate
scenarios, to estimate pollutant loads and to simulate plant growth. Even though the model
was claimed to be run at the daily time step, the validation through independent streamflow
measurements was carried out at the annual scale. In addition, this was the only variable
used to validate such a complex model, however the authors gave examples of the
successful validation of SWAT for other fluxes (like surface runoff, groundwater flow, and
evapotranspiration) at other sites in USA while producing monthly and daily time series of
streamflow.
Some authors tested the sensitivity of WB components to different spatial variants in
distributed models. For instance:
- Bormann (2006) assessed the effect of aggregating the spatial input datasets for the
TOPLATS grid-based hydrological model, on the estimation of regional WB components,
in a forest dominated basin of central Germany (55% of land cover is deciduous or
coniferous forest, but grassland, agricultural crops, fallow land and urban area are also
present). Although the absolute scale of the studied basin (area of 693 km2) is probably
conditioning the absolute results of this study, the author concludes that no significant errors
51
are detected on the WB components if the grid size is kept below 300 m, and that the errors
are gradually increasing for grids up to 1 km, from which significant errors arise. Another
important remark is that, in presence of significant vertical fluxes in the basin (e.g.,
evapotranspiration in basins with dense vegetation cover) from all model inputs, the land
use controlling SVAT processes is the most sensitive to spatial aggregation procedures. Soil
physical properties and topography are also sensitive factors though.
- Tripathi et al. (2006) studied the effect that different basin subdivision schemes had on the
modelled water balance using the SWAT semi-distributed model in a 90 km2 agricultural-
dominated basin in eastern India. They used measured data of rainfall, climatology and
streamflow, and found that the overall streamflow simulation was very accurate and little
sensitive to changing subdivision pattern in the basin, while this election can significantly
affect evapotranspiration, percolation and soil water content.
3.2 Plot erosion and basin sediment yield: measurements and upscaling
A summary of experimental studies is given below under three main themes: (1) erosion at
plot scale, (2) sediment yield at basin scale, (3) upscaling from plot erosion to basin sediment
yield.
3.2.1 Erosion at plot scale
Few studies were found on the installation of experimental setups to record erosion rates in
coffee systems. For instance:
- Ataroff and Monasterio (1997) used 6×2 m2 area and 31% sloped plots, installed under AF
and MC coffee systems in Mérida, Venezuela; to conclude that human activities can be a
much important factor than runoff and rainfall erosivity in the production of plot erosion.
They also showed the increase of erosion rates in the period just after the elimination of
shade trees in coffee plantations.
- Afandi et al. (2002a) isolated coffee plots of 20×5 m2 size and 27% slope in Lampung,
Indonesia; in order to record seasonal runoff coefficients in the presence of Paspalum
conjugatum weed. Higher runoff coefficients were detected under natural weed cover and
clean-weeded ground. Erosion rates were one order of magnitude higher in the coffee
52
plantation with clean-weeded cover, with respect to coffee with Paspalum conjugatum
cover and to coffee with natural weeds cover.
- Dariah et al. (2004) made observations in 15×8 m2 size plots on steep slopes (50% - 60%)
in Lampung, Indonesia; recording very similar runoff coefficients in a coffee AF system
with Gliricidia sepium, and in a coffee MC system. They also found very similar erosion
values for AF and MC, being the latter slightly greater.
- Gómez-Gómez (2005) worked on 7×2 m2 coffee plots with slopes between 25% and 30%,
located in Alajuela, Costa Rica. The author compared the measurements of water yield and
plot erosion from plots treated with the paraquat herbicide, with those from plots treated
with the glyphosate herbicide, the former producing two thirds of the erosion observed in
the latter.
- Rijsdijk et al. (1990) worked on Wischmeier plots installed in East Java, Indonesia; and
measured erosion rates in coffee plantations, which ranked in-between of those in
forest/scrubland (the lowest rates) and in maize fields (the highest rates).
- Hairiah et al. (2005) used 10×4 m2 size plots installed in West Lampung, Indonesia; to
verify that a coffe AF system (Gliricidia sepium and Paraserianthes falcataria) yielded an
average erosion rate that is a fifth part of that under MC, for coffee systems over 10 years.
3.2.2 Sediment yield at basin scale
A. Measurements of sediment yield in coffee basins
Only one study was found providing estimations of sediment yield in a coffee basin. This is
the work of Afandi (2005), which presents measurements of water and sediment yield in a
small (2 ha) high sloped (55%) basin on volcanic soils in Sumberjaya, Indonesia; and mostly
covered by coffee under AF and MC schemes, but also by other land uses in minor degree. He
measured annual runoff coefficients and annual sediment rates under undisturbed conditions,
and using manually collected sediment samples. One main finding of this work is that
sediment yield increased during farm footpath improvement, involving cutting of the original
slope and accumulation of the loose soil clods downslope of the road, which accelerated the
transport processes to the stream.
53
B. Sediment yield at basin scale using optical backscatter sensors
Basin sediment has been measured in the most diverse environments, through very different
methodologies, depending on the time-space scales, the required precision and the
technological development of new measurement techniques, as described by Downing (2006).
However, it has been shown that continuous turbidity recording by the use of optical
backscatter sensors (OBS) has performed much better to estimate suspended sediment
concentrations under many conditions.
Measurements during storm events
Several authors used an OBS to measure sediment concentrations only during storm events.
For instance:
- Walling and Webb (1982) proposed that the shape of the suspended sediment
concentration/discharge relationship may prove useful in the development of realistic
models of storm-period sediment yield, including “exhaustion effects” in the form of this
relationship. They analized data from several basins in Devon, UK, and highlighted some
relevant implications for the development of sediment yield models: (1) the importance of
the surface runoff component of total streamflow in sediment generation, which make
necessary to incorporate a realistic representation of storm runoff production; (2) the
reproduction of the typical hysteretic behaviour of sediment concentration during individual
events should be attempted, by considering potential contrasts in detachment and transport
capacities during the rising and falling limbs of the hydrograph; (3) a time-variant measure
of sediment availability should be incorporated to take account of exhaustion effects
operating both, within multiple events and during a sequence of events, and involving
parameterization of the “recovery process” (time of sediment replenishment); and (4) the
recognition that a realistic representation of sediment availability, exhaustion and recovery
will necessitate consideration of the partial/variable source area concept of storm runoff
production.
- Gippel (1989) points out the confounding effects that are often present in natural stream
systems, for which device calibration requires careful consideration of site characteristics.
His analyses, based on data from five sites in a small forested basin of Australia, indicated
variability in the turbidity - suspended sediment concentration relationship. Improvements
54
were obtained by considering the effects of very fine sediment and background water
colour. The report concludes that a good linear relationship between turbidity and
suspended sediment concentration should not necessarily be expected and any observed
hysteresis could actually help explain erosion and transport processes.
- Lenzi and Marchi (2000) studied suspended sediment load in a small, high-gradient stream
located in the Dolomites, Italy. They analyzed the ratio of suspended to total sediment yield
and the relations between sediment concentration and water discharge, for summer and
autumn seasonal floods. These authors linked different patterns of hysteresis in the relation
suspended sediment - discharge, to the types and locations of active sediment sources. A
coarsening of transported material for increasing streamflow was detected looking at within-
storm variation of the particle size of suspended sediment during a major flood, while the
analysis of grain sizes helped to identify the main sources of suspended load (erosion areas
on hillslopes). The data made clear the influence of particle size variation on the turbidity-
concentration relationship.
- Sun et al. (2001) developed a storm-based sediment estimation based on turbidity and
sediment sampling data by collecting several datasets in a 2.9 km2 basin of Adelaide,
Australia. The datasets were used to develop a number of turbidity - sediment relationships
with which basin yield was estimated and its variability analyzed. An annual load was
derived by accumulating the storm sediment loads, finding that large storms dominated
sediment yield in the basin, while its rate depends more on the peak storm flow. The authors
affirm that erosion based on infrequent, non-storm based or extrapolated data is exposed to
potentially large errors, and the results may only be relied upon as a general guide rather
than serious estimation of basin sediment yield. The estimation error was found to partly
depend on the forms of functional relationships, working better the polynomial ones: of 2nd
order for general purposes and of 4th order for data on the larger range.
- Seeger et al. (2004) worked in a 284 ha basin in the Central Spanish Pyrenees, and tried to
link basin soil moisture and rainfall characteristics with the hysteretic loops in the
streamflow - suspended sediment relationship, basin on turbidimeter readings during storm
events. Three types of hysteretic loops were found at the studied experimental basin and
associated with different levels of humidity and rainfall, here used as indicators of different
processes of runoff and sediment transport (1) clockwise loops (the most frequent) are
55
produced under “normal” storm flow conditions, when the basin is very moist and runoff
generation and sediment supply is limited to areas next to the channel (i.e. sediments are
removed, transported and depleted rapidly), (2) counter-clockwise curves occur under very
high moisture and high antecedent rainfall conditions. In this case, flood propagation occurs
as a kinematic wave. Sediment sources are incorporated all over the basin. In both cases,
saturation excess overland flow generates the surface runoff. (3) The eight-shaped loop
(partial clockwise followed by counter-clockwise) occurs with low water content. Here, the
runoff generation process is supposed to be infiltration excess overland flow, which causes
a rapid extension of the contributing areas both near the channel and over the whole basin.
- Lana-Renault et al. (2007) worked on the same basin that Seeger et al. (2004), but made a
different classification of storm flows according to a cluster analysis, finding four types of
storm events according to precipitation, streamflow, suspended sediment and antecedent
moisture conditions. Two clusters were associated to dry conditions, facing rainfall of
moderate volume and low intensity and presenting very fast response of streamflow and of
suspended sediment concentration. Other cluster was defined under dry conditions but for
intense rainfall events, featuring a fast response on streamflow and sediment concentration,
and showing an enlargement of the contributing areas. Finally, under very wet conditions
relatively moderate rainfall produces a great response in streamflow but keeping a mild
reaction in suspended sediment concentration, which the authors interpreted as a dilution
effect occurring when the entire basin is contributing.
- Minella et al. (2008) assessed the relationship between suspended sediment concentration
and turbidity for a small (1.19 km2) rural basin of Arvorezinha, Brazil; and evaluated two
calibration methods by comparing the estimates of suspended sediment concentration
obtained from the calibrated turbidity readings with direct measurements obtained using a
suspended sediment sampler. An interesting remark from this experience is the need to
consider the complex nature of the relationship between sediment concentration and
turbidity, reflecting the influence of factors such as particle size and shape, and of water
colour. It seems important that the calibration method adequately represents these naturally
occurring variations, which are difficult to replicate in a laboratory setting.
56
Measurements at the annual scale
The sediment yield for longer time spans has also been calculated using turbidimeters. A
selection of studies developed at that time scale is given below:
- Walling (1977) calculated annual sediment loads using a turbidimeter in a 258 km2 basin
in Devon, England; in order to test the accuracy of suspended sediment rating curves, which
were considered to introduce from large to huge errors. The author highlighted the fact that
the calibration of this device is made for a particular site in a small- or medium-sized basin
with relatively homogeneous rock type and a predominantly silt- and clay-sized load.
- Jansson (1992) used two types of turbidimeters to estimate the sediment transport in a 371
km2 basin of Costa Rica, contrasting the variations during day/night and during a long
period of time. Three main recommendations are given for the design of representative
water sampling programs through a manual technique, aiming to obtain a reliable sediment
rating curve: (1) many streamflow peaks must be included because of the great
concentration variations between different water peaks, (2) water sampling must be
undertaken during evenings and nights in order to include high concentrations of short
duration, and (3) the crucial streamflow range to be sampled should be estimated by using
water duration curves and a preliminary sediment rating curve.
- Wass and Leeks (1999) reported the results of the measurements from an extensive
turbidity-based sediment monitoring network, involving ten of the main tributaries of a
large basin in the Humber District, UK; and aiming to measure the total flux of suspended
sediment to the estuary. Linear relationships were established between suspended sediment
concentration and turbidity. Some potential uncertainties were identified and quantified.
Large temporal and spatial variations in the flux were measured during the monitoring
period, in response to factors such as geology, climate, land use, basin scale, deposition and
reservoir trapping. The systematic variations owing to sensor positioning compared with
measurements made using depth-integrated samplers across a section were relatively low.
Some properties were advised as desirable in the development of new instruments that
provide consistent turbidity records, free from the effects of temperature, sunlight, algal
accumulation and large river debris.
57
- Jansson and Erlingsson (2000) established the sediment budget (inflow, deposition,
flushing removal and river outflow) in the Cachí hydropower reservoir in Costa Rica; by
combining turbidity records, the traditional sediment rating curves, side-scan sonar, echo-
soundings and excavating pits in the reservoir, techniques that are applicable at different
time scales (from annual to event). The balance of the sediment budget seemed to indicate
that the major components of the sediment budget were estimated in the right order of
magnitude. Because of the wide scatter of data, the authors proposed to develop the
sediment rating curves on mean sediment loads in discharge classes, to avoid the bias of log
regressions. The uncertainty analysis is proposed as a tool to decrease the costs of gathering
adequate data to achieve the required precision (sampling strategy), so in the study different
types of error were identified, and then calculated or estimated.
- Schoellhamer and Wright (2003) worked in rivers of California and Utah, USA; and
calibrated an optical backscatter sensor to streamflow-weighted, cross-sectionally averaged
suspended sediment concentration, which was measured with the equal streamflow- or
width-increment, methods and an isokinetic sampler. The calibration of the sensor was
found to be affected by particle-size variability. The authors consider that the results of
optical sensors directly deployed in rivers should be evaluated on a site-specific basis and
measurement objectives, while three main potential problems were identified: biological
fouling, particle-size variability, and particle-reflectivity variability.
- García-Ruiz et al. (2008) reported the monitoring of three small basins in the Central
Spanish Pyrenees, to analyse the hydrological and geomorphological consequences of
different land covers under the same climate scenario: (1) a dense, undisturbed forest
environment; (2) an old, abandoned cultivated area subjected to colonisation by plants; and
(3) some active badlands. They point to plant cover is a key factor, influencing the
seasonality and intensity of floods, the annual volume of discharge, and the suspended
sediment concentration, total sediment yield and proportions of different types of sediment.
Sediment outputs from the badland basin are two orders of magnitude higher than those
from the other two basins.
58
3.2.3 Upscaling, from plot erosion to basin sediment yield
In tropical regions, the erosion monitoring programs assessing the effectiveness of the
implemented soil/water conservation practices are not well documented, and are usually
focused on the plot scale (Verbist et al., 2010), thus carrying to scrutiny for not delivering the
desired reductions in downstream sediment rates (Rijsdijk et al., 2007a). The scaling up of
plot results to the basin scale is not straightforward, given that basin sediment yield is the
product of all generating and transporting processes within a basin (de Vente and Poesen,
2005). Scaling sediment fluxes has been recognized as a difficult task, since factors as the
dominant geomorphic processes controlling sediment yield, the dependency on the spatial
distribution of sediment sources and sinks, and the influence of point sediment sources such
as runoff-generating unpaved roads (Vanacker et al., 2007) can produce a wide variability in
the overall yield response.
The upscaling issues between plot erosion and basin-sediment yield have been reported by the
next authors:
- Afandi et al. (2002b) stresses that soil erosion from various land use types can be easily
measured using erosion plot studies, although in a landscape scale, the existence of “border
strips” makes difficult the extrapolation of the results from erosion plot studies. The
sediment transfer was measured across various types of land uses on coffee culture based
and natural field borders in Lampung, Indonesia; using a “point measurement” technique.
The aim was to determine the extent of sediment transfer from various land use types and
natural vegetative filters, and it was found that the existence of natural vegetative strips was
very effective in entrapping sediment. Due to this filtering effect, small sediment yields
were observed in land uses with relatively high plot erosion rates.
- Sidle et al. (2006) presented a deep study case review, underlining the effects different
types of land uses and management schemes on surface and landslide erosion in Southeast
Asia. The authors consider that frequent misconceptions exist regarding the
conceptualization and relative importance of these two processes, and the actual
implications of management activities and avoidance/control measures. The discrepancies
found were attributed to the short-term and small-scale nature of many studies as well as the
emphasis on agronomic production at the farm or field-plot scale; although even in longer-
59
term basin studies, the occurrence of landslides and their influence on sediment budgets was
considered under-studied, ignoring the fact that this process may contribute with
comparable levels of sediment in steep terrain. A lack of study was also detected dealing
with the effects of scale on sediment delivery to streams. According to the authors, plot
scale erosion is generally higher than basin scale erosion because average infiltration
capacity increases with scale and deposition occurs within the basin. However, in some
cases, larger scale measurements will include other processes (e.g., gully development, bank
erosion, dry ravel) that cannot be captured at the plot scale, and thus may yield higher soil
losses. Besides of landslides, the role of roads, trails and footpaths was also discussed as
part of the sediment scaling issue, as well as the potential sediment transport mitigation
through terrain interspersion by trees or brush lands, which provide areas of high infiltration
along with ‘roughness elements’ on the landscape, where sediment deposition can occur.
Some studies intended to match the sediment yield at both, the plot and basin scales, tracing
the main sources and the transport mechanisms turning the erosion into river sediment yield.
- Brasington and Richards (2000) established an intensive three-year project, carrying out
the instrumentation, calibration and analysis of high frequency infrared turbidimetric
records from five small sub-basins in Kathmandu, Nepal; in order to obtain basin sediment
yields. The comparison with area-weighted plot-scale measurements from other studies in
the same basins revealed that all estimates of basin sediment yield are significantly higher
with respect to plot erosion, suggesting the importance of other sediment sources like active
gully systems, slope failure, riparian and channel erosion, as well as a highly efficient
sediment delivery ratio.
- Rijsdijk et al. (2007a) analyzed measured data on runoff and sediment yield from a gully
system, riparian mass wastage and bank erosion, in order to highlight the contribution of
these sources to the total (measured) sediment yield in a basin of East Java, Indonesia. The
non-sheet erosion transported to the basin outlet reached a maximum of 27% of the total
annual load, which in addition to similar estimations of the effect of impervious surfaces
like roads, trails and settlements, made evident the inaccuracy of the direct upscaling of plot
measures to basin scales.
- Raclot et al. (2009) found opposite results in a Mediterranean, 0.91 km2 experimental
basin of Southern France, in which sediment yield over 18 rainfall-runoff erosive events
60
was significantly lower than the measured plot erosion. According to the authors, this result
confirms that basin soil loss cannot be estimated by the sum of individual field losses, and
that the scaling transition can be analyzed in terms of the spatial variability of soil
management and the connectivity between sediment-producing and deposition areas. This
latter element was fundamental in the reduction of sediment loads at basin scale, and was
linked to filtering land discontinuities in the shape of ditches (located in a depression area)
and field boundaries.
- Verbist et al. (2010), tried to relate plot erosion and basin sediment yield with a variety of
possible clarifying factors like land use, geology, soil and topography. Working in three
sub-basins of West Lampung, Indonesia; sediment yield at basin scale per unit area was
found to be much higher than soil loss measured in erosion plots. A regression analysis
showed that the dominant factors explaining sediment yield differences at basin scale were a
particular lithology and slope angle, followed by the silt fraction of the top soil. Soil loss at
the plot scale was measured under bare soils, monoculture and shaded coffee, and forest
plots. The analysis through spatial scales revealed that less than 20% of the basin area
produced almost 60% of the sediment yield. Both, a detailed assessment of the lithologies in
volcanic landscapes and the consideration of potential sediment sources and sink areas were
identified as key factors, for plot to basin sediment quantification. Sheet or rill erosion at the
plot scale were discarded to be the main contributors to sediment yield at the basin scale,
while the dominant erosive processes could be rather river bank/bed erosion, landslides
and/or the concentrated flow erosion on footpaths and roads. Filtering effects were
associated to a major flood plain in the basin. A final remark of this study is that the
currently promoted forest protection and ‘re-treeing’ by converting sun coffee to
agroforestry systems with shade trees will reduce over time soil loss and surface runoff at
plot scale in lithologically sensitive areas. However, the impact at basin scale is likely less,
because these measures will affect only partly the dominant sediment generating processes.
61
3.3 Concluding remarks: proposing measurement and modelling methods
for combined ecophysiological, hydrological and sedimentological assessments in coffee agroforestry basins
3.3.1 Water balance: ecophysiological and hydrological approaches
According to the reviewed literature, there exist many different ways of closing the WB, and
the chosen methodology is often restricted by the available measurements and the targeted
WB component. There seem to be two main different approaches for closing a water balance,
the engineering (hydrological) approach aiming to understand surface and return flow at basin
scales, and the agronomical (ecophysiological) approach, more interested in soil WB at plot
scale and on the effects of the local microclimatology on plant physiological responses. Some
combinations of these approaches, emphasizing one or the other, have been examined above.
For instance, when WBs at larger scales are produced basing on ecophysiological models, it is
often observed that few or no lateral flows are included (especially for groundwater flows),
and that the basin structure and accumulation processes are roughly conceptualized. On the
other hand, it is frequent that engineer hydrologists do not go further analyzing time-space
variations in the vegetation cover and its physiological behaviour, even though this
component can be an important (or the main) control of the WB through evapotranspiration
processes. An interdisciplinary multi-scale work targeting clear quantitative objectives is
desirable, in order to respect the importance of the different components of the biophysical
system, and the hidden interaction between components, which is commonly neglected when
a single-field approach is carried out.
Modelling has been commonly used as a complement of WB measurements for many
different purposes, from the gap filling of incomplete datasets or the reproduction of the
hydrological behaviour of unknown WB components, to the assessment of the dynamics of
other linked biophysical systems. The complexity of the chosen model is a function of the
modelling purposes and the available observations. However, some approaches making gross
assumptions (e.g. actual evapotranspiration equalling the potential one, no subsurface lateral
flow, no seasonal accumulation properties in streamflow, no deep-percolation in small basins)
might lead to either unsatisfactory model performances or to acceptable univariate
62
performances (on streamflow, for instance) that are however based on unrealistic or non-
validated representations of reality.
3.3.2 The problem of upscaling plot erosion to basin sediment yield
Concerning the experimental measurement of surface runoff and erosion, many antecedents of
research were found in coffee plots, including the effect of shade trees, weeding and
herbicides. In particular, the agroforestry schemes seemed to promote a reduction in both
runoff and erosion processes. However, a high variability was found in the experimental
methods, and most importantly, in the results, making clear that these processes are highly
site-dependent, and that universal estimations must be treated carefully.
Only one experimental study measuring water and sediment at basin scale was found to be
developed on coffee systems, which make obvious the lack of knowledge and understanding
of this production scheme. The most accurate and current technique to quantify sediment
fluxes, according to the several researches presented above, has proven to be the continuous
measurement of turbidity, which has never been used to study a coffee basin. This device
performs satisfactorily during both storm events and recession times, giving the possibility of
following the basin response to different erosion-generating and sediment-transporting
mechanisms. The importance of a careful in situ calibration was discussed, while different
sampling methods and experimental considerations were also shown. A generalized
conceptual deficiency was found, however, in the treatment of all measurement, sampling and
analytical errors, and in the estimation of the uncertainties on sediment concentration/yield.
Finally, it was widely demonstrated in this literature review that it is conceptually incorrect
the simple upscaling of plot erosion measurements or estimations to provide basin scale
estimates by a direct integration of the individual outputs. Consequently, the impact of
agricultural management practices intending to reduce soil erosion at the plot scale, will not
have a proportional effect in the reduction of sediment yield at basin scale.
3.3.3 Proposing a methodology for the assessment of HES based on
ecophysiological, hydrological and sedimentologic approaches
A combination of experimental and modelling approaches, taken into account the particular
biophysical conditions of the studied basin, the human influence on the basin properties
63
(agriculture, infrastructure, management practices, etc.) and its time-space variations, seems
to be, by far, the most complete feasible methodology that can be applied in order to assess
the actual condition and continuous evolution of a given Hydrological Environmental Service.
In order to assess the effectiveness of management practices carrying to propose the payment
of HES it is necessary to identify the target variable. For the sake of an example, in the
present study the most interesting HES were found to be linked to water quantity (aquifer
recharge and return flow properties) as well as to water quality (sediment concentration in the
streamflow and possible effect of agroforestry on erosion reduction), both HES operating at
basin scale. The first objective (studying water quantity, aquifer recharge and flow seasonal
regulation) can be achieved through the systematic measurement, at the basin scale, of as
many WB components as possible. The closure and understanding of the water balance of the
studied ecosystem is indispensible at this stage. Then, a realistic basin-scale model including
ecophysiological and hydrological processes, calibrated and validated in the particular
agricultural context, could provide complementary information on non-measured components.
The accuracy of the modelled WB components and processes can be improved using all
available ecophysiological measurements as input information to constraint the model. The
second objective (studying water quality) should also be addressed using field measurements.
For the evaluation of basin sediment yield, in particular, the most accurate and current
technique to quantify sediment fluxes, according to the several researches presented above,
has proven to be the continuous turbidity meters. However, a proper in situ calibration of
these devices is decisive, and particle properties and sampling procedures become central for
the achievement of valid measurements. Given the lack of bibliographic information
concerning the uncertainty in sediment-related estimations, an effort should be done in order
to define standard methods to assess the reliability of basin sediment estimations.
64
65
CHAPTER 4
Study site, experimental setup and data
4.1 Study site: location, climate and soil
The area of interest is located in Reventazón river basin, in the Central-Caribbean region of
Costa Rica (Figs. 18 and 19a,b). It lies on the slope of the Turrialba volcano (central volcanic
mountain range of the country) and drains to the Caribbean Sea.
Figure 18 Turrialba river basin, Mejías creek micro-basin and coffee
plantations. (Source of coffee plantations map: Icafé).
The Aquiares coffee farm5 is one of the largest in Costa Rica (6.6 km
2), “Rainforest
AllianceTM
”6 certified, 15 km from CATIE
7 (Centro Agronómico Tropical de Investigación y
Enseñanza). Within the Aquiares farm, we selected the Mejías creek micro-basin (Fig. 19c)
for installing the “Coffee-Flux”8 experiment from scratch in late 2008. Coffee-Flux became
5 http://cafeaquiares.com 6 http://www.rainforest-alliance.org 7 http://www.catie.ac.cr 8 http://www.montpellier.inra.fr/ecosols/recherche/projets_de_recherche_finances/coffeeflux
66
part of the TGIR/SOERE9 in September 2010, which makes it a long-term observatory,
together with the other instrumented sites of Cirad/UPR80/UMR Eco&SOLS. Since 2009,
CoffeeFlux became also part of the Trees Carbon & Water10
project from SLU-Sweden.
Figure 19. a) Location of Reventazón river basin in Costa Rica, Central America. b) Position
of experimental basin inside of Reventazón basin. c) The “Coffee-Flux” experimental basin
in Aquiares farm and its experimental setup to measure the water balance components.
The basin is placed between the coordinates -83º44’39” and -83º43’35” (West longitude), and
between 9º56’8” and 9º56’35” (North latitude) and is homogeneously planted with coffee
(Coffea arabica L., var Caturra) on bare soil, shaded by free-growing tall Erythrina
poeppigiana trees. The initial planting density for coffee was 6,300 plants ha-1
, with a current
age >30 years, 20% canopy openness and 2.5 m canopy height. It is intensively managed and
selectively pruned (20% per year, around March). Shade trees have a density of 12.8 trees ha-1
(Taugourdeau, 2010), with 12.3% canopy cover and ca. 20 m canopy height. The
experimental basin has an area of 0.9 km2, an elevation range from 1,020 up to 1,280 m.a.s.l.
9 http://www.insu.cnrs.fr/co/files/ao.pdf 10 http://sites.google.com/site/treescarbonwater/home
67
and a mean slope of 20% (45° being 100%). Permanent streams extend along 5.6 km,
implying a drainage density of 6.2 km km-2
. The average slope of the main stream is 11%.
According to the classification by Mora-Chinchilla (2000), the experimental basin is located
along a 1.3 km wide strip of volcanic avalanche deposits, characterized by chaotic deposits of
blocks immersed in a matrix of medium-to-coarse sand, which is the product of the collapse
of the south-eastern slope of Turrialba volcano’s ancient crater. The general classification
given by the geological map of Costa Rica (MINAE-RECOPE, 1991) describes the general
stratigraphy as shallow intrusive volcanic rocks, and the particular region as proximal facies
of modern volcanic rocks (Quaternary), with presence of lava flows, agglomerates, lahars and
ashes. Soils belong to the order of andisols according to the USDA soil taxonomy, which are
soils developing from volcanic ejecta, under weathering and mineral transformation
processes, very stable, with high organic matter content and biological activity and very large
infiltration capacities.
The climate is tropical humid with no dry season and strongly influenced by the climatic
conditions in the Caribbean hillside, according to Köppen-Geiger classification (Peel et al.,
2007). The mean annual rainfall in the study region for the period 1973-2009 was estimated
as 3014 mm at the Aquiares farm station (Fig. 20). At the experimental basin the rainfall in
2009 (3208 mm) was close to the annual mean, but showed a monthly average deviation of
±100 mm around the historical regime. Mean monthly net radiation ranged in 2009 from 5.7
to 13.0 MJ m-2
d-1
, air temperature from 17.0 to 20.8 °C, relative humidity from 83 to 91 %,
windspeed at 2 m high from 0.4 to 1.6 m s-1
and potential evapotranspiration (Allen et al.,
1998) from 1.7 to 3.8 mm d-1
.
0
100
200
300
400
500
600
J F M A M J J A S O N DMonth
Monthly
rainfall
(mm) Aquiares farm station, Average 1973!2009
Aquiares experimental basin 2009
Figure 20. Mean monthly rainfall at Aquiares farm station (period
1973-2009) compared to the rainfall in the experimental basin in 2009.
68
4.2 Experimental setup
The “Coffee-Flux”11
experimental basin and instrument layout was designed to trace the main
water balance components employing spatially representative methods (Fig. 19c). It is part of
the FLUXNET12
network for the monitoring of greenhouse gases of terrestrial ecosystems.
The hydrological measurements used in this thesis were recorded from December 2008 up to
March 2010.
4.2.1 Rainfall (R) and climate
Rainfall was monitored at 2-3 m above ground (Fig. 21) in the middle of 3 transects of the
basin, using three lab-intercalibrated ARG100 tipping-bucket (R.M. Young, MI, USA)
connected to CR800 dataloggers (Campbell Scientific, Shepshed, UK), and integrated every
10 min. Other climate variables were logged on top of the eddy-flux tower with a CR1000,
every 30 s, integrated half-hourly and using: Net radiation: NR-Lite (Kipp & Zonen, Delft,
The Netherlands); PPFD: Sunshine sensor BF3 (Delta-T devices Ltd, U.K.); temperature and
humidity: HMP45C in URS1 shelter (Campbell Scientific); wind-speed and direction: 03001
Wind Sentry (R.M. Young, MI, USA). The theoretical evapotranspiration from a wet grass
placed under local climate conditions, ET0, was computed in accordance with FAO (1998).
Figure 21 Rainfall stations in the (a) middle and the (b) uppest parts of the experimental basin.
11 http://www.montpellier.inra.fr/ecosols/recherche/projets_de_recherche_finances/coffeeflux
12 http://www.fluxnet.ornl.gov/fluxnet/index.cfm
b)a)
69
4.2.2 Streamflow (Q)
A long-throated steel flume (length: 3.9 m; width: 2.8 m; height: 1.2 m) was home-built to
measure the streamflow at the outlet of the experimental basin, to record up to 3 m3 s
-1, the
maximum estimated discharge for the study period from an intensity-duration-frequency
analysis (Fig. 22).
Figure 22: (a) Long throat flume at the outlet of the experimental basin. (b) Calibration of the
flume by pygmy current meter (courtesy of ICE).
The flume was equipped with a PDCR-1830 pressure transducer (Campbell Scientific) to
record water head at gauge point (30 s, 10 minutes integration), while the rating curve was
calculated considering the geometric and hydraulic properties of the flume using Winflume
software (Wahl et al., 2000). A validation of the rating curve was made successfully using the
salt dilution method as well as a pygmy current meter.
4.2.3 Soil water content (!)
A frequency-domain-reflectometry portable probe (FDR Diviner2000, Sentek Pty Ltd) was
used to survey 20 access tubes distributed in the three study transects to provide the mean
volumetric soil water in the basin. The sensor measures at 10 cm intervals, reaching a total
depth of 1.6 m. A measurement campaign through the 20 sites was carried out every week.
The sensors were calibrated by digging sampling pits in the vicinity of six test tubes, to obtain
the actual volumetric soil water content from gravimetric content and dry bulk density.
b)a)
70
Figure 23: Measuring on a coffee slope with Diviner2000 FDR.
4.2.4 Evapotranspiration (ET)
The actual evapotranspiration from the soil, coffee plants and shade trees was measured at
reference height (26 m) on the eddy-covariance tower (Fig. 24), similarly to Roupsard et al.
(2006). 3D wind components and temperature were measured with a WindMaster sonic
anemometer (Gill Instruments, Lymington, UK) at 20 Hz. H2O fluctuations were measured
with a Li-7500 open path (LiCor, Lincoln, NE, USA). Raw data were collected and pre-
processed by “Tourbillon” software (INRA-EPHYSE, Bordeaux, France) for a time-
integration period of 300 s, then post-processed using EdiRe software (University of
Edinburgh, UK) into half-hourly values and quality checked. A validation was made by direct
comparison of the measured net radiation Rn with the sum of sensible heat flux (H) and latent
heat flux ( E): at daily time step, this yielded H+ E=0.92 Rn (R2 = 0.93) which was
considered sufficiently accurate to assume that advection effects on E could be neglected
here. Due to lighting and sensor breakdown, 45 days of data were lost between July and
August 2009. To gap-fill the missing period we used the Penman-Monteith model, whose
canopy conductance was adjusted using measured values.
71
Figure 24: Eddy covariance tower (26 m high), recording the actual evapotranspiration, energy and
CO2 fluxes of the coffee AF system. In the rear, one can see the Turrialba volcano
4.2.5 Leaf Area Index (LAI)
The coffee light transmittance was measured monthly in diffuse light conditions, for five rings
at different zenital angles (LAI2000, Li-COR Corvallis, USA, Fig. 25), along three 50 m-long
transects through the flux tower plot, similarly to Roupsard et al. (2008), but here using 2
devices, one for reference impinging light (in a clearing) and one mobile, measuring above
coffee (=below tree canopy) and below coffee. Effective coffee plant area index, PAI,
obtained from this light transmittance, was converted into actua leaf area index (LAI)
according to Nilson (1971), using a ratio of effective PAI to actual LAI that was estimated
from a dedicated calibration detailed below.
Figure 25: LAI-2000 device to measure light transmittance.
72
The LAI and upscaling work using remote sensing is detailed below and is a courtesy of S.
Taugourdeau (Taugourdeau, 2010). The actual coffee LAI was measured directly on a small
plot by counting total leaf number of 25 coffee plants, measuring leaf length and width every
20 leaves and using empirical relationships (courtesy of B. Rapidel) between leaf length and
width and leaf area (LI-3100C, Li-COR) (R2 > 0.95). On the same small plot, the effective
PAI was measured with LAI2000. The ratio of effective PAI to actual LAI was then calculated
on this small plot (1.34) and was considered to be constant with time and space in the micro-
basin, allowing the estimation of the actual LAI on the three LAI2000 transects. The LAI for
shade trees was estimated using their crown cover projection (on average 12.3% over the
whole basin) observed on a very high resolution panchromatic satellite image (WorldView
image, February 2008, 0.5 m resolution). As we did not have measurements of LAI for shade
trees, we considered this LAI in the order of magnitude of coffee LAI on a crown-projected
basis, and therefore we multiplied the actual coffee LAI measured on transects by 1.123, to
estimate the ecosystem LAI (tree and coffee). In order to monitor the time-course of
ecosystem LAI at the basin scale, we combined these ground measurements with time series
of remotely-sensed images. We used time series of Normalized Difference Vegetation Index
(NDVI) from Moderate Resolution Imaging Spectroradiometer (MODIS) data products
MOD13Q1 and MYD13Q1 (16-Day composite data, 250m resolution). NDVI is known to be
correlated with the green LAI if it is low (< 6), for most ecosystems (Rouse et al., 1974).
Twenty-three MODIS pixels covering the experimental basin were selected, and their NDVI
time series were downloaded. We filtered the raw NDVI time series according to quality
criterion given in the MODIS products, and we adjusted a smooth spline function on it as in
Marsden et al. (2010). Then, a linear regression between the smoothed NDVI of the pixel
including the flux tower and ground values of actual ecosystem LAI was calibrated (R2 =
0.69). This regression was used on other pixels of the basin, and averaged to have the annual
time-course of actual LAI.
4.2.6 Water table level (z)
Four piezometric wells measuring up to 4 m depth were built in the three main transects of
study (Fig. 26). They were equipped with pressure transducers (Mini-Divers, Schlumberger
Water Services) that measure and record the water table level every 30 minutes.
73
Figure 26: A piezometric well and a Mini-Diver sensor in the
experimental basin.
4.2.7 Turbidity (")
An optical backscatter sensor (OBS) or turbidimeter OBS-3 (Campbell Scientific) was
deployed in the flume at the outlet of the basin (Fig. 27), allowing to measure water ! every
20 s and integrate it every 10 and 30 min. NTU range: 0-250 (low) and 0-1000 (high range).
The sensor window was placed facing the main stream and protected with a grid properly
adapted to the source beam and detector acceptance cone.
Figure 27: Location of the backscatter sensor in the flume, with respect to the water inflow.
Flume
Turbidimeter
Streamflow
74
4.2.8 Plot surface runoff (SRplot) and plot erosion
Two experimental plots of 1000 m2 each (14 m by 72 m, approximately) were placed
following the slope direction within the experimental basin (Fig. 28). Both plots have a slope
of 20%, which was also the average slope of the basin. One of the plots displays an
agroforestry scheme (AF), coffee plants under E. poeppigiana shade trees, while the other
plot displays coffee in monoculture (MC), i.e. there is no tree in this subplot. They are limited
in the upper side by the basin water divider, while their lateral boundaries were isolated by 20
cm tall barriers. PVC gutter collectors were installed at the lower end of each plot to trap
coarse sediment and route runoff to two measuring tanks equipped with tipping buckets
(1000-20, Sierra Misco, CA, USA). The soil surface within 30 cm of the gutters was
compacted and sealed with a plastic sheet. The erosion at the plot scale was measured by
weekly collection, drying and weighting of the coarse sediment accumulated in the gutters at
the end of the experimental plots, along with the finer particles trapped by plastic filters
located just before the entrance of runoff into the measuring tipping buckets.
Figure 28: Experimental plots (a) in their whole extension: the shaded plot (AF) at left, the non-shaded
plot (MC) at right; (b) at the lower end where the gutters collect water and plot erosion (the plastic
canopy over the gutters has been retracted for the photo).
4.3 Data: measurement period, missing values and gap!filling
The information utilized to model and close the water balance partition (Chapter 5) is given in
Table 2 for the five measured hydrologic variables. The frequency of measurement varies, but
b)a)
75
is finally calculated at the 30 minutes time step (except for soil water content and sediment
collection at plots, which were performed on a weekly base). When gaps are present in the
measurements, a gap filling method was applied.
Concerning the estimation of water and sediment yield (Chapter 6), for the sake of having one
complete year of turbidity observations, our measurement period was chosen from 07/03/2009
to 06/03/2010. The frequency of measurement depends on the time scale of the analysis: for
an annual based analysis we worked at the 30 min time step (!t), while for storm events we
chose the 10 min !t. The procedure for the in situ recording of turbidity by OBSs presents
some relevant shortcomings, especially those linked to chemical and biological fouling in
some streams and brackish water environments (Downing, 2006). Those factors caused the
loss of 19% of the ! measurements during our measurement period.
Finally, the model comparison given in Chapter 7 was based on rainfall and streamflow time
series, in the same period studied in the Chapter 6 (March 2009 - March 2010).
Table 2. Measured hydrologic variables, record and gaps periods.
Variable
Frequency of
measurement
Measurement
period Data gaps Gap filling method
Rainfall R 10 minutes 01/01/09 - 06/03/10 No -
Streamflow
Q
10 minutes 01/01/09 - 18/07/09
23/07/09 - 06/03/10
18/07/09 - 23/07/09 Peak estimation by
peakflow/peak rainfall
analysis. Recesion
approached using the
current model
Evapo-
transpiration
ETR
20 Hertz 04/03/09 - 17/07/09
30/08/09 - 06/03/10
01/01/09 - 04/03/09
17/07/09 - 30/08/09
Penman-Monteith model
adjusting the canopy
conductance
Soil water
content "
7 to 15 days 02/04/09 - 07/12/09 01/01 - 02/04 -
Water table
level z
30 minutes 02/06/09 - 06/03/10 01/01 - 02/06 -
Turbidity ! 10 minutes 07/03/09 - 06/03/10 12/11/09 - 15/11/09
16/11/09 - 20/11/09
28/11/09 - 04/12/09
19/12/09 - 24/12/09
14/01/10 - 17/01/10
26/01/10 - 28/01/10
29/01/10 - 01/02/10
03/02/10 - 18/02/10
19/02/10 - 25/02/10
Explained in detail in the
Section 6.3.2 of this
thesis.
Plot erosion 1 week 10/03/09 - 06/03/10 - -
76
77
CHAPTER 5
Modelling the hydrological behaviour of a coffee agroforestry
basin in Costa Rica13
!" #$%&'()&*+,-./010" 2!" 3.456,7-/080" #!" *&"9,:7&/0" ;!" <,4+.47-&,4/0" =!" >?7&'80"9!" @,A"2:B&AC0" >!"
D,,6E/0"F!"3,5:-&*G080"H!"9!"I,7%,A-/0"9!"D.*E'J0"H!"9!"F.AA&K.A-L0">!"M%N,OP8",A-"3!"9.466,J"
/"QM3=)0"R93"SO.T;.*60" (8CUJU"9.AE5&**:&70" 7,AO&"
1"MQS0"QSV>S0"/UU81";,A"H.6?0"Q.6E,"3:O,"
8"Q=<MS0"L/LU0"8UGU/"<477:,*N,0"Q.6E,"3:O,"
C"QSI"S-:AN47+P0"F46P"S6E,E&0">&A:O4:W"SI1J"UXF0"RA:E&-"Y:A+-.%"
G"QM3=)0"R93";Z;<S90" (8CUJU"9.AE5&**:&70" 7,AO&"
J"MV3=0"R93"[M;=I0"1">*,O&"D:,*,0"8CUJU"9.AE5&**:&70" 7,AO&"
L"MV3=0"S>IZ;S0"F>"\/0"88\\8"D:**&A,@&"-]27A.A0" 7,AO&"
Q.77&65.A-&AO&"E.^" !"#$%&'()&*+,-."_ #.%&')`:O&!+.!O7a
Abstract
The profitability of hydropower in Costa Rica is affected by soil erosion and sedimentation in
dam reservoirs, which are in turn influenced by land use, infiltration and aquifer interactions
with surface water. In order to foster the provision and payment for Hydrological
Environmental Services (HES), a quantitative assessment of the impact of specific land uses
on the functioning of drainage-basins is required. The present paper aims to study the water
balance partitioning in a volcanic coffee agroforestry micro-basin (1 km2, steep slopes) in
Costa Rica, as a first step towards evaluating sediment or contaminant loads. The main
13 Paper accepted in Hydrology and Earth System Sciences.
Gómez-Delgado, F., Roupsard, O., le Maire, G., Taugourdeau, S., Pérez, A., van Oijen, M., Vaast, P., Rapidel,
B., Harmand, J. M., Voltz, M., Bonnefond, J. M., Imbach, P., and Moussa, R.: Modelling the hydrological
behaviour of a coffee agroforestry basin in Costa Rica, Hydrol. Earth Syst. Sci., 15, 369-392, doi:10.5194/hess-
15-369-2011, 2011.
http://www.hydrol-earth-syst-sci.net/15/369/2011
78
hydrological processes were monitored during one year, using flume, eddy-covariance flux
tower, soil water profiles and piezometers. A new Hydro-SVAT lumped model is proposed,
that balances SVAT (Soil Vegetation Atmosphere Transfer) and basin-reservoir routines. The
purpose of such a coupling was to achieve a trade-off between the expected performance of
ecophysiological and hydrological models, which are often employed separately and at
different spatial scales, either the plot or the basin. The calibration of the model to perform
streamflow yielded a Nash-Sutcliffe (NS) coefficient equal to 0.89 for the year 2009, while
the validation of the water balance partitioning was consistent with the independent
measurements of actual evapotranspiration (R2=0.79, energy balance closed independently),
soil water content (R2=0.35) and water table level (R2=0.84). Eight months of data from 2010
were used to validate modelled streamflow, resulting in a NS=0.75. An uncertainty analysis
showed that the streamflow modelling was precise for nearly every time step, while a
sensitivity analysis revealed which parameters mostly affected model precision, depending on
the season. It was observed that 64% of the incident rainfall R flowed out of the basin as
streamflow and 25% as evapotranspiration, while the remaining 11% is probably explained by
deep percolation, measurement errors and/or inter-annual changes in soil and aquifer water
stocks. The model indicated an interception loss equal to 4% of R, a surface runoff of 4% and
an infiltration component of 92%. The modelled streamflow was constituted by 87% of
baseflow originating from the aquifer, 7% of subsurface non-saturated runoff and 6% of
surface runoff. Given the low surface runoff observed under the current physical conditions
(andisol) and management practices (no tillage, planted trees, bare soil kept by weeding), this
agroforestry system on a volcanic soil demonstrated potential to provide valuable HES, such
as a reduced superficial displacement-capacity for fertilizers, pesticides and sediments, as well
as a streamflow regulation function provided by the highly efficient mechanisms of aquifer
recharge and discharge. The proposed combination of experimentation and modelling across
ecophysiological and hydrological approaches proved to be useful to account for the
behaviour of a given basin, so that it can be applied to compare HES provision for different
regions or management alternatives.
79
5.1 Introduction
The ability of ecosystems to infiltrate rainfall, sustain aquifers, and avoid erosion is a key
determinant for the provision of hydrological environmental services (HES), especially in the
humid tropics where surface fluxes can be very high (MEA, 2005). Woody plants and, in
particular, agroforestry (AF) systems associating shade trees and perennial crops with deep
root systems, are assumed to enhance these HES in comparison to traditional intensive
cropping systems (Ataroff and Monasterio, 1997; Vaast et al., 2005; Siles et al., 2010a).
Costa Rica is renowned as a promoter of HES by charging water users for the HES they
receive from land owners (e.g. forest conservation), focusing on water quality (Pagiola,
2008). Hydropower producers, generating 78% of the total electricity consumption in Costa
Rica during 2008 (ICE, 2009), are major HES payers. Coffee is one of the most traded
agricultural commodities in the world employing 100 million people (Vega and Rosenquist,
2001). In Costa Rica, coffee accounted for 15% of the agricultural exports in 2008 and
covered 2% of the territory (SEPSA, 2009). As coffee plantations are present in the main
basins used for hydroelectric generation in Costa Rica, the possible trade-offs of the payment
for HES from hydropower producers to coffee farmers become evident. Negotiation for these
payments is facilitated between providers and purchasers when the service, and/or the impact
of a given practice on the provision of the service, are clearly evaluated. However, links
between land use, management practices (like tree planting) and hydrology in Costa Rica
have not been thoroughly investigated by quantitative research (Anderson et al., 2006). There
is a need of both, experimentation at the basin scale in order to evaluate the main hydrological
processes, and of integrated modelling to understand the behaviour of all water compartments,
including hidden ones (e.g. the aquifer).
The partitioning of the water balance (WB) is a pre-requisite to evaluate HES such as
infiltration, aquifer regulation capacity, erosion control and contaminants retention in coffee
AF systems. Comprehensive WB studies at basin scale, including closure verification by
independent methods, have been carried out in the developed world and for other land covers,
like those reported by Roberts and Harding (1996), Dawes et al. (1997), Ceballos and
Schnabel (1998), Wilson et al. (2001) and Maeda et al. (2006). Some experimental basins are
located in the tropics, like those in Brazil, Costa Rica, Guadeloupe and Panamá (Fujieda et
al., 1997; Kinner and Stallard, 2004; Genereux et al., 2005; Charlier et al., 2008), but no
80
coffee AF basins have been equipped so far. Some reports are available for coffee AF systems
but at the plot level and for some particular fluxes such as throughfall and stemflow (Siles et
al., 2010b), tree and coffee transpiration (Dauzat et al., 2001; van Kanten and Vaast, 2006),
surface runoff (Harmand et al., 2007), energy balance and latent heat flux (Gutiérrez et al.,
1994). To our knowledge, there is no comprehensive study of the water balance partitioning
of coffee AF systems at the basin level, including the behaviour of the aquifer.
Truly balanced combinations of hydrological and ecophysiological experiments and models
remain scarce, although they intrinsically carry a more realistic and comprehensive
representation of plant, soil and aquifer components at plot and basin scales. Most
hydrological studies at basin scale use flumes for monitoring the streamflow and simply
estimate evapotranspiration (ET), which prevents a true verification of the water balance
closure.
As in the tropics we assumed that ET, including the re-evaporation of intercepted water (RIn),
is an important component of the water balance, even for precipitations around 3000 mm
year-1, we decided to measure it directly by eddy-covariance (Baldocchi and Meyers, 1998;
Wilson et al., 2001; Roupsard et al., 2006), choosing a 0.9 km2 micro-basin embedded in a
very homogeneous coffee AF plantation. As an additional advantage, the eddy-covariance
method can be validated itself by closing the energy balance (Falge et al., 2001).
Lumped, conceptual rainfall-streamflow models have been used in hydrology since the 1960s
(e.g. Crawford and Linsley, 1966; Cormary and Guilbot, 1969; Duan et al., 1992; Bergström,
1995; Donigan et al., 1995; Havnø et al., 1995; Chahinian et al., 2005). These models
consider the basin as an undivided entity, and use lumped values of input variables and
parameters. For the most part (for a review, see Fleming, 1975; Singh, 1995), they have a
conceptual structure based on the interaction between storage compartments, representing the
different processes with mathematical functions to describe the fluxes between the
compartments. Most hydrological models simplify the ET component based on reference ET
routines (Allen et al., 1998) or using very empirical, non-validated models for actual ET. For
instance, many models assume that actual evapotranspiration equals the reference ET, or a
constant fraction of it, or a variable fraction that depends only on the soil water content in the
root zone. However, improper parameterization of the crop coefficient may severely affect the
parameterization of hydrological resistances and fluxes. In constrast, ecophysiological models
81
may operate efficiently at plot level but miss the partitioning between lateral non-saturated
(subsurface) runoff and vertical drainage, and the dynamics of water in aquifers and rivers.
This is a major limitation for the assessment of HES, which is mainly desired at the basin
scale.
In the present study we attempted to couple two lumped models into a new integrative one,
chosen to be scalable and parsimonious: a basin reservoir model similar to the CREC model
(Cormary and Guilbot, 1969) and employing the Diskin and Nazimov (1995) production
function as proposed by Moussa et al. (2007a; 2007b), and the SVAT model proposed by
Granier et al. (1999). While the basin model was considered appropriate for its simplicity and
capacity to support new routines, the SVAT model was chosen for its parsimony (three
parameters in its basic formulation), its robustness (uses simple soil and stand data in order to
produce model runs for many years, avoiding hydraulic parameters that are difficult to
measure and scale up), its ability to quantify drought intensity and duration in forest stands,
and for its successful past validation in various forest stands and climatic conditions,
including tropical basins (Ruiz et al., 2010).
This paper aims to explain and model the hydrological behaviour of a coffee AF micro-basin
in Costa Rica, assessing its infiltration capacity on andisols. The methodology consists of
experimentation to assess the main water fluxes and modelling to reproduce the behaviour of
the basin. First, we present the study site and the experimental design. Second, we develop a
new lumped eco-hydrological model with balanced ecophysiological/hydrological modules
(that we called Hydro-SVAT model). This model was tailored to the main hydrological
processes that we recorded (streamflow, evapotranspiration, water content in the non-
saturated zone and water table level) and that are described in the subsequent sections. Third,
we propose a multi-variable calibration/validation strategy for the Hydro-SVAT model so we
calibrate using the streamflow in 2009 and validate using the remaining three variables in
2009 and the streamflow in 2010. Fourth, we make an uncertainty analysis to produce a
confidence interval around our modelled streamflow values, and a sensitivity analysis to
assess from which parameters this uncertainty might come. Fifth, we attempt to simplify the
initially proposed model based on the results of the sensitivity analysis. Finally, we discuss
the main findings concerning the water balance in our experimental basin.
82
5.2 The study site
5.2.1 Location, soil and climate
The area of interest is located in Reventazón river basin, in the Central-Caribbean region of
Costa Rica (Fig. 19a,b). It lies on the slope of the Turrialba volcano (central volcanic
mountain range of the country) and drains to the Caribbean Sea. The Aquiares coffee farm is
one of the largest in Costa Rica (6.6 km2), “Rainforest Alliance
TM” certified, 15 km from
CATIE (Centro Agronómico Tropical de Investigación y Enseñanza). Within the Aquiares
farm, we selected the Mejías creek micro-basin (Fig. 19c) for the “Coffee-Flux” experiment.
The basin is placed between the coordinates -83º44’39” and -83º43’35” (West longitude), and
between 9º56’8” and 9º56’35” (North latitude) and is homogeneously planted with coffee
(Coffea arabica L., var Caturra) on bare soil, shaded by free-growing tall Erythrina
poeppigiana trees. The initial planting density for coffee was 6,300 plants ha-1
, with a current
age >30 years, 20% canopy openness and 2.5 m canopy height. It is intensively managed and
selectively pruned (20% per year, around March). Shade trees have a density of 12.8 trees ha-
1, with 12.3% canopy cover and 20 m canopy height. The experimental basin has an area of
0.9 km2, an elevation range from 1,020 up to 1,280 m.a.s.l. and a mean slope of 20%.
Permanent streams extend along 5.6 km, implying a drainage density of 6.2 km km-2
. The
average slope of the main stream is 11%.
According to the classification by Mora-Chinchilla (2000), the experimental basin is located
along a 1.3 km wide strip of volcanic avalanche deposits, characterized by chaotic deposits of
blocks immersed in a matrix of medium-to-coarse sand, which is the product of the collapse
of the south-eastern slope of Turrialba volcano’s ancient crater. The general classification
given by the geological map of Costa Rica (MINAE-RECOPE, 1991) describes the general
stratigraphy as shallow intrusive volcanic rocks, and the particular region as proximal facies
of modern volcanic rocks (Quaternary), with presence of lava flows, agglomerates, lahars and
ashes. Soils belong to the order of andisols according to the USDA soil taxonomy (USDA,
1999), which are soils developing from volcanic ejecta, under weathering and mineral
transformation processes, very stable, with high organic matter content and biological activity
and very large infiltration capacities.
83
According to Köppen-Geiger classification (Peel et al., 2007), the climate is tropical humid
with no dry season and strongly influenced by the climatic conditions in the Caribbean
hillside. The mean annual rainfall in the study region for the period 1973-2009 was estimated
as 3014 mm at the Aquiares farm station (Fig. 20). At the experimental basin the rainfall in
2009 (3208 mm) was close to the annual mean, but showed a monthly mean deviation of ±100
mm around the historical regime. Mean monthly net radiation ranged in 2009 from 5.7 to 13.0
MJ m-2
d-1
, air temperature from 17.0 to 20.8 °C, relative humidity from 83 to 91%,
windspeed at 2 m high from 0.4 to 1.6 m s-1
and Penman-Monteith reference
evapotranspiration (Allen et al., 1998) from 1.7 to 3.8 mm d-1
.
5.2.2 Experimental setup
The “Coffee-Flux” experimental basin and instrument layout was designed to trace the main
water balance components employing spatially representative methods (Fig. 19c). It is part of
the FLUXNET network for the monitoring of greenhouse gases of terrestrial ecosystems. The
hydrological measurements were recorded from December 2008 up to February 2010.
Rainfall and climate: rainfall was monitored at 3 m above ground in the middle of 3 transects
of the basin, using three lab-intercalibrated ARG100 tipping-bucket (R.M. Young, MI, USA)
connected to CR800 dataloggers (Campbell Scientific, Shepshed, UK), and integrated every
10 min. Other climate variables were logged on top of the eddy-flux tower with a CR1000,
every 30 s, integrated half-hourly and using: Net radiation: NR-Lite (Kipp & Zonen, Delft,
The Netherlands); PPFD: Sunshine sensor BF3 (Delta-T devices Ltd, U.K.); temperature and
humidity: HMP45C in URS1 shelter (Campbell Scientific); wind-speed and direction: 03001
Wind Sentry (R.M. Young, MI, USA). The theoretical evapotranspiration from a wet grass
placed under local climate conditions, ET0, was computed in accordance with Allen et al.
(1998).
Streamflow: a long-throated steel flume (length: 3.9 m; width: 2.8 m; height: 1.2 m) was
home-built to measure the streamflow at the outlet of the experimental basin, to record up to 3
m3 s
-1, the maximum estimated discharge for the study period from an intensity-duration-
frequency analysis. The flume was equipped with a PDCR-1830 pressure transducer
(Campbell Scientific) to record water head at gauge point (30 s, 10 minutes integration), while
the rating curve was calculated considering the geometric and hydraulic properties of the
84
flume using Winflume software (Wahl et al., 2000). A validation of the rating curve was
made successfully using the salt dilution method as well as a pygmy current meter.
Soil water content: a frequency-domain-reflectometry portable probe (FDR Diviner2000,
Sentek Pty Ltd) was used to survey 20 access tubes distributed in the three study transects to
provide the mean volumetric soil water in the basin. The sensor measures at 10 cm intervals,
reaching a total depth of 1.6 m. A measurement campaign through the 20 sites was carried out
every week. The sensors were calibrated by digging sampling pits in the vicinity of six test
tubes, to obtain the actual volumetric soil water content from gravimetric content and dry bulk
density.
Evapotranspiration: the actual evapotranspiration from the soil, coffee plants and shade trees
was measured at reference height (26 m) on the eddy-covariance tower, similarly to Roupsard
et al. (2006). 3D wind components and temperature were measured with a WindMaster sonic
anemometer (Gill Instruments, Lymington, UK) at 20 Hz. H2O fluctuations were measured
with a Li-7500 open path (LiCor, Lincoln, NE, USA). Raw data were collected and pre-
processed by “Tourbillon” software (INRA-EPHYSE, Bordeaux, France) for a time-
integration period of 300 s, then post-processed using EdiRe software (University of
Edinburgh, UK) into half-hourly values and quality checked. A validation was made by direct
comparison of the measured net radiation Rn with the sum of sensible heat flux (H) and latent
heat flux ( E): at daily time step, this yielded H+ E=0.92 Rn (R2 = 0.93) which was
considered sufficiently accurate to assume that advection effects on E could be neglected
here. Due to lighting and sensor breakdown, 45 days of data were lost between July and
August 2009. To gap-fill the missing period we used the Penman-Monteith model, whose
canopy conductance was adjusted using measured values.
Leaf Area Index (LAI): the coffee light transmittance was measured monthly in diffuse light
conditions, for five rings at different zenital angles (LAI2000, Li-COR Corvallis, USA), along
three 50 m-long transects through the flux tower plot, similarly to Roupsard et al. (2008).
Effective coffee LAI, obtained from this light transmittance, was converted into actual LAI
according to Nilson (1971), using a ratio of effective to actual LAI that was estimated from a
dedicated calibration. The actual coffee LAI was measured directly on a small plot by
counting total leaf number of 25 coffee plants, measuring leaf length and width every 20
leaves and using empirical relationships between leaf length and width and leaf area (LI-
85
3100C, Li-COR) (R2 > 0.95). On the same small plot, the effective LAI was measured with
LAI2000. The ratio of effective to actual LAI was then calculated on this small plot (1.75) and
was considered to be constant with time and space in the micro-basin, allowing the estimation
of the actual LAI on the three LAI2000 transects. The LAI for shade trees was estimated using
their crown cover projection (on average 12.3% over the whole basin) observed on a very
high resolution panchromatic satellite image (WorldView image, February 2008, 0.5 m
resolution). As we did not have measurements of LAI for shade trees, we considered this LAI
in the order of magnitude of coffee LAI on a crown-projected basis, and therefore we
multiplied the actual coffee LAI measured on transects by 1.123, to estimate the ecosystem
LAI (tree and coffee). In order to monitor the time-course of ecosystem LAI at the basin scale,
we combined these ground measurements with time series of remotely-sensed images. We
used time series of Normalized Difference Vegetation Index (NDVI) from Moderate
Resolution Imaging Spectroradiometer (MODIS) data products MOD13Q1 and MYD13Q1
(16-Day composite data, 250m resolution). NDVI is known to be correlated with the green
LAI if it is low, for most ecosystems (Rouse et al., 1974). Twenty-three MODIS pixels
covering the experimental basin were selected, and their NDVI time series were downloaded.
We filtered the raw NDVI time series according to quality criterion given in the MODIS
products, and we adjusted a smooth spline function on it as in Marsden et al. (2010). Then, a
linear regression between the smoothed NDVI of the pixel including the flux tower and
ground values of actual ecosystem LAI was calibrated (R2 = 0.69). This regression was used
on other pixels of the basin, and averaged to have the annual time-course of actual LAI.
Water table level: four piezometric wells measuring up to 4 m depth were built in the three
main transects of study. They were equipped with pressure transducers (Mini-Divers,
Schlumberger Water Services) that measure and record the water table level every 30
minutes.
Period of measurement, data gaps and gap-filling: the recording information is given in Table
2 for the five hydrologic variables. The frequency of measurement varies, but is finally
calculated at the 30 minutes time step (except for soil water content that is a non-continuous
measurement). When gaps are present in the measurements, a gap filling method was applied.
86
5.3 Hydro!SVAT lumped model
We designed a lumped, five-reservoir-layer eco-hydrological model in order to predict the
water balance (WB) partitioning (stocks and fluxes) at the scale of the entire basin. It is based
on the water balance models developed by Moussa et al. (2007a; 2007b) and Granier et al.
(1999), and built to reproduce the main hydrological processes measured at the experimental
basin, which will be presented in Sect. 5.4. The model of Moussa et al. (2007a; 2007b) works
at the basin scale and simulates the ecosystem evapotranspiration rather roughly, while the
one of Granier et al. (1999) works at the plot scale and totally ignores the lateral water fluxes
through the soil and the role of the basin aquifers. The main novelties of the Hydro-SVAT
model with respect to the model structure of Moussa et al. (2007a; 2007b) are the inclusion of
a land cover reservoir to separate the intercepted rainfall from the combined
throughfall/stemflow component, and the partition of non-saturated soil into two reservoirs,
one with and one without roots of plants and trees. The first of these innovations intends to
take into account the non-negligible interception loss in coffee AF systems, as reported by
Jiménez (1986), Harmand et al. (2007) and Siles (2007). The second new addition to the
model is to better represent the water dynamics in the non-saturated soil, given that only its
upper layer will lose humidity by root extraction. The water balance model of Granier et al.
(1999) is incorporated in this superficial reservoir but in a simplified form, so that both, the
root distribution and the soil porosity, are homogeneous through the vertical, non-saturated
profile. Hence, the water content in this reservoir is the state variable linking our two parent
models.
The modelling hypotheses governing the model architecture were: a) the interception loss
component is not negligible in the WB and is a function of rainfall intensity, b) infiltration is a
function of the soil water content in the non-saturated reservoirs, c) evapotranspiration is a
significant component in the WB and is best described using a SVAT model that couples
evapotranspiration to root water extraction from the soil, d) the aquifer has a higher discharge
rate above a threshold level, e) any lateral inputs to the basins system are considered
negligible in comparison to rainfall inputs, and f) there is a net water outflow from the system
as deep percolation. The model was implemented using Matlab®
V. R2007a (The MathWorks
Inc., USA).
87
5.3.1 Model structure
The model structure is presented in Fig. 29. The next three sections will describe the model
structure according to its three major routines and five layers.
Figure 29. The lumped conceptual hydrological model proposed for the experimental basin.
The first layer is called “land cover reservoir” and separates the total rainfall into an
intercepted loss and a joint throughfall/stemflow component. The second layer or “surface
reservoir” regulates the surface runoff. The infiltration process from the second layer is
controlled by the joint water content at the third and fourth layers, called “non-saturated root
88
reservoir” and “non-saturated non-root reservoir”, respectively. The evapotranspiration flux is
calculated at the “non-saturated root reservoir”, while both non-saturated layers control the
drainage, the percolation and the non-saturated runoff processes. The fifth and last layer is the
“aquifer reservoir”, which determines the baseflow and the deep percolation. Finally, we will
explain the sum of the total runoff and baseflow components and the routing procedure to
generate the modelled streamflow. Let A(t), B(t), C(t), D(t) and E(t) [L] be the water levels at
time t in the five reservoirs A, B, C, D and E, respectively (or land cover reservoir, surface
reservoir, non saturated root reservoir, non-saturated no-root reservoir and aquifer reservoir).
Let AX, BX, CX, DX and EX [L] be the water levels corresponding to the maximum holding
capacities for the five reservoirs.
A. Infiltration and actual evapotranspiration
Infiltration
The infiltration process i [LT-1
] occurs from the second layer (surface reservoir) to the third
one (non-saturated root reservoir), and eventually to the fourth one (non-saturated non-root
reservoir) when i fills the third one. The infiltration capacity fi(t) [LT-1
] is a state variable that
depends on the water level in the non-saturated root reservoir, given by C(t). In addition to CX
(maximum) we define CF as the water level in the root reservoir at field capacity. Then, fi(t) is
calculated as (see Fig. 30a):
If ! FCtC " then ! ! ! 1
00
##$% Fci CtCffftf (38)
If ! FCtC & then ! ci ftf % (39)
where f0 [LT-1
] is the maximum infiltration capacity ( cff '%0 ) and fc [LT-1
] is the
infiltration rate at field capacity. The infiltration i both modifies and depends on !tB( , which
is the water availability in the second reservoir before i is extracted, according to:
If ! !tfttB i")( #1 then ! 1#)(% ttBi and ! 0%tB (40)
If ! !tfttB i&)( #1 then !tfi i% and ! ! ! ttftBtB i )#(% (41)
The infiltration module calculates the infiltration i as output variable, using the state variables
B(t), C(t) and fi(t). Four parameters (CX, CF, fc and !) are demanded.
89
Figure 30. a) Infiltration from surface reservoir as a function of soil water
content in the non-saturated non-root reservoir, b) The ratio r = T ET0-1 as a
function of relative extractable water REW and for different values of LAI
(here, LAI1 > LAI2 > … > LAIi), c) Surface runoff from the surface reservoir
as a function of its water content, d) Baseflow from the aquifer reservoir as a
function of its water content.
Evapotranspiration
The evapotranspiration component ET [LT-1
] acts directly on the third layer (non-saturated
root reservoir) and is the sum of Eu [LT-1
] the understory and soil evaporation, and of T [LT-1
]
the transpirational water uptake by roots.
TEET u $% (42)
According to Shuttleworth and Wallace (1985), the fraction of total evapotranspiration
originating from the plants is close to 100% of the total evapotranspiration of the ecosystem
when LAI >3 an when the soil is not saturated at its surface, which was always the case in our
study. We thus assumed for simplicity that Eu, the evaporation from the soil, was nil.
90
Transpiration T is obtained by solving T from the slightly modified ratio: r = T ET0-1
[dimensionless] proposed by Granier et al. (1999). We substituted the original Penman
potential evapotranspiration PET in that ratio by the Penman-Monteith reference
evapotranspiration ET0 [LT-1
] (Allen et al., 1998). While ET0 was calculated at each time step
t, we estimated r as a function of the relative extractable water REW(t) [dimensionless], a
state variable given by Granier et al. (1999) as:
! ! 1#% FCtCtREW (43)
The REW(t) is linked to the soil water content according to:
! !rf
rttREW
****
#
#% (44)
with "(t): volumetric soil water content [L3L
-3] at time t, "r: residual soil water content [L
3L
-3]
and "f: soil water content at field capacity [L3L
-3].
The parameter REWc [dimensionless] is the critical REW(t) below which the transpiration of
the system begins to decrease. Figure 30b shows an example of some r curves as a function of
REW(t). Each curve can be defined only by REWc and the rmLAI, a maximum value for the
ratio r that depends on the LAI of the system as:
mXLAI rLAILAIrm 1#% (45)
where LAIX is the maximum measured LAI during the modelling period and rm is a parameter
indicating the maximum ratio T ET0-1
that can be found in this system. Then:
If ! cREWtREW " then ! 1#% cLAI REWtREWrmr (46)
If ! cREWtREW & then LAIrmr % (47)
Finally, we find the transpiration as 0ETrT % . The total modelled evapotranspiration
including the interception loss, can be calculated as: Inum RTEETR $$% , with RIn [LT-1
]
being the intercepted/evaporated rainfall loss that will be explained in the next section. Hence,
ETRm can be directly compared to the evapotranspiration that we measured at the flux tower.
This module provides the evapotranspiration ET as a function of the state variable C(t), two
input variables (LAI and ET0) and three parameters (CX, REWc and rm).
91
B. Water balance in the model reservoirs
Land cover reservoir
The first layer of the model, denoted “land cover reservoir”, represents the soil cover in the
basin and controls the partition of the total incident rainfall R [LT-1
] into intercepted (then
evaporated) rainfall loss RIn [LT-1
] and the combined troughfall/stemflow RTS [LT-1
]. A simple
water balance of this reservoir is established to calculate a proxy !tA( [L] of the final water
level A(t) for each time step t, by adding the incident rainfall R and subtracting the Penman
potential evapotranspiration PET [LT-1
] from the existing land cover humidity level !1#tA :
! ! PETRtAtA #$#%( 1 (48)
We calculated the water level A(t) in this reservoir as well as RTS and RIn by differentiating
three cases:
If ! 0+( tA then ! 0%tA and ! RtARIn $#% 1 and 0%TSR (49)
If ! XAtA "("0 then ! !tAtA (% and PETRIn % and 0%TSR (50)
If ! XAtA &( then ! XAtA % and PETRIn % and ! XTS AtAR #(% (51)
The land cover module calculates at each time t the water level A(t) as a state variable,
demanding two input variables (R and PET) and one parameter (AX). It yields the partition of
R into RIn and RTS.
Surface reservoir
The second layer is called “surface reservoir” and acts as a sheet top soil with a given
roughness and surface runoff delaying properties. The water balance in this surface reservoir
for a given interval t is:
! ! iQQRtBtB BBTS ###$#% 211 (52)
where RTS [LT-1
] is the combined throughfall/stemflow component from the previous layer
and QB1 and QB2 [LT-1
] are the non-immediate and immediate surface runoffs calculated as:
!tBkQ BB %1 (53)
92
where kB [T-1
] is a discharge parameter, and:
If ! XBtB + then 02 %BQ (54)
If ! XBtB , then !- . 1
2
#)#% tBtBQ XB (55)
If QB2 > 0 then the water level B(t) is reset to BX. The infiltration i [LT-1
] is a function of the
water content in the third layer and it is the last component to be evaluated in the surface
reservoir.
This surface reservoir module calculates the water level B(t) as a state variable and demands
one input variable (RTS), and two parameters for the reservoir (BX and kB). It produces three
output variables: i, QB1 and QB2. The two latter variables constitute the surface runoff in the
basin (Fig. 30c).
Non!saturated root reservoir
The “non-saturated root reservoir” is the third layer of the model and it represents a soil layer
with presence of root systems from trees and plants. The water balance here is:
! ! CQddETitCtC ####$#% 211 (56)
where C(t) [L] is the state variable of the water level at a given time t, i is the infiltration from
the second layer, ET [LT-1
] is the evapotranspiration, d1 and d2 [LT-1
] are the non-immediate
and immediate drainages to the fourth layer, respectively and QC [LT-1
] is the non-saturated
runoff from the root reservoir.
There will be immediate drainage d2 if at anytime the RTS component fills the reservoir above
CX. Then !- . 1
2
#)#% tCtCd X goes to the fourth layer and C(t) is reset to CX. Both non-
immediate drainage d1 and non-saturated runoff QC occur whenever C(t) is higher than the
field capacity threshold CF [L] according to:
!- . CF kCtC #%/ (57)
where # [LT-1
] is the total outflow capacity in this reservoir and kC [T-1
] a discharge
parameter. The partition of # in d1 and QC depends on a parameter $ [dimensionless], with 0 <
01< 1. Then:
! /0#% 11d and /0%CQ (58)
93
The root soil module calculates the water level C(t) as state variable using two input variables
(i and ET) and four parameters (CX, CF, kC and 0). It provides three outputs (d1, d2 and QC).
Non!saturated non!root reservoir
The fourth layer of the model is denoted “non-saturated non-root reservoir” and represents a
soil layer with total absence of root systems and hence, of root water extraction. The water
balance here is given by:
! ! 21211 ggQddtDtD D ###$$#% (59)
where D(t) [L] is the state variable of the water level at a given time t, d1 and d2 [LT-1
] are the
non-immediate and immediate drainages from the third layer, respectively; QD [LT-1
] is the
non-saturated runoff from the non-root reservoir and g2 and g1 [LT-1
] are the immediate and
non-immediate percolation to the fifth model layer, respectively.
Immediate percolation g2 will be produced if at anytime the drainage (d2 and/or d1) fills the
reservoir above DX. Then !- . 1
2
#)#% tDtDg X moves to the aquifer reservoir and D(t) is
reset to DX. Both non-immediate percolation g1 and non-saturated runoff QD occur whenever
D(t) is higher than the field capacity threshold DF [L]:
!- . DF kDtD #%2 (60)
where % [LT-1
] is the total outflow capacity of this reservoir and kD [T-1
] a discharge
parameter. The partition of % in g1 and QD depends on the parameter $ [dimensionless]. Then:
!20#% 11g and 20%DQ (61)
The non-root soil module calculates the water level D(t) as state variable using two input
variables (d1 and d2) and four parameters (DX, DF, kD and 0). It provides three outputs (g1, g2
and QD).
Aquifer reservoir
A fifth layer called “aquifer reservoir” represents the groundwater system and controls
baseflow and deep percolation. The reservoir is composed by a shallow aquifer that acts
whenever the water level in the reservoir is higher than EX, and by a deep aquifer with a
permanent contribution. The water balance here is:
94
! ! DPQQggtEtE EE ###$$#% 21211 (62)
where E(t) [L] is the state variable of the water level at a given time t, g1 and g2 [LT-1
] are
respectively the non-immediate and immediate percolation from the fourth layer, QE1 and QE2
[LT-1
] are the baseflow from deep and shallow aquifers respectively (Fig. 30d), and DP [LT-1
]
is the deep percolation.
If ! XEtE + then !tEkQ EE 11 % and 02 %EQ (63)
If ! XEtE , then XEE EkQ 11 % and !- .XEE EtEkQ #% 22 (64)
!tEkDP E3% (65)
where kE1, kE2, and kE3 are discharge parameters controlling deep/shallow aquifers and deep
percolation, respectively.
This module calculates the water level E(t) as state variable using two input variables (g2 and
g1) and four parameters (EX, kE1, kE2 and kE3), to provide three outputs (QE1, QE2 and DP).
C. Total runoff, baseflow and streamflow
The components of surface runoff, non-saturated runoff and baseflow are added to obtain the
total runoff QT [LT-1
]:
QT = QB + QC + QD + QE (66)
As explained in Moussa and Chahinian (2009) the streamflow Q [LT-1
] at the outlet of the
basin is obtained by the routing of QT using a transfer function (to take into account the water
travel time). The Hayami (1951) kernel function (an approximation of the diffusive wave
equation) is developed to obtain a unit hydrograph linear model for this purpose. That is:
! ! !3 #%t
T dtHQtQ0
444 (67)
H(t) is the Hayami kernel function, equal to:
! !35
67
89:
;##
%67
89:
;%
0
23
2
2
1
1and dttHt
ezwtH
t
w
w
tz
F
F
< (68)
95
where w [T] is a time parameter that represents the centre of gravity of the unit hydrograph (or
the travel time) and zF [dimensionless] a form parameter. Q in [LT-1
] units can be transformed
to volume units [L3T
-1] multiplying it by the basin area [L
2].
5.3.2 Model parameterization, calibration and validation
Summarizing, this Hydro-SVAT model uses four input variables: rainfall R, Penman-
Monteith ET0, Penman PET and leaf area index LAI to generate five main output variables:
interception RIn, infiltration i, evapotranspiration ET, discharge components Q = QB + QC +
QD + QE (from surface, non-saturated and aquifer reservoirs, respectively) and deep
percolation DP.
Five state variables are calculated for every time step, the water levels in the five reservoirs:
A(t), B(t), C(t), D(t) and E(t). A coupled discharge QCD for the two non-saturated reservoirs is
obtained by adding QC and QD.
In our experimental basin we applied the model for a one-year period (2009), a time step
t=30 minutes (1800 s) and a basin area equal to 0.886 km2.
This model contains 20 parameters that are used to calculate infiltration (CX, DX, CF, DF, fc
and !), evapotranspiration (REWc and rm), the exchange between reservoirs (AX, BX, kB, kC, kD,
0, EX, kE1, kE2 and kE3) and the basin transfer function (w and zF).
Four out of these twenty parameters (CX, DX, CF, DF) were estimated using field data. For
instance, two excavation experiments down to 3.5 m showed that very few roots were present
below 1.5 m, where the andisol layer turns into a more clayey, compact and stony deposit.
Then, the depth of the non-saturated root soil layer was fixed at CH = 1.6 m (for simplicity,
equal to the length of our FDR probe tubes). The depth of the non-root layer was estimated in
DH = 1.0 m. Following the relationships ! HrsX CC ** #% and ! HrsX DD ** #% , the levels
for maximum water holding capacities in the non-saturated reservoirs can be calculated, as
well as the levels for field capacities, using the equations ! HrfF CC ** #% and
! HrfF DD ** #% . The volumetric soil water contents " were estimated as "r=0.37: the
residual water content equal to the minimum " observed in the basin during the study period;
"s =0.63: the " at saturation for a typical andisol, according to Hodnett and Tomasella (2002);
96
and "f =0.43: the " at field capacity equal to the average van Genuchten value for a matric
potential = -10 kPa (Hodnett and Tomasella, 2002).
Three parameters were taken from literature reviews and expert criteria (AX, REWc and rm).
We set the surface reservoir maximum storage capacity AX for our coffee AF system equal to
4×10-4
m using data from Siles et al. (2010b). The two parameters of the evapotranspiration
routine were taken as: REWc=0.4 from Granier et al. (1999) and rm=0.8 from field
measurements of T ET0-1
(data not shown).
One parameter (BX) was obtained at the end of the optimization process in order to fit the
maximum observed streamflow peak (this parameter is very sensitive as it acts directly on the
highest peaks). Two other parameters (w and zF) were separately estimated by a trial and error
procedure, given the low sensitivity of the model to their variation.
The remaining 10 empirical parameters (fc, !, kB, kC, kD, 0, EX, kE1, kE2 and kE3) were
simultaneously optimized. For this purpose we used the Nelder-Mead (Nelder and Mead,
1965) simplex algorithm included in Matlab (following Lagarias et al., 1998) on 17520 semi-
hourly time steps (one year). Convergence was reached within 1000 runs and the stabilization
of all parameter values by the end of the iteration process was checked. A two-step calibration
procedure was applied: a) selection of an initial value for each parameter, falling within the
respective range (fourth column, Table 3), and b) simultaneous estimation of parameter values
that maximize an objective function (sixth column of Table 3, identified as M1), in this case
the Nash and Sutcliffe (1970) efficiency coefficient.
We calibrated the model using one year of streamflow Q, and then validated it using the
remaining three measured variables: evapotranspiration ETR, water content in the non-
saturated zone * and water table level z. For ETR we grouped the measured and modelled
values (given in the same units) at the daily time scale, which is the original time scale in
Granier’s model, and then we excluded the gap-filled values. To calculate the modelled *
from the water level in the root reservoir C(t) we used the relation
! ! rrsXCtC **** $#% #1, while the observed values were obtained as the average of the 20
FDR point-measurements throughout the basin. To validate z we proposed an effective
porosity of the aquifer nA= 0.39, to be able to directly link piezometric measurements z with
the modelled water level in our aquifer reservoir E(t), according to z=E(t)/nA.
97
Tab
le 3
. P
aram
eter
des
crip
tio
n,
ran
ge
for
op
tim
izat
ion
, o
pti
miz
ed v
alu
e an
d r
ang
e fo
r se
nsi
tiv
ity
an
aly
sis.
M1:
ori
gin
al 1
0-p
aram
eter
mo
del
, M
2:
sim
pli
fied
7-p
aram
eter
mo
del
.
Pa
ram
eter
Des
crip
tio
n
Un
its
Ra
ng
e fo
r
op
tim
iza
tio
n R
efer
ence
O
pti
mu
m
va
lue
M1
Op
tim
um
va
lue
M2
0 V
erti
cal/
late
ral
spli
t co
effi
cien
t to
div
ide
ou
tpu
ts f
rom
no
n-s
atu
rate
d r
eser
vo
irs
into
no
n-s
atu
rate
d r
un
off
and
ver
tica
l fl
ow
s
frac
tio
n
[0 -
0.8
4]
Mo
uss
a an
d C
hah
inia
n
(20
09
)
0.0
32
0
.031
k B
Dis
char
ge
rate
fo
r su
rfac
e re
serv
oir
s-1
[0
- 1
]×10
-4
Em
piri
cal p
aram
eter
2.
12×
10-5
2.11
×10
-5
f c
Infi
ltra
tion
rat
e at
fie
ld c
apac
ity
m s
-1
[0 -
1]×
10-5
M
inim
um s
tead
y st
ate
infi
ltra
b. in
the
expe
rim
. bas
in (
Kin
oshi
ta, p
ers.
com
m.,
2009
)7.
45×
10-6
-
Coe
ffic
ient
to c
alcu
late
the
max
imum
in
filt
rati
on c
apac
ity
from
fie
ld c
apac
ity
dim
ensi
onle
ss[1
- 7
0]
Mou
ssa
and
Cha
hini
an
(200
9)
102
-
k C
Dis
char
ge c
oeff
icie
nt, t
otal
out
puts
fro
m n
on-
satu
rate
d ro
ot r
eser
voir
s-1
[0
- 1
]×10
-4
Em
piri
cal p
aram
eter
1.
02×
10-4
1.06
×10
-4
k D
Dis
char
ge c
oeff
icie
nt f
or to
tal o
utpu
ts f
rom
no
n-sa
tura
ted
non-
root
res
ervo
ir
s-1
[0 -
1]×
10-4
E
mpi
rica
l par
amet
er
6.65
×10
-5-
EX
Thr
esho
ld le
vel i
n th
e aq
uife
r re
serv
oir,
ab
ove
whi
ch a
sha
llow
-aqu
ifer
out
let i
s fo
und
m
[0 -
1]
Em
piri
cal p
aram
eter
0.
341
0.33
9
k E2
Dis
char
ge c
oeff
icie
nt f
or b
asef
low
fro
m
shal
low
aqu
ifer
res
ervo
ir
s-1
[0 -
2.4
]×10
-6 C
harl
ier
et a
l. (
2008
) 1.
02×
10-6
9.81
×10
-7
k E1
Dis
char
ge c
oeff
icie
nt f
or b
asef
low
fro
m d
eep
aqui
fer
rese
rvoi
r s-1
[0
- 2
.1]×
10-6
Cha
rlie
r et
al.
(20
08)
1.58
×10
-71.
50×
10-7
k E3
Dis
char
ge c
oeff
icie
nt f
or d
eep
perc
olat
ion
from
the
aqui
fer
rese
rvoi
r s-1
[0
- 1
]×10
-7
Em
piri
cal p
aram
eter
4.
36×
10-8
4.29
×10
-8
98
We also carried out an independent validation of Q by running the calibrated Hydro-SVAT
model over 8 months of data from 2010 and verifying the model performance.
In addition, we performed an analysis of model residuals: zero expectancy, normality,
homoscedasticity and standardized residuals.
5.3.3 Uncertainty and sensitivity analysis
In order to investigate the uncertainty in model predictions we performed a Monte-Carlo
approach on a restricted subset of parameter combinations, as suggested by Helton (1999).
For each of the 10 parameters a range was created with a deviation of ±30% around the
optimum value found in the calibration process (like in White et al., 2000; Ines and Droogers,
2002; Lenhart et al., 2002; Zaehle et al., 2005; Droogers et al., 2008), following the “one-at-
a-time” method described by Hamby (1994; 1995) and Frey and Patil (2002). Then, we
assumed a uniform distribution for all the parameters, and used the Latin Hypercube function
of Simlab 2.2 (http://simlab.jrc.ec.europa.eu) in order to produce a sample from a joint
probability distribution, of size equal to ten times the number of parameters as recommended
by SimLab developers, i.e., 100 parameter combinations. These combinations were
introduced into the calibrated Hydro-SVAT model and we retrieved 100 streamflow output
series over the 17520 semi-hourly time steps. Then we generated 17520 empirical confidence
intervals (95% and 99%) from the respective frequency distributions over the 100 parameter
combinations.
A sensitivity analysis (SA) was carried out to determine which of the input variables
contributed significantly to this modelling uncertainty. Two separate assessments were
produced. First, a summary (through the time) index of model performance was studied: the
Nash-Sutcliffe coefficient (NS), being evaluated by four sensitivity indexes: Pearson,
Spearman, standardized regression and standardized rank regression coefficients (Hamby,
1994, 1995). Then, a second approach of sensitivity analysis was tested to try to follow the
behaviour of the Spearman coefficient for each of the 10 parameters included in the SA
through all time steps, as in Helton (1999).
The sensitivity indexes are calculated departing from the joint probability sample matrix xij of
size m×n, where m is the sample size and n the number of independent variables (here our 10
99
parameters) to study. The Monte Carlo evaluation of xij in the model produces the result
vector yi, configuring the matrix system [yi : xij]. Then, the Pearson product moment
correlation (PEAR) for a given parameter j is the linear correlation coefficient between the
variables xij and yi over the m samples. To account for non-linear relationships that can be
hidden by indicators like PEAR, a simple rank transformation is applied, replacing the
original x-y series with their corresponding ranks R(x) and R(y). Then, the Spearman
coefficient (SPEA) is obtained by calculating the correlation on the transformed data, as
SPEA(x, y) = PEAR[R(x), R(y)]. The standardized regression coefficients (SRC) are the result
of a linear regression analysis performed on [yi : xij] but previously standardizing all the
variables. This is useful to evaluate the effect of the independent variables on the dependent
one, without regarding their units of measurement, and can be computed by standard
statistical methods. Finally, the standardized rank regression coefficients are obtained as
SRRC(x, y) = SRC[R(x), R(y)].
5.4 Results
In the 0.9 km2 micro-basin of Mejías creek, within the Cafetalera Aquiares AF coffee farm,
we obtained one full year (2009) of comprehensive experimental results of streamflow,
evapotranspiration, soil water content and piezometry that we used to calibrate and validate
the Hydro-SVAT model. First we present the hydrological behaviour of the basin, then the
ecophysiological behaviour, and finally the water balance.
5.4.1 Hydrological behaviour of the basin
The time series of streamflow Q and rainfall R are given at a semi-hourly time-step in Fig.
31a for 2009. Rainfall (Fig. 20) was quite evenly distributed (no marked dry spell), although
the period from January to June clearly received less rain (later named the ‘drier season’, in
opposition to the ‘wetter season’). From the total R (3208 mm), the measured Q at the outlet
was 2048 mm, yielding an annual streamflow coefficient of 0.64. The Q hydrograph (Fig.
31a) displays a continuous baseflow with episodes of groundwater recharge after rainfall
events, followed by marked recessions controlled by the baseflow. Q peaks reached an annual
maximum of 0.84 m3 s
-1, i.e. 28% of the nominal capacity of the flume, indicating that the
size chosen for the flume was adequate.
100
Figure 31. Measurements in the experimental basin for 2009: a) rainfall R and streamflow
Q, t=30 min, b) leaf area index (LAI) within 1 std. dev. confidence bands (grey), c)
reference (ET0) and measured (ETR) daily evapotranspiration, d) soil water content ! with
one standard deviation bars and e) water table level z in the four piezometers throughout
the basin. Graphs on the right side correspond to a two-month period (June-July 2009) for
better illustration.
101
It can be observed that similar rainfall events resulted in higher Q peaks during the wetter
season.The lower Q peaks in response to rainfall events during the drier season could thus be
interpreted as the consequence of higher infiltration rates when belowground was less
saturated. Soil water content in the 0 - 1.6 m layer (Fig. 31d) remained above 37% all-year
round and rose up to a maximum of 47% during the transition between the drier and the
wetter season. In order to understand the large baseflow shaping the hydrograph of Q, four
piezometers were installed in early June 2009 (Fig. 31e) and showed the existence of a
permanent aquifer at levels varying between 0.7 and 3.2 m deep, according to their respective
distance to water channels. Their behaviour was extremely variable, as expected, from
responsive (piezos #1 and #3) to conservative behaviours (piezos #2 and #4). The continuous
baseflow observed by the flume originated mainly from an important aquifer, covering a large
(although undefined) area within the basin.
5.4.2 Ecophysiological behaviour of the basin
The ecosystem LAI (Fig. 31b) changed seasonally quite severely from 2.8 (in March), as a
result of coffee pruning during the drier season and leaf shedding by E. poeppigiana, coffee
flowering and new leafing just after the beginning of the wetter season, to a maximum of 4.8
(in September), then coffee leaf shedding during the main coffee-berry harvest (in October).
The actual evapotranspiration (ETR) obtained by eddy-covariance (Fig. 31c) accounted for the
sum of coffee and shade-tree transpiration, understory evaporation (mainly bare soil), and
rainfall interception loss. The total ETR in 2009 amounted to 818 mm (25% of R), fluctuating
daily according to atmospheric demand and seasonally according to LAI and canopy
conductance. It always stayed below 4.5 mm d-1 and, in average, around 60% of reference
ET0, clearly invalidating the use of ET0 as a reliable indicator of ETR in coffee system eco-
hydrological models. The crop coefficient (the ETR ET0-1 ratio) was clearly lower during the
drier season, as a consequence mainly of a lower LAI. We did not observe a period during
which the relative extractable water (REW) of the soil drop below the critical value (REWc) of
0.4, confirming that the coffee plants probably encountered no seasonal water stress.
102
5.4.3 Water balance partitioning and closure
Figure 32 shows the water balance partition and closure for 2009, as obtained by the water
flows measured by independent experimental methods, rainfall (R), streamflow (Q), and
evapotranspiration (ETR) (including the rainfall interception loss). It was observed on
cumulative values that the sum of Q+ETR was 11% lower than annual R. However, Q+ETR
matched or exceeded R once at the end of April, just after three episodes of lower rainfall,
which confirmed the occurrence of an important storage in the basin, namely the soil and
aquifer reservoirs, creating a seasonal hysteresis between R and Q. On an annual basis,
measured Q represented 64% of R and measured ETR amounted 25%. The remaining 11%
was attributed to deep percolation, measurement errors and/or inter-annual changes in soil and
aquifer water stocks.
Figure 32. Water balance in the experimental basin for 2009. Only measured values are presented here.
Figure 33a shows the result of the streamflow modelling for the year 2009, after calibrating
the 10 parameters from Table 3. The model yielded a NS coefficient of 0.89 and R2 = 0.88 for
N = 17520 semi-hourly time-steps. Baseflow and peakflow events appeared to be
satisfactorily represented for the whole time series. The modelled partitioning of streamflow
into surface runoff, non-saturated runoff and baseflow is presented in Fig. 33b. It indicated a
prominent contribution of baseflow, as already inferred from visual inspection of the Q time
series.
103
Figure 33. Model results: a) measured vs. modelled streamflow and b) components of modelled streamflow. Water level in the model reservoirs: c) surface reservoir, d) non-saturated root reservoir (offset by !r CH to show the absolute water content in the soil layer), e) non-saturated non-root reservoir (offset by !r DH) and f) aquifer reservoir. Graphs on the right side correspond to a two-month period (June-July 2009) for better illustration.
Hillslope surface runoff and non saturated runoff were a minor part of Q. The water stored in
the 4 lowest reservoirs of the model is shown in Figs 33c to 33f. The water level at the surface
104
reservoir was rarely greater than zero, displaying some few events of surface storage beyond
the modelling time step (30 minutes). The non-saturated reservoirs were replenished during
storm events and depleted by non-saturated runoff and evapotranspitation (this latter acting
only in the root reservoir). Finally, the aquifer reservoir displayed the largest magnitude of
variation, fluctuating seasonally by a factor of almost 4 (this factor is reduced to 1.3 when the
aquifer effective porosity is taken into account).
5.5 Discussion
In the next sections we first discuss the validation, uncertainty and sensitivity analysis of the
model, second some model simplification attempts, third the main hydrological processes that
we observed, and finally we examine the hydrological services in our coffee AF basin.
5.5.1 Validation, uncertainty and sensitivity analysis for the Hydro!SVAT model
A. Model validation
Model validation was done by direct comparison of model output variables with three field
measurements: actual evapotranspiration ETR, soil water content ! and aquifer water level z.
For ETR, the determination coefficient between the daily sum of observed and modelled ETR
was R2=0.79 (Fig. 34a). Concerning !, the observed and modelled time series appeared to be
consistent, but divergent during the driest season (Fig. 34b), reaching a rather low R2 = 0.35.
However, considering that this global ! was obtained as the arithmetic average of only 20
observations for the whole basin, the uncertainty of these measured values is high. The large
amount of rocks hindering the tubes might also affect the FDR readings. Finally, we obtained
a good approximation for the behaviour of the aquifer (Fig. 34c), with R2=0.84. The use of the
two piezometers that displayed the highest stability (probably representing the larger and
more relevant aquifer systems) out of the four piezometers installed was crucial at this step.
We considered to be recording two different processes, the typical gradual aquifer response
(piezos #2 and #4 in Fig. 19c) and the local quick-varying shallow-water accumulations
(piezos #1 and #3 in Fig. 19c), which might be associated with rapid changes in soil water
contents at smaller and less representative aquifer units.
105
Figure 34. Validation of: a) daily evapotranspiration ETRm, b) soil water content ! at the 1.6 m root-reservoir, c) water table level z and d) streamflow Q for eight months of 2010. (R2: determination coefficient, RRMSE: relative root mean squared error).
106
Though the representativeness of these few piezometers of the behaviour of the main basin-
aquifer could be questioned, the very high similarity that we found between the modelled
values and the average measurements from piezometers #2 and #4 (located in two opposite
ends of the basin), supported the idea of a correct performance of both, the field method for
aquifer monitoring and the corresponding model routine that is based on a linear reservoir.
Concerning the precision of our model, the NS = 0.89 on streamflow seems to be good
enough for a semi-hourly time step, considering the detail with which the hydrological
processes need to be described. Some studies that have evaluated hydrological models using
this coefficient for different time scales have shown the decline in NS values as the modelling
time step is shortened. For instance, at calibration stages, Notter et al. (2007) achieved
maximum NS values from 0.8 to 0.69 for decadal to daily time steps (basin area = 87 km2),
while Bormann (2006) obtained 0.8, 0.9, 0.85 and 0.73 for annual, monthly, weekly and daily
time steps, respectively (basin area = 63 km2). García et al. (2008) reached NS coefficients of
0.93, 0.91 and 0.61 for quarterly, monthly and daily modelling in a 162 km2 basin.
In addition, the streamflow validation over an independent eight-month test period (January-
August 2010) produced a NS=0.75, which is considered satisfactory (Fig. 34d). During this
first eight months of 2010 the total rainfall amounted to 1978 mm (compared to 2222 mm in
the first eight months of 2009), the total streamflow summed 1057 mm (against 1445 mm in
2009) and 11 storm events exceeded the threshold of 0.3 m3 s-1 (compared to 13 events in
2009).
To study possible systematic errors in the Hydro-SVAT model, we examined the distribution
of residuals between measured and modelled streamflows (since this was the optimized
variable). A t-test with a confidence level of 95% indicated that we cannot conclude that the
mean of our residuals (equal to -1×10-4 m3 s-1) is significantly different from zero (p=0.36). At
our short time step it is common to find highly autocorrelated residuals, so we found
significant partial autocorrelations up to the seventeen time lag (8.5 hours). The Kolmogorov-
Smirnov test revealed that the distribution of residuals is not normal, though it is highly
symmetrical around zero. Studying residuals as functions of time, rainfall R, streamflow Q,
soil water content ! and water table level z (data not shown), we did not find any trends, but
noticeable changes in their variability make clear that the homoscedasticity condition was not
107
properly fulfilled. However, it is accepted that these ordinary least squares assumptions are
often not satisfied in streamflow modelling (Xu and Singh, 1998).
B. Uncertainty analysis
If we accept that the model reproduces efficiently the actual streamflow at the outlet of the
experimental basin, the uncertainty in the modelled values needs to be known. A Monte-Carlo
(MC) uncertainty analysis was produced as detailed in Sect. 5.3.3 to yield the empirical 95%
and 99% confidence limits (CL) (the 95% CL are presented in Fig. 35, along with measured
streamflow). 82% of the measured values fell within the 95% CL produced by our model,
while 87% of them fell within the 99% CL. The ranges of the 95% confidence intervals (CI)
along each of the time steps varied from 20% to 131% of the MC mean value, while for the
99% CI the ranges were from 24% up to 157% of the MC mean. From these analyses it is
possible to state that the model is efficient (high NS coefficient), precise (relatively small
confidence intervals, Fig. 35) and accurate (given the high percentages of measured Q values
falling within the 95% and 99% confidence intervals).
Figure 35. Measured streamflow against empirical 95% confidence interval.
108
C. Sensitivity analysis
The sensitivity analysis was carried out to determine the responsiveness of the model
predictions to variations in our main parameters. The first assessment consisted in testing the
statistical significance of the relationships between each of our 10 calibration parameters and
the NS coefficient, using four dimensionless sensitivity indexes: Pearson (PEAR), Spearman
(SPEA), standardized regression and standardized rank regression coefficients (SRC and
SRRC, respectively). Table 4 presents the results of these tests revealing that, with the
exception of the discharge coefficients from the root non-saturated and shallow-aquifer
reservoirs (kC and kE2), and of the shallow-aquifer threshold (EX), none of the parameters
seemed to significantly influence the global model efficiency. However, these results may be
misleading given the enormously variable conditions under which the model works through
time. Therefore, in a second approach we selected the Spearman test (a simple rank
transformation index that can identify non linear relationships) to assess the influence of each
of these 10 parameters on streamflow, at each time step.
Table 4. Sensitivity indexes for each of the 10 calibration parameters vs. the NS coefficient. PEAR: Pearson product moment correlation coefficient, SPEA: Spearman coefficient, SRC: Standardized regression coefficient and SRRC: Standardized rank regression coefficients. Index values in bold are statistically significant at 95% confidence level. Sensitivity index
Parameter PEAR SPEA SRC SRRC
0.08 0.07 0.09 0.08
kB -0.04 -0.04 -0.06 -0.05
fc -0.06 -0.10 -0.05 -0.07
-0.13 -0.11 -0.04 -0.06
kC -0.20 -0.24 -0.13 -0.16
kD -0.07 -0.11 -0.08 -0.10
EX 0.18 0.05 0.22 0.11
kE2 -0.29 -0.29 -0.27 -0.24
kE1 0.10 0.09 0.06 0.05
kE3 0.02 -0.06 -0.01 -0.06
109
Figure 36 presents the results, displaying the dimensionless Spearman index in the vertical
axis and the time in the horizontal axis. A positive Spearman index reveals a proportional
influence of the parameter on the model result, while negative values indicate inverse
proportionality. Figure 36a shows the time series for all the parameters, revealing an
alternation in their influence on streamflow over time and model state. The two black
horizontal lines represent the 95% confidence limits above (or below) which the correlation is
significantly different from zero. In Figs. 36b to 36k the significant values are plotted as black
points in contrast to non-significant in grey, for each parameter separately. Figure 36h
suggests that the parameter EX is again one of the most influential, because it is permanently
displaying high positive or negative correlations, together with the discharge coefficient (DC)
for the deep-aquifer kE1, which most of the time has a proportional influence on model outputs
(Fig. 36j). These two parameters reach the highest Spearman indexes during sustained periods
(not only during storm events) and are clearly relevant during long streamflow recessions,
when the modelled water table level z is around or below EX. The DC for the surface reservoir
kB (Fig. 36c) is noticeably influencing model outputs during each individual storm event,
while the vertical/lateral split coefficient (Fig. 36b) is relevant only during the main ones.
The DC for the deep percolation kE3 (Fig. 36k) is significant over recession periods, while DC
for root reservoir kC is significant when its water content exceeds field capacity (Fig. 36f).
The DC for the shallow aquifer kE2 (Fig. 36i) has proportional influence on streamflow when z
is above EX, or else, inverse influence during the driest part of the recessions. Finally, the
infiltration rate at field capacity (fc), the coefficient for maximum infiltration rate ( ) and the
DC for non-root reservoir (kD) seem to have no relevant effects on the streamflow modelling
(Figs. 36d, 36e and 36g, respectively).
5.5.2 Model simplification
A model simplification was carried out by assuming that the three less sensitive parameters
could be considered as modelling constants. The original 10-parameter model (identified as
M1) was converted to a 7-parameter model (called M2) in the search for model parsimony and
faster convergence. The parameters fc, and kD were fixed according to the optimal values
from the original model. Hence, we assumed that fc = 7.45×10-6 m s-1, = 102 and kD =
6.65×10-5 s-1.
110
Figure 36. Spearman indexes between Q and: a) all the parameters: black horizontal lines indicate the 95% confidence interval out of which the Spearman is significant, b) partition parameter , c) surface discharge kB, d) infiltration rate at field capacity fc, e) maximum infiltration , f) root-reservoir discharge kC, g) non-root discharge kD, h) threshold for shallow aquifer EX, i) shallow aquifer discharge kE2, j) deep aquifer discharge kE1 and k) deep percolation discharge kE3. The black dots indicate the time steps for which the Spearman is significant at the 95% confidence level, while grey dots indicate non-significant correlations.
111
This simplified model M2 converged after 1524 runs and reached a NS = 0.88, a performance
that is almost as good as that of M1. The optimal values for M2 are presented in the seventh
column of Table 3, and are very close to the optimal for M1. Given the algorithmic
configuration of Nelder-Mead procedure, the convergence of M2 to the same parametric
optimal than M1 is not assured (since the new model structure defines a completely different
simplex problem, on a smaller 7-dimensional space). From this result, an almost identical
water balance is obtained from M2. In any case, M2 should be validated under different
conditions (other basins and spatio-temporal scales), in order to corroborate that the numerical
values that we assigned to the three fixed parameters can be exported to different contexts,
and can always be treated like modelling constants. The simplification here undertaken also
proves the effectiveness of the time-varying sensitivity analysis here proposed, as a tool to
identify the main parametric sources of modelling uncertainty and, therefore, the potentially
redundant model parameters.
An additional simplification step was carried out by separately removing two lateral
flowpaths that could be unnecessary in the model structure (QB2 and QE2), given that
reservoirs B (surface) and E (aquifer) already have primary flowpaths (QB1 and QE1).
However, while suppressing QB2 affects the simulation of the main peakflow (that decreases
from 0.84 m3 s-1 to 0.52 m3 s-1), the exclusion of QE2 enormously worsens the model’s ability
to shape the hydrograph recessions (and therefore the aquifer recharge/discharge processes),
yielding a NS=0.71 for the optimized model. From this exercise we conclude that these two
flowpaths (and their respective parameters) play a clearly identifiable and relevant role in the
model structure.
5.5.3 Hydrological processes in the experimental basin
The main hydrological processes and components observed using this measuring/modelling
approach are presented in the following four sub-sections.
A. Interception, throughfall, stemflow and surface runoff
A review of water balance (WB) partitioning in comparable situations is proposed in Table 5
(interception loss, evapotranspiration, surface/non-saturated runoff, baseflow, change in soil
water content and deep percolation).
11
2
Tab
le 5
. Com
pari
son
of a
nnua
l wat
er b
alan
ce c
ompo
nent
s (a
s %
of
rain
fall
) fo
r di
ffer
ent s
tudi
es a
t plo
t and
bas
in s
cale
s, in
trop
ical
reg
ions
. S
ou
rce
Lo
cati
on
C
lim
ate
B
asi
n a
rea
(km
2)
La
nd
co
ver
L
AI
(m2 m
-2)
Ra
infa
ll
R (
mm
)
RIn
a
Ta
To
tal
ET
Ra
ET
0a
NT
d
QB
aQ
CD
aQ
E a
To
tal
Qa
S
aD
a
DP
,
SO
a
Jim
énez
(1
986)
C
osta
Ric
a H
umid
tr
opic
al
plot
sca
le
Cof
fee
and
E.
po
epp
igia
na
-
26
42
b 1
6
53
69
- -
- -
- -
9 22
-
Imba
ch e
t a
l.
(198
9)
Cos
ta R
ica
Hum
id
trop
ical
pl
ot s
cale
C
offe
e an
d E
.
po
epp
igia
na
-
19
19
4
42
46
60
-
- -
- -
- 54
-
Har
man
d et
al.
(200
7)
Cos
ta R
ica
Hum
id
trop
ical
pl
ot s
cale
C
offe
e an
d E
.
deg
lup
ta
3.5
26
22
b 1
5
23
3
8
- -
2-3
- -
- 6
54
-
Sil
es (
2007
),
Can
navo
et
al.
(2
011)
Cos
ta R
ica
Hum
id
trop
ical
pl
ot s
cale
C
offe
e an
d I.
den
sifl
ora
5.
0-6.
02
68
4-
32
45
1
1-1
5 2
8-3
44
1-4
639
-44
0.14
3-6
- -
- -1
-144
-55
-
Les
ack
(199
3)B
razi
l H
umid
tr
opic
al
0.23
R
ain
fore
st
- 2
87
0
- -
39
- -
3 -
- 5
7
2
- 1
Fuj
ieda
et
al.
(1
997)
B
razi
l H
umid
su
btro
p.
0.56
T
hree
laye
r fo
rest
-
23
19
1
5
15
30
32
- 5
6 59
7
0
- -
-
Kin
ner
and
Sta
llar
d (2
004)
Pan
ama
Hum
id
trop
ical
0.
10
Tro
pica
l fo
rest
-
24
00
-
- 53
56
-
4 17
20
4
1
6c -
6c
Gen
ereu
x et
al.
(20
05)
Cos
ta R
ica
Hum
id
trop
ical
0.
26
Tro
pica
l rai
n fo
rest
-
49
74
-
- 32
-46
- -
- -
- 5
4-6
8-
- -
Cha
rlie
r et
al.
(2
008)
G
uade
loup
e M
arit
ime
hum
. tro
p.0.
18
Ban
ana
- 4
22
9
- -
31
31
- 10
- 17
2
7
0 59
42
Thi
s st
udy
Cos
ta R
ica
Hum
id
trop
ical
0.
90
Cof
fee
and
E.
po
epp
igia
na
3.
8 3
20
8
4 20
2
5
32
0.16
4 4
56
64
0
69
11-1
2
* N
umbe
rs in
bol
d ar
e m
easu
red
quan
titi
es.
a Giv
en a
s %
of
R:
RIn
: in
terc
epti
on l
oss,
T:
tran
spir
atio
n, E
TR
: ev
apot
rans
pira
tion
, E
T0:
ref
eren
ce e
vapo
tran
spir
atio
n, Q
B:
surf
ace
runo
ff,
QC
D:
non-
satu
rate
d ru
noff
(su
bsur
face
flo
w),
QE:
base
flow
, Q
: st
ream
flow
, D
: dr
aina
ge,
S:
chan
ge i
n so
il w
ater
con
tent
, D
P:
deep
pe
rcol
atio
n, S
O: s
ubsu
rfac
e ou
tflo
w.
b Ann
ual e
stim
atio
n fo
r ex
peri
men
ts c
ondu
cted
in s
hort
term
per
iods
(le
ss th
an a
yea
r).
c The
aut
hors
igno
re w
heth
er th
is f
ract
ion
is b
eing
sto
red
in s
oil,
or it
bec
ame
subs
urfa
ce o
utfl
ow d
owns
trea
m th
e ga
ugin
g si
te.
d NT
= n
orm
aliz
ed tr
ansp
irat
ion
equa
l to
T (
ET
0 L
AI)
-1
113
In our basin, the adjusted interception loss (RIn) equalled 4% of input rainfall (R). This value
is the same as found by Imbach et al. (1989) in a WB experiment under similar coffee and E.
poeppigiana land cover, but is lower than other reports obtained by direct measurements at
plot scale in Costa Rica: Jiménez (1986) found 16% under the same AF system; Harmand et
al. (2007) found 15% for coffee and Eucalyptus deglupta (but this RIn could be lower because
stemflow was not separately measured); Siles (2007) found 11-15% under coffee and Inga
densiflora. Interception loss can be greatly affected by the local LAI of both layers, the
specific architecture of the coffee and trees and the rainfall regime.
From the throughfall/stemflow component in our model (equal to 96% of R), 4% of R came
out of the basin as surface runoff QB. This value is not far from other reports at plot scale, like
those by Ávila et al. (2004): 1-9% (coffee and E. deglupta); Harmand et al. (2007): 2% and
Siles (2007): 3-6% (coffee and I. densiflora). At basin scale, Fujieda et al. (1997) measured 5%
on a 0.56 km2 basin under a three layer forest in Brazil, Lesack (1993) modelled 3% in a rain-
forest basin in Brazil (area: 0.23 km2), Kinner and Stallard (2004) modelled 4% in a
Panamanian tropical-forest basin (0.10 km2) and Charlier et al. (2008) modelled 10% in a
banana-plantation basin in Guadeloupe (0.18 km2). A major source of surface runoff and
discrepancy between plot and basin scale studies could be the presence of roads. In our
experimental basin the total length of roads is 10 km and it represents 4.5% of the basin area.
Then, assuming a direct road-runoff of 80% (Ziegler et al., 2004), this component could
represent up to 95% of the total basin surface runoff.
B. Infiltration
The non-intercepted and non-runoff fraction of incident rainfall was infiltrated (i = 92% of R).
This large i/R ratio and a QB/i ratio close to 4% give indication of very high infiltration and
drainage capacities, which are typical of andic-type volcanic soils (Poulenard et al., 2001;
Cattan et al., 2006) and are further enhanced for perennial crops in the absence of tillage and
in presence of substantial macroporosity (Dorel et al., 2000). It was indicated by preliminary
measurements carried out in our experimental basin with a Cornell infiltrometer (Ogden et al.,
1997) that steady state infiltrability values could be as high as 4.7×10-5 m s-1 (168 mm h-1)
(Kinoshita, pers. comm., 2009). Recent experiments found hydraulic conductivity values as
high as 3.4×10-5 m s-1 (122 mm h-1) and 2.1×10-5 m s-1 (75 mm h-1) for andisols in Costa Rica
114
(Cannavo et al., 2011) and Guadeloupe (Charlier et al., 2008), respectively. From our
infiltration capacity (that was modelled as a function of soil water content in the non-saturated
reservoirs), we calculated the infiltration/rainfall (i/R) and the runoff/infiltration (QB/i) ratios
at the storm-event scale, analyzing 78 events with cumulative rain higher than 10 mm. No
significant changes were found in any of these two ratios as a function of time (or season),
and only slight reductions were detected for increasing storm cumulative rainfall or soil water
content. This fact, added to the permanently high i/R ratios for each individual event (65-
98%), are explained by the constantly high infiltration capacity of andisols and the stable
behaviour of soil water content throughout the year. We therefore observed very efficient soil
and aquifer recharge mechanisms (Fig. 33d to 33f). According to Dorel et al. (2000) the soil
properties of non-tilled perennial crops in andisols are mainly determined by wetting-drying
cycles and by biological activity in the soil. Those are relatively stable factors in our
experimental basin, given the absence of a well-defined dry season on this side of the country
(Caribbean influence), and the permanently high organic matter content in these soils.
C. Evaporation and transpiration
Transpiration of coffee plants and trees T accounts for 20% of R (645 mm y-1, obtained from
modelling). Higher transpiration values were reported by van Kanten and Vaast (2006) in a
coffee AF system under E. poeppigiana (29%, 897 mm y-1), and also by Siles (2007), who
measured values ranging from 31% (1008 mm y-1) to 34% (905 mm y-1) (see Table 5). In
other experimental plots in Costa Rica, estimations of T varied between 53% (750 mm y-1)
and 42% (811 mm y-1) (Jiménez, 1986; Imbach et al., 1989, respectively). T can be highly
dependent on local ET0, on the effect of drought on stomatal closure, and on LAI. For a better
site comparison, we computed a simple “normalized transpiration” index NT = T (ET0 LAI)-1
in Table 5 and found that our value (0.16) is very close to the respective ratio in the study of
Siles (2007).
As mentioned earlier, we measured the actual evapotranspiration (ETR=T+Eu+RIn) in our
experimental basin by the eddy-covariance method and it represented 25% of R (804 mm y-1).
This is the smallest value reported in comparison to the other studies on coffee AF systems
(Table 5). The closest value (38%, 956 mm in nine months) was measured by Harmand et al.
(2007), but it is about 1.5 times our ETR; while a maximum of 69% (985 mm y-1) was
reported by Jiménez (1986). It must be stressed that our study is the only one that brings
115
independent validation of ETR through energy balance closure, whereas many plot studies
might carry errors due to the calibration of sapflow, the model of Eu or the sampling of RIn. At
the basin scale, many authors modelled ETR values around 30% and 46% in tropical
experimental basins. In absolute terms these values are: 1120, 554-682, 1961-2345 and 1300
mm y-1 (Lesack, 1993; Fujieda et al., 1997; Genereux et al., 2005; Charlier et al., 2008,
respectively).
D. Drainage, deep percolation and streamflow
Another component of interest at plot scale is drainage (vertical flow beyond root reservoir),
which we modelled as 69% of R and seems, on average, slightly higher than the values
reported in the literature (Table 5). The modelled deep percolation (or subsurface outflow
downstream the basin outlet) was 12%, much lower than the 42% reported by Charlier et al.
(2008), but similar to the 6% encountered by Kinner and Stallard (2004). As deep percolation
is calculated from water balance closure, its accuracy is enhanced when Q and ETR are
precisely measured.
The first (or the second) highest WB basin output is usually streamflow Q, which in our
experimental basin was recorded and modelled as 64% of R. It seems to be very close to
similar measurements in the tropics (Table 5). The baseflow from the aquifer accounted for
87% of total Q, while surface runoff was only 6% and non-saturated runoff 7%.
5.5.4 The coffee agroforestry basin and hydrological services
Modelling the hydrological behaviour of this experimental, coffee AF basin gave some
insights on the provision of HES by this system. This is particularly relevant in the Costa
Rican context where HES payments have already been implemented as national
environmental protection policies (Pagiola, 2008). Two main services related to water quality
can be recognized, both linked to the observed high infiltration i (92% of R) and low surface
runoff QB (4% of R).
At first, the low QB in the basin is closely associated to low surface displacement of fertilizers,
pesticides and sediments (Leonard and Andrieux, 1998; Bruijnzeel, 2004; Cattan et al., 2006;
Cattan et al., 2009). We found very constant QB/i ratios through the time, or under different
rainfall intensities, which may come from the expected stability in soil hydraulic properties
116
(e.g. high infiltration capacity) and the absence of either a marked dry season that controls
soil desiccation (Dorel et al., 2000; Park and Cameron, 2008) or mechanized agricultural
practices like tillage affecting soil compaction, surface roughness, continuity of pores,
macroporosity, soil cover and organic matter content (Le Bissonnais et al., 2005; Chahinian et
al., 2006). However, the high drainage capacity of these andisols might be a disadvantage in
terms of percolation and groundwater contamination by agrochemicals (Cattan et al., 2007;
Saison et al., 2008) given our model estimates for groundwater recharge of around 67% of R.
In addition, the modelled surface runoff for the experimental basin includes possible
discharges from unpaved roads and ditches, which needs to be controlled to avoid excessive
water, sediment and contaminant flux concentrations.
A second HES might be the streamflow regulation function provided by this AF basin through
aquifer recharge/discharge mechanisms. With a measured evapotranspiration close to 25% of
R (which is presumably much lower than the equivalent for forests), soil and aquifer water
depletion seems unlikely under the observed hydrogeological and climatic conditions,
favouring water availability during dry seasons (Robinson et al., 2003; Bruijnzeel, 2004). On
the other hand, during intense rainfalls and tropical storms the aquifer is efficiently recharged,
as we have observed in our piezometric measurements. The result is a homogeneous seasonal
distribution of streamflow, with a high rainfall recuperation fraction of 64% (Q/R).
5.6 Conclusions
This paper gives some insights into the assessment of Hydrological Environmental Services
(HES) by studying the hydrological processes in a particular micro-basin. The water balance
partition is proposed as a baseline for analyses and negotiations leading to the payment for
HES, for which measuring and modelling approaches are complementary. The understanding
of water dynamics supplied a better knowledge about the main services provided by the
studied ecosystem, as well as some potential vulnerabilities.
The general behaviour of the coffee AF basin (1 km2) on andisols can be summarized by the
fact that 92% of R was infiltrated through the highly permeable andisol, 64% of R was
measured as streamflow, 25% of R was measured as evapotranspiration, no major seasonal
variation was detected in the soil water stock, and a large aquifer contribution to streamflow
was observed in the shape of baseflow (Fig. 33b). These are characteristics of a system prone
117
to generate important HES at basin scale, which is a result infrequently reported for coffee
systems.
We proposed an original modelling approach coupling a hydrological and a SVAT model,
calibrated using the streamflow at the outlet of the basin, but validated by independent and
direct measurements of streamflow (test period), evapotranspiration, soil water content and
water table level.
We presented a standard uncertainty analysis in order to built simulation confidence intervals
around our modelled streamflow values, as well as a sensitivity analysis to investigate the
source of such uncertainty. The first parameterization of the model is considered adequate,
though model simplification could be attempted centred on the three less sensitive parameters.
Special attention needs to be given to direct measurement of a representative field capacity
and the associated probability distribution function.
The conceptual nature of our Hydro-SVAT model allows a wide time/space domain of
application, conditional only on knowledge of some general properties of the basin of interest
and on the acquisition of basic hydrological data. Different environments can be configured in
terms of climate, land cover, soils and hydrogeology, and further applications under different
conditions are desired to test the generality of the model. Complementary studies like
hillslope and channel surface runoff, basin water losses through roads, temporal variation in
soil and ecophysiological properties and ground water dynamics and composition are
expected.
Acknowledgements
This work was funded by the Centre de Coopération Internationale en Recherche
Agronomique pour le Développement (CIRAD, France), the European project CAFNET
(EuropAid/121998/C/G) and the Cafetalera Aquiares farm (http://www.cafeaquiares.com).
CATIE (Centro Agronómico Tropical de Investigación y Enseñanza) provided collaboration
and infrastructure. We gratefully thank the Cafetalera Aquiares farm for allowing us to install
the experimental basin and for providing us research facilities. Rintaro Kinoshita contributed
to saturated infiltrability observations and Álvaro Barquero and his family backed the field
work. Discussions with Lars Gottschalk, Irina Krasovskaia, Sadí Laporte and Rafael Chacón
118
provided useful insight. We would also like to thank the two anonymous reviewers for their
constructive and effective comments. Federico Gómez-Delgado was supported by a CIRAD
fellowship during his PhD.
References
Allen, R. G., Pereira, L. S., Raes, D., and Smith, M.: Crop evapotranspiration: Guidelines for
computing crop water requirements, Food and Agriculture Organization, Rome, 301 pp.,
1998.
Ataroff, M., and Monasterio, M.: Soil erosion under different management of coffee
plantations in the Venezuelan Andes, Soil Technology, 11, 95-108, 1997.
Ávila, H., Harmand, J. M., Dambrine, E., Jiménez, F., Beer, J., and Oliver, R.: Dinámica del
nitrógeno en el sistema agroforestal Coffea arabica con Eucalyptus deglupta en la Zona Sur de
Costa Rica, Agroforestería en las Américas, 41-42, 83-91, 2004.
Baldocchi, D., and Meyers, T.: On using eco-physiological, micrometeorological and
biogeochemical theory to evaluate carbon dioxide, water vapor and trace gas fluxes over
vegetation: a perspective, Agr. Forest Meteorol., 90, 1-25, 1998.
Bergström, S.: The HBV model, in: Computer Models of Watershed Hydrology, edited by:
Singh, V. P., Water Resources Publications, Colorado, USA, 1995.
Bormann, H.: Impact of spatial data resolution on simulated catchment water balances and
model performance of the multi-scale TOPLATS model, Hydrol. Earth Syst. Sci., 10, 165-
179, 2006.
Bruijnzeel, L. A.: Hydrological functions of tropical forests: not seeing the soil for the trees?,
Agriculture, Ecosystems & Environment, 104, 185-228, 2004a.
Bruijnzeel, L. A.: Hydrological functions of tropical forests: not seeing the soil for the trees?,
Agr. Ecosyst. Environ., 104, 185-228, 2004b.
Cannavo, P., Sansoulet, J., Harmand, J. M., Siles, P., Dreyer, E., and Vaast, P.: Agroforestry
associating coffee and Inga densiflora results in complementarity for water uptake and
119
decreases deep drainage in Costa Rica, Agr. Ecosyst. Environ., 140, 1-13, doi:
10.1016/j.agee.2010.11.005, 2011.
Cattan, P., Cabidoche, Y. M., Lacas, J. G., and Voltz, M.: Effects of tillage and mulching on
runoff under banana (Musa spp.) on a tropical Andosol, Soil Till. Res., 86, 38-51, 2006.
Cattan, P., Voltz, M., Cabidoche, Y. M., Lacas, J. G., and Sansoulet, J.: Spatial and temporal
variations in percolation fluxes in a tropical Andosol influenced by banana cropping patterns,
J. Hydrol., 335, 157-169, 2007.
Cattan, P., Ruy, S. M., Cabidoche, Y. M., Findeling, A., Desbois, P., and Charlier, J. B.:
Effect on runoff of rainfall redistribution by the impluvium-shaped canopy of banana
cultivated on an Andosol with a high infiltration rate, J. Hydrol., 368, 251-261, 2009.
Ceballos, A., and Schnabel, S.: Hydrological behaviour of a small catchment in the dehesa
landuse system (Extremadura, SW Spain), J. Hydrol., 210, 146-160, 1998.
Cormary, Y., and Guilbot, A.: Relations pluie-débit sur le bassin de la Sioule, Rapport
D.G.R.S.T. No. 30, Université des Sciences et Techniques, Montpellier, France, 1969.
Crawford, N. H., and Linsley, R. K.: Digital simulation in hydrology: Stanford watershed
model IV, Technical Report 39, Department of Civil Engineering, Stanford University,
California, USA, 210, 1966.
Chahinian, N., Moussa, R., Andrieux, P., and Voltz, M.: Comparison of infiltration models to
simulate flood events at the field scale, J. Hydrol., 306, 191-214, 2005.
Chahinian, N., Moussa, R., Andrieux, P., and Voltz, M.: Accounting for temporal variation in
soil hydrological properties when simulating surface runoff on tilled plots, J. Hydrol., 326,
135-152, 2006.
Charlier, J. B., Cattan, P., Moussa, R., and Voltz, M.: Hydrological behaviour and modelling
of a volcanic tropical cultivated catchment, Hydrol. Process., 22, 4355-4370, 2008a.
Charlier, J. B., Cattan, P., Moussa, R., and Voltz, M.: Hydrological behaviour and modelling
of a volcanic tropical cultivated catchment, Hydrol. Process., 22, 4355-4370, 2008b.
120
Dauzat, J., Rapidel, B., and Berger, A.: Simulation of leaf transpiration and sap flow in virtual
plants: model description and application to a coffee plantation in Costa Rica, Agr. Forest
Meteorol., 109, 143-160, 2001.
Dawes, W. R., Zhang, L., Hatton, T. J., Reece, P. H., Beale, G. T. H., and Packer, I.:
Evaluation of a distributed parameter ecohydrological model (TOPOG_IRM) on a small
cropping rotation catchment, Journal of Hydrology, 191, 64-86, 1997.
Diskin, M. H., and Nazimov, N.: Linear reservoir with feedback regulated inlet as a model for
the infiltration process, J. Hydrol., 172, 313-330, 1995.
Donigan, A., Bicknell, B., and Imhoff, J. C.: Hydrological simulation program – Fortran
(HSPF), in: Computer Models of Watershed Hydrology, edited by: Singh, V. P., Water
Resources Publications, Colorado, USA, 1995.
Dorel, M., Roger-Estrade, J., Manichon, H., and Delvaux, B.: Porosity and soil water
properties of Caribbean volcanic ash soils, Soil Use Manage., 16, 133-140, 2000.
Droogers, P., Van Loon, A., and Immerzeel, W. W.: Quantifying the impact of model
inaccuracy in climate change impact assessment studies using an agro-hydrological model,
Hydrol. Earth Syst. Sci., 12, 669-678, 2008.
Duan, Q., Sorooshian, S., and Gupta, V.: Effective and efficient global optimization for
conceptual rainfall-runoff models, Water Resour. Res., 28, 1015-1031,
doi:10.1029/91WR02985, 1992.
Falge, E., Baldocchi, D., Olson, R., Anthoni, P., Aubinet, M., Bernhofer, C., Burba, G.,
Ceulemans, R., Clement, R., Dolman, H., Granier, A., Gross, P., Grünwald, T., Hollinger, D.,
Jensen, N.-O., Katul, G., Keronen, P., Kowalski, A., Ta Lai, C., Law, B. E., Meyers, T.,
Moncrieff, J., Moors, E., William Munger, J., Pilegaard, K., Rannik, Ü., Rebmann, C.,
Suyker, A., Tenhunen, J., Tu, K., Verma, S., Vesala, T., Wilson, K., and Wofsy, S.: Gap
filling strategies for long term energy flux data sets, Agr. Forest Meteorol., 107, 71-77, 2001.
Fleming, G.: Computer Simulation Techniques in Hydrology, Environmental Science Series,
Elsevier, New York, USA, 1975.
Frey, H. C., and Patil, S. R.: Identification and Review of Sensitivity Analysis Methods, Risk
Anal., 22, 553-578, 2002.
121
Fujieda, M., Kudoh, T., de Cicco, V., and de Calvarcho, J. L.: Hydrological processes at two
subtropical forest catchments: the Serra do Mar, São Paulo, Brazil, J. Hydrol., 196, 26-46,
1997.
García, A., Sainz, A., Revilla, J. A., Álvarez, C., Juanes, J. A., and Puente, A.: Surface water
resources assessment in scarcely gauged basins in the north of Spain, J. Hydrol., 356, 312-
326, 2008.
Genereux, D. P., Jordan, M. T., and Carbonell, D.: A paired-watershed budget study to
quantify interbasin groundwater flow in a lowland rain forest, Costa Rica, Water Resour.
Res., 41, W04011, doi:dx.doi.org/10.1029/2004WR003635, 2005.
Granier, A., Breda, N., Biron, P., and Villette, S.: A lumped water balance model to evaluate
duration and intensity of drought constraints in forest stands, Ecol. Model., 116, 269-283,
1999.
Gutiérrez, M. V., Meinzer, F. C., and Grantz, D. A.: Regulation of transpiration in coffee
hedgerows: covariation of environmental variables and apparent responses of stomata to wind
and humidity, Plant Cell Environ., 17, 1305-1313, 1994.
Hamby, D. M.: A review of techniques for parameter sensitivity analysis of environmental
models, Environ. Monit. Assess., 32, 135-154, 1994.
Hamby, D. M.: A Comparison of Sensitivity Analysis Techniques, Health Phys., 68, 195-204,
1995.
Harmand, J.-M., Ávila, H., Dambrine, E., Skiba, U., de Miguel, S., Renderos, R., Oliver, R.,
Jiménez, F., and Beer, J.: Nitrogen dynamics and soil nitrate retention in a Coffea arabica -
Eucalyptus deglupta agroforestry system in Southern Costa Rica, Biogeochemistry, 85, 125-
139, 2007.
Havnø, K., Madsen, M. N., and Dørge, J.: MIKE 11 – a generalized river modelling package,
in: Computer Models of Watershed Hydrology, First ed., edited by: Singh, V. P., Water
Resources Publications, Colorado, USA, 1995.
Hayami, S.: On the propagation of flood waves, Disaster Prev. Res. Inst. Bull., 1, 1-16, 1951.
Helton, J. C.: Uncertainty and sensitivity analysis in performance assessment for the Waste
Isolation Pilot Plant, Computer Physics Communications, 117, 156-180, 1999.
122
Hodnett, M. G., and Tomasella, J.: Marked differences between van Genuchten soil water-
retention parameters for temperate and tropical soils: a new water-retention pedo-transfer
functions developed for tropical soils, Geoderma, 108, 155-180, 2002.
ICE: Plan de Expansión de la Generación Eléctrica, Período 2010–2021, Instituto
Costarricense de Electricidad, San José, Costa Rica, 2009.
Imbach, A. C., Fassbender, H. W., Beer, J., Borel, R., and Bonnemann, A.: Sistemas
agroforestales de café (Coffea arabica) con laurel (Cordia alliodora) y café con poró
(Erythrina poeppigiana) en Turrialba, Costa Rica. VI. Balances hídricos e ingreso con lluvias
y lixiviación de elementos nutritivos, Turrialba, 39, 400-414, 1989.
Ines, A. V. M., and Droogers, P.: Inverse modelling in estimating soil hydraulic functions: a
Genetic Algorithm approach, Hydrol. Earth Syst. Sci., 6, 49-65, 2002.
Jiménez, F.: Balance hídrico con énfasis en percolación de dos sistemas agroforestales: café-
poró y café-laurel, en Turrialba, Costa Rica, MSc Thesis, CATIE, Turrialba, Costa Rica, 104
pp., 1986.
Kinner, D. A., and Stallard, R. F.: Identifying storm flow pathways in a rainforest catchment
using hydrological and geochemical modelling, Hydrol. Process., 18, 2851-2875, 2004.
Lagarias, J. C., Reeds, J. A., Wright, M. H., and Wright, P. E.: Convergence Properties of the
Nelder-Mead Simplex Method in Low Dimensions, SIAM J. Optimiz., 9, 112-147, 1998.
Le Bissonnais, Y., Cerdan, O., Lecomte, V., Benkhadra, H., Souchère, V., and Martin, P.:
Variability of soil surface characteristics influencing runoff and interrill erosion, Catena, 62,
111-124, 2005.
Lenhart, T., Eckhardt, K., Fohrer, N., and Frede, H. G.: Comparison of two different
approaches of sensitivity analysis, Phys. Chem. Earth, 27, 645-654, 2002.
Leonard, J., and Andrieux, P.: Infiltration characteristics of soils in Mediterranean vineyards
in Southern France, Catena, 32, 209-223, 1998.
Lesack, L. F. W.: Water Balance and Hydrologic Characteristics of a Rain Forest Catchment
in the Central Amazon Basin, Water Resour. Res., 29, 759-773, doi:10.1029/92WR02371
1993.
123
Maeda, K., Tanaka, T., Park, H., and Hattori, S.: Spatial distribution of soil structure in a
suburban forest catchment and its effect on spatio-temporal soil moisture and runoff
fluctuations, J. Hydrol., 321, 232-256, 2006.
Marsden, C., le Maire, G., Stape, J.-L., Seen, D. L., Roupsard, O., Cabral, O., Epron, D.,
Lima, A. M. N., and Nouvellon, Y.: Relating MODIS vegetation index time-series with
structure, light absorption and stem production of fast-growing Eucalyptus plantations, Forest
Ecol. Manag., 259, 1741-1753, 2010.
MEA: Ecosystems and Human Well-Being: Synthesis., Millennium Ecosystem Assessment,
Island Press, Washington, USA, 155, 2005.
MINAE-RECOPE: Mapa geológico de Costa Rica, Scale 1:200000, San José, Costa Rica,
1991.
Mora-Chinchilla, R.: Geomorfología de la cuenca del Río Turrialba (unpublished),
Universidad de Costa Rica, San José, Costa Rica, 2000.
Moussa, R., Chahinian, N., and Bocquillon, C.: Erratum to "Distributed hydrological
modelling of a Mediterranean mountainous catchment - Model construction and multi-site
validation" [J. Hydrol. 337 (2007) 35-51], J. Hydrol., 345, 254-254, 2007a.
Moussa, R., Chahinian, N., and Bocquillon, C.: Distributed hydrological modelling of a
Mediterranean mountainous catchment - Model construction and multi-site validation, J.
Hydrol., 337, 35-51, 2007b.
Moussa, R., and Chahinian, N.: Comparison of different multi-objective calibration criteria
using a conceptual rainfall-runoff model of flood events, Hydrol. Earth Syst. Sci., 13, 519-
535, 2009.
Nash, J. E., and Sutcliffe, J. V.: River flow forecasting through conceptual models part I - A
discussion of principles, J. Hydrol., 10, 282-290, 1970.
Nelder, J. A., and Mead, R.: A Simplex Method for Function Minimization, Comput. J., 7,
308-313, 10.1093/comjnl/7.4.308, 1965.
Nilson, T.: A theoritical analysis of the frequency of gaps in plant stands, Agr. Meteorol., 8,
25-38, 1971.
124
Notter, B., MacMillan, L., Viviroli, D., Weingartner, R., and Liniger, H.-P.: Impacts of
environmental change on water resources in the Mt. Kenya region, J. Hydrol., 343, 266-278,
2007.
Ogden, C. B., van Es, H. M., and Schindelbeck, R. R.: Miniature Rain Simulator for Field
Measurement of Soil Infiltration, Soil Sci. Soc. Am. J., 61, 1041-1043,
10.2136/sssaj1997.03615995006100040008x, 1997.
Pagiola, S.: Payments for environmental services in Costa Rica, Ecol. Econ., 65, 712-724,
2008.
Park, A., and Cameron, J. L.: The influence of canopy traits on throughfall and stemflow in
five tropical trees growing in a Panamanian plantation, Forest Ecol. Manag., 255, 1915-1925,
2008.
Peel, M. C., Finlayson, B. L., and McMahon, T. A.: Updated world map of the Köppen-
Geiger climate classification, Hydrol. Earth Syst. Sci., 11, 1633-1644, 2007.
Poulenard, J., Podwojewski, P., Janeau, J.-L., and Collinet, J.: Runoff and soil erosion under
rainfall simulation of Andisols from the Ecuadorian Páramo: effect of tillage and burning,
Catena, 45, 185-207, 2001.
Roberts, G., and Harding, R. J.: The use of simple process-based models in the estimate of
water balances for mixed land use catchments in East Africa, Journal of Hydrology, 180, 251-
266, 1996.
Robinson, M., Cognard-Plancq, A. L., Cosandey, C., David, J., Durand, P., Führer, H. W.,
Hall, R., Hendriques, M. O., Marc, V., McCarthy, R., McDonnell, M., Martin, C., Nisbet, T.,
O'Dea, P., Rodgers, M., and Zollner, A.: Studies of the impact of forests on peak flows and
baseflows: a European perspective, Forest Ecol. Manag., 186, 85-97, 2003.
Roupsard, O., Bonnefond, J.-M., Irvine, M., Berbigier, P., Nouvellon, Y., Dauzat, J., Taga, S.,
Hamel, O., Jourdan, C., Saint-André, L., Mialet-Serra, I., Labouisse, J.-P., Epron, D., Joffre,
R., Braconnier, S., Rouzière, A., Navarro, M., and Bouillet, J.-P.: Partitioning energy and
evapo-transpiration above and below a tropical palm canopy, Agr. Forest Meteorol., 139, 252-
268, 2006.
125
Roupsard, O., Dauzat, J., Nouvellon, Y., Deveau, A., Feintrenie, L., Saint-André, L., Mialet-
Serra, I., Braconnier, S., Bonnefond, J.-M., Berbigier, P., Epron, D., Jourdan, C., Navarro, M.,
and Bouillet, J.-P.: Cross-validating Sun-shade and 3D models of light absorption by a tree-
crop canopy, Agr. Forest Meteorol., 148, 549-564, 2008.
Rouse, J. W., Haas, R. H., Schell, J. A., and Deering, D. W.: Monitoring vegetation systems
in the Great Plains with ERTS, Proc. Third Earth Resources Tech. Satellite-1 Symp. Vol. 1:
Technical Presentations, Sec. A, NASA SP-351, Washington, USA, 1974.
Ruiz, L., Varma, M. R. R., Kumar, M. S. M., Sekhar, M., Maréchal, J.-C., Descloitres, M.,
Riotte, J., Kumar, S., Kumar, C., and Braun, J.-J.: Water balance modelling in a tropical
watershed under deciduous forest (Mule Hole, India): Regolith matric storage buffers the
groundwater recharge process, J. Hydrol., 380, 460-472, 2010.
Saison, C., Cattan, P., Louchart, X., and Voltz, M.: Effect of Spatial Heterogeneities of Water
Fluxes and Application Pattern on Cadusafos Fate on Banana-Cultivated Andosols, J. Agr.
Food Chem., 56, 11947-11955, 2008.
SEPSA: Boletín Estadístico Agropecuario 19, Secretaría Ejecutiva de Planificación Sectorial
Agropecuaria, San José, Costa Rica, 2009.
Shuttleworth, W. J., and Wallace, J. S.: Evaporation from sparse crops-an energy combination
theory, Q. J. Roy. Meteor. Soc., 111, 839-855, 1985.
Siles, P.: Hydrological processes (water use and balance) in a coffee (Coffea arabica L.)
monoculture and a coffee plantation shaded by Inga densiflora in Costa Rica, PhD Thesis,
Université Henri Poincaré, Nancy-I, Nancy, France, 153 pp., 2007.
Siles, P., Harmand, J.-M., and Vaast, P.: Effects of Inga densiflora on the microclimate of
coffee Coffea arabica L.) and overall biomass under optimal growing conditions in Costa
Rica, Agroforest. Syst., 78, 269-286, 2010a.
Siles, P., Vaast, P., Dreyer, E., and Harmand, J.-M.: Rainfall partitioning into throughfall,
stemflow and interception loss in a coffee (Coffea arabica L.) monoculture compared to an
agroforestry system with Inga densiflora, J. Hydrol., 395, 39-48, doi:
10.1016/j.jhydrol.2010.10.005, 2010b.
126
Singh, V. P.: Computer Models of Watershed Hydrology, First Edition ed., Water Resources
Publications, Colorado, USA, 1995.
USDA: Soil Taxonomy, A Basic System of Soil Classification for Making and Interpreting
Soil Surveys, Agriculture Handbook Number 436, Washington, USA, 1999.
Vaast, P., Beer, J., Harvey, C., and Harmand, J. M.: Environmental services of coffee
agroforestry systems in Central America: a promising potential to improve the livelihoods of
coffee farmers’ communities, in: Intregrated Management of Environmental Services in
Human-Dominated Tropical Landscapes, IV Henri A. Wallace Inter-American Scientific
Conference Series, CATIE, Turrialba, Costa Rica, 2005.
van Kanten, R., and Vaast, P.: Transpiration of Arabica Coffee and Associated Shade Tree
Species in Sub-optimal, Low-altitude Conditions of Costa Rica, Agroforest. Syst., 67, 187-
202, 2006.
Vega, F. E., and Rosenquist, E.: The coffee berry borer and coffee research at the United
States Department of Agriculture, Proc. First ICO World Coffee Conference, London, UK,
2001.
Wahl, T. L., Clemmens, A. J., Replogle, J. A., and Bos, M. G.: WinFlume - Windows-based
software for the design of long-throated measuring flumes, Fourth Decennial National
Irrigation Symposium, Phoenix, Arizona, 2000.
White, M. A., Thornton, P. E., Running, S. W., and Nemani, R. R.: Parameterization and
Sensitivity Analysis of the BIOME-BGC Terrestrial Ecosystem Model: Net Primary
Production Controls, Earth Interact., 4, 1-85, doi:10.1175/1087-
3562(2000)004<0003:PASAOT>2.0.CO;2, 2000.
Wilson, K. B., Hanson, P. J., Mulholland, P. J., Baldocchi, D. D., and Wullschleger, S. D.: A
comparison of methods for determining forest evapotranspiration and its components: sap-
flow, soil water budget, eddy covariance and catchment water balance, Agr. Forest Meteorol.,
106, 153-168, 2001.
Xu, C. Y., and Singh, V. P.: A Review on Monthly Water Balance Models for Water
Resources Investigations, Water Resour. Manag., 12, 31-50, 1998.
127
Zaehle, S., Sitch, S., Smith, B., and Hatterman, F.: Effects of parameter uncertainties on the
modeling of terrestrial biosphere dynamics, Global Biogeochem. Cy., 19, GB3020,
doi:dx.doi.org/10.1029/2004GB002395, 2005.
Ziegler, A. D., Giambelluca, T. W., Sutherland, R. A., Nullet, M. A., Yarnasarn, S., Pinthong,
J., Preechapanya, P., and Jaiaree, S.: Toward understanding the cumulative impacts of roads
in upland agricultural watersheds of northern Thailand, Agr. Ecosyst. Environ., 104, 145-158,
2004.
128
129
CHAPTER 6
Water and sediment yield in a coffee agroforestry system at
various spatio!temporal scales: from plot to basin and from
event to annual scale14
!"!#$%&'()*!+,-!./0"& !"!#!$%&'(')*$+,-./0*1 !2!$+,3)*$4,-//05$
$67+89!$:4+$;<,=>,&/!$?@25ABA$4,CD.)&&')*!$?*0C<)$
"$76;!$6;EF;!$ AA2"$>0C$G,/H!$6,/D0$+'<0$
2$68I7;!$J JA!$2AKA $I-**'0&L0!$6,/D0$+'<0$
5$7E+8!$:4+$M7>8N!$"$F&0<)$O'0&0!$25ABA$4,CD.)&&')*!$?*0C<)$
#6,**)/.,C1'C3$0-DP,*Q$
)@R0'&Q$?S,R)T9U'<)V3,V<*$
I)&).P,C)Q$WKABX$"K$KY$"K$ZJ$
?0[Q$WKABX$"K$KY$"A$5K$
Abstract
In many studies aiming at assessing the surface runoff and sediment yield in agricultural
basins with little infrastructure area (like roads), the initial research assumptions are that basin
sediment yield originates mostly from plot erosion, that plots with trees (agroforestry) erode
less than monoculture, and that most of basin sediment is produced during storm events. A
quantification of elemental hydrological and soil loss processes in a coffee agroforestry
system on andisoils of Costa Rica is presented here, aiming to document (1) the plot erosion
in comparison with the sediment yield at a basin scale, (2) the effects of trees on plot erosion,
and (3) the complex sources of sediment in the basin and their occurrence according to
rainfall events or recession periods. The study intends to verify the initial assumptions and
14 Paper in preparation for Agriculture, Ecosystems & Environment:
Gómez-Delgado, F., Roupsard, O., Moussa, R., 2011. Water and sediment yield in a coffee agroforestry system
at various spatio-temporal scales: from plot to basin and from event to annual scale. Agriculture, Ecosystems &
Environment, in prep.
130
assess the performance of these agro-ecosystems in terms of sediment control, as eventual
providers of Hydrological Environmental Services.
We continuously measured water and sediment yield during one year in an experimental
coffee agroforestry (CAF) micro-basin (1 km2), and in two contrasting coffee plots (shaded,
non-shaded) located within this basin. The streamflow and turbidity at the outlet of the basin
were recorded by a flume and a turbidimeter, respectively. In the plots, surface runoff was
measured with tipping buckets and soil erosion by manual collection. An uncertainty analysis
was developed to account for estimation errors of short- and long-term basin sediment yields.
The main findings were: (1) basin sediment yield of 1.0 t ha-1
y-1
, of which only 1/3 was
produced during storms events, and 2/3 during recessions; (2) plot erosion contributing to
only 16% of storm sediment yields, i.e. only 5% of total annual sediment yield (95% of the
basin area is covered by plots, 5% by roads); (3) plot storm runoff supplying only 4% of basin
surface runoff, and estimated runoff on roads explaining the remainder.
These findings confirm low runoff/erosion on CAF basins on andisoils, mostly coming from
recession times, and the huge importance of roads and river bed/banks on runoff and sediment
generation during storms.
Key words: Erosion, Sediment yield, Coffee, Agroforestry, Hydrology, Hydrological
processes, Streamflow, Surface runoff, Turbidity, Plot, Basin, Scale effects
6.1 Introduction
6.1.1 Hydrological Environmental Services
Water and sediment yield from tropical lands has been reported in many studies, widely
summarized by authors like Bruijnzeel (2004) or Aylward (2004). However, there is a dearth
of information on the effects of land rehabilitation on basin sediment yield, be it through
physical soil conservation schemes, the introduction of agroforestry-based cropping systems
or full-scale reforestation (Bruijnzeel, 2004). In Costa Rica, some of these practices are being
promoted by hydropower producers through the payment of Hydrological Environmental
Services HES (Pagiola, 2008; Wunder et al., 2008), or in the frame of basin management
131
programs (Jaubert, 2001). Coffee is one of the most traded agricultural commodities in the
world employing 100 million people (Vega and Rosenquist, 2001). In Costa Rica, coffee
accounted for 15% of the agricultural exports in 2008 and covered 2% of the territory
(SEPSA, 2009), being present in the main basins used for hydropower generation over the
country, and making evident the eventual trade-offs of the payment of HES from hydropower
producers to coffee farmers.
6.1.2 Spatial scales from the plot to the basin
While the scale of interest for hydropower producers is the basin, any management practice in
the coffee plantations must be implemented at the plot scale, making the scaling up an
essential step. These management practices, addressing to lower suspended sediment
concentrations in dam-reservoir tributaries, reduce the operational costs linked to cleaning the
useful storage volume of the reservoirs. However, monitoring programs to assess the
effectiveness of the implemented practices are not well documented in Costa Rica and, like in
other tropical regions, are mainly focused on the plot scale (Verbist et al., 2010), thus
carrying to scrutiny for not delivering the desired reductions in downstream sediment rates
(Rijsdijk et al., 2007a).
The scaling up of plot results to the basin scale is not straightforward, given that basin
sediment yield is the product of all generating and transporting processes within a basin (de
Vente and Poesen, 2005). Scaling sediment fluxes has been recognized as a difficult task by
various researchers (Afandi et al., 2002b; Sidle et al., 2006; Verbist et al., 2010), since factors
as the dominant geomorphic processes controlling sediment yield, the dependency on the
spatial distribution of sediment sources and sinks, and the influence of point sediment sources
such as runoff-generating unpaved roads (Vanacker et al., 2007) can produce a wide
variability in the overall yield response.
6.1.3 Temporal scales from the event to the whole year
Studies comparing the sediment yield at these two scales (Brasington and Richards, 2000;
Rijsdijk et al., 2007a; Vanacker et al., 2007; Raclot et al., 2009; Verbist et al., 2010) had
greater explanation capacity than those working at the plot or the basin scale only, with regard
to the location of the main erosion sources and the transport mechanisms turning the erosion
into river sediment yield. But only the first of the above listed references reported continuous
132
turbidity recording at the basin outlets, which has performed much better to estimate either
suspended sediment concentrations (S) during storm events (Walling and Webb, 1982;
Gippel, 1989; Lenzi and Marchi, 2000; Sun et al., 2001; Verbist et al., 2003; Seeger et al.,
2004; Lana-Renault et al., 2007; Minella et al., 2008) or suspended sediment yields (Y) for
longer time spans (Walling, 1977; Jansson, 1992; Wass and Leeks, 1999; Jansson and
Erlingsson, 2000; Schoellhamer and Wright, 2003; García-Ruiz et al., 2008). A continuous
turbidity recording is also useful in streams with rapidly varying sediment concentrations
(Lewis, 1996), for the assessment of sedimentological variations due to wildfires (Lane et al.,
2006), in barren regions or agricultural lands with very concentrated sediment transport
during storm events (Dabney et al., 2006; Downing, 2006), or for estimating loads of
transported substances, such as nutrients, heavy metals and radionuclides (Walling et al.,
1992).
6.1.4 Uncertainty in sediment estimations
Another concern in the quantification of basin sediment yield is the uncertainty in the
measurement and estimation methods, which has been approached by researchers using
different techniques. Turbidity ( ) measurement errors due to particle size, shape, organic
matter and the like are well recognized by sensor manufacturers and commonly found in the
literature (Foster et al., 1992; Clifford et al., 1995; Gippel, 1995; Downing, 2006). The
attribution of total errors to the prediction by regression methods (like the regression of
streamflow against sediment concentration, sensor output against sediment concentration, or
simply against sediment concentration) is, to our knowledge, the most used technique
(Gilroy et al., 1990; Lewis, 1996; Wass et al., 1997; Brasington and Richards, 2000;
Schoellhamer and Wright, 2003; Lane et al., 2006). Other authors assessed either the relative
error in total sediment loads obtained by different rating methods, with respect to a reference
one (Walling, 1977), or the difference between total loads calculated with vs S rating curves
and loads from detailed sediment sampling or from different sub-sampling schemes (Lewis,
1996; Sun et al., 2001). However, fewer studies expounded other types of uncertainties, like
those attributed to the instrumentation (measurement error), random measurement errors, the
point sampling type represented by turbidimeters, or the estimation of streamflow by rating
curves on water stage height (Wass and Leeks, 1999; Jansson and Erlingsson, 2000).
133
6.1.5 Objectives and methodology
This study aims to estimate the water and sediment yield in different spatial and temporal
scales: from plot to basin and from storm events to annual, in a coffee agroforestry system of
Costa Rica. The approach addresses to find out where is the sediment coming from
(shaded/non-shaded plots, roads, stream channel), and when (storm events, recession periods).
In order to achieve these objectives, we set up an agroforestry experimental micro-basin (1
km2) in Costa Rica to continuously measure the main water and sediment fluxes at plot and
basin scales. Different estimation uncertainties (i.e., random, systematic and analytic) were
taken into account by following standard rules for its determination and propagation (JCGM,
2008). To study the spatial source of sediment and the potential effects of shade trees we
measured separately the surface runoff and the erosion in coffee agroforestry and coffee
monoculture plots. If basin sediment yield is not entirely coming from plot erosion, other
sources like roads or stream channels (bed, banks) might be contributing as well. To
understand when the sediment is being yielded, the common assumption that most of it is
produced only during storm events (Webb and Walling, 1982; Chappell et al., 1999; Lewis,
2003) was verified by comparing the water and sediment yields during storm events (in which
surface runoff may occur on plots) with those occurring during flow recessions (in which
basin outflow is only due to baseflow). Then, our research hypotheses are that (1) basin
sediment yield originates from plot erosion, (2) agroforestry coffee plots erode less than
monoculture coffee plots and (3) most of basin sediment is produced during storm events.
The paper was structured as follows: first we introduce the study site, the experimental design
and the calculation methodology. Second, we present the results of the vs S calibration.
Third, we report the water and sediment yields during storm events for both plot and basin
scales, in comparison with the annual water and sediment yields. Fourth, we attempt to
upscale these results from storm to annual scale and from plot to basin scale. Finally, we
discuss the proposed scaling analyses in order to identify the sources of basin sediment, the
main transporting mechanisms and the uncertainties in our estimations.
134
6.2 Study site: location, soil and climate
The area of interest is located in Reventazón river basin, in the Central-Caribbean region of
Costa Rica (Fig. 19a,b). It lies on the slope of the Turrialba volcano (central volcanic
mountain range of the country) and drains to the Caribbean Sea. The Aquiares coffee farm is
one of the largest in Costa Rica (6.6 km2), “Rainforest Alliance TM” certified, 15 km from
CATIE (Centro Agronómico Tropical de Investigación y Enseñanza). Within the Aquiares
farm, we selected the Mejías creek micro-basin (Fig. 19c) for the “Coffee-Flux” experiment.
The basin is placed between the coordinates -83º44’39” and -83º43’35” (West longitude), and
between 9º56’8” and 9º56’35” (North latitude) and is homogeneously planted with coffee
(Coffea arabica L., var Caturra) on bare soil, shaded by free-growing tall Erythrina
poeppigiana trees. The initial planting density for coffee was 6,300 plants ha-1
, with a current
age >30 years, 20% canopy openness and 2.5 m canopy height. It is intensively managed and
selectively pruned (20% per year, around March). Shade trees have a density of 12.8 trees ha-
1, with 12.3% canopy cover and 20 m canopy height. The experimental basin has an area of
0.9 km2, an elevation range from 1,020 up to 1,280 m.a.s.l. and a mean slope of 20%.
Permanent streams extend along 5.6 km, implying a drainage density of 6.2 km km-2
. The
average slope of the main stream is 11%.
According to the classification by Mora-Chinchilla (2000), the experimental basin is located
along a 1.3 km wide strip of volcanic avalanche deposits, characterized by chaotic deposits of
blocks immersed in a matrix of medium-to-coarse sand, which is the product of the collapse
of the south-eastern slope of Turrialba volcano’s ancient crater. The general classification
given by the geological map of Costa Rica (MINAE-RECOPE, 1991) describes the general
stratigraphy as shallow intrusive volcanic rocks, and the particular region as proximal facies
of modern volcanic rocks (Quaternary), with presence of lava flows, agglomerates, lahars and
ashes. Soils belong to the order of andisols according to the USDA soil taxonomy, which are
soils developing from volcanic ejecta, under weathering and mineral transformation
processes, very stable, with high organic matter content and biological activity and very large
infiltration capacities.
The climate is tropical humid with no dry season and strongly influenced by the climatic
conditions in the Caribbean hillside, according to Köppen-Geiger classification (Peel et al.,
2007). The mean annual rainfall in the study region for the period 1973-2009 was estimated
135
as 3014 mm at the Aquiares farm station. At the experimental basin the rainfall in the one-
year measurement period (2849 mm) was close to the annual mean, but showed a monthly
deviation of ±100 mm around the historical regime. Mean monthly net radiation ranged in
2009 from 5.7 to 13.0 MJ m-2
d-1
, air temperature from 17.0 to 20.8 °C, relative humidity
from 83 to 91%, windspeed at 2 m high from 0.4 to 1.6 m s-1
and potential evapotranspiration
(Allen et al., 1998) from 1.7 to 3.8 mm d-1
.
6.3 Materials and methods
6.3.1 Experimental setup
The “Coffee-Flux” experimental basin and instrument layout was designed to trace the main
water balance components and sediment yield, employing spatially representative methods
(Fig. 19c). It is part of the FLUXNET network for the monitoring of greenhouse gases of
terrestrial ecosystems. The experimental setup has been described in more details in Gómez-
Delgado et al. (2011a).
A. Rainfall (R) and climate
Rainfall was monitored at 3 m above ground in the middle of 3 profiles of the basin, using
three lab-calibrated ARG100 tipping-bucket (R.M. Young, MI, USA) connected to CR800
dataloggers (Campbell Scientific, Shepshed, UK), and integrated every 10 min. Other climate
variables were logged on top of an eddy-flux tower with a CR1000, every 30 s, integrated
half-hourly and using: Net radiation: NR-Lite (Kipp & Zonen, Delft, The Netherlands);
PPFD: Sunshine sensor BF3 (Delta-T devices Ltd, U.K.); temperature and humidity:
HMP45C in URS1 shelter (Campbell Scientific); wind-speed and direction: 03001 Wind
Sentry (R.M. Young, MI, USA). The theoretical evapotranspiration from a wet grass placed
under local climate conditions, ET0, was computed in accordance with Allen et al. (1998).
B. Streamflow (Q)
A long-throated steel flume (length: 3.9 m; width: 2.8 m; height: 1.2 m) was home-built to
measure the streamflow at the outlet of the experimental basin, with recording capacity up to
3 m3 s
-1, the maximum estimated discharge for the study period from an intensity-duration-
frequency analysis. The flume was equipped with a PDCR-1830 pressure transducer
(Campbell Scientific) to record water head at gauge point (20 s, 10 and 30 min integration),
136
while the rating curve was calculated considering the geometric and hydraulic properties of
the flume using WinFlume software (Wahl et al., 2000). A validation of the rating curve was
made successfully using the salt dilution method as well as a pygmy current meter.
C. Soil water content
A frequency-domain-reflectometry portable probe (FDR Diviner2000, Sentek Pty Ltd) was
used to survey 20 access tubes distributed in the three study profiles to provide the mean
volumetric soil water in the basin. The sensor measures at 10 cm intervals, reaching a total
depth of 1.6 m. A measurement campaign through the 20 sites was carried out every week.
The sensors were calibrated by digging sampling pits in the vicinity of six test tubes, to obtain
the actual volumetric soil water content from gravimetric content and dry bulk density.
D. Turbidity (!)
An optical backscatter sensor (OBS) or turbidimeter OBS-3 (Campbell Scientific) was
deployed in the flume at the outlet of the basin, allowing to measure water every 20 s and
integrate it every 10 and 30 min. NTU range: 0-250 (low) and 0-1000 (high range). The
sensor window was placed facing the main stream and protected with a grille properly
adapted to the source beam and detector acceptance cone.
E. Plot surface runoff (SRplot)
Two experimental plots of 1000 m2 each (14 m by 72 m, approximately) were placed
following the slope direction within the experimental basin. Both have a slope of 20%. One of
the plots displays an agroforestry scheme (AF), coffee plants under E. poeppigiana shade
trees, while the other plot displays coffee in monoculture (MC). They are limited in the upper
side by the basin water divider, while their lateral boundaries were isolated by 20 cm tall
barriers. PVC gutter collectors were installed at the lower end of each plot to trap coarse
sediment and route runoff to two measuring tanks equipped with tipping buckets (1000-20,
Sierra Misco, CA, USA). The soil surface within 30 cm of the gutters was compacted and
sealed with a plastic sheet.
F. Plot erosion
The erosion at the plot scale was measured by collection, drying and weighting of the coarse
sediment accumulated in the gutters at the end of the experimental plots, along with the finer
137
particles trapped by plastic filters located just before the entrance of runoff into the measuring
tipping buckets.
6.3.2 Measurement period, missing values and gap!filling
For the sake of having one complete year of observations, our measurement period was
chosen from 07/03/2009 to 06/03/2010. The frequency of measurement depends on the time
scale of the analysis: for an annual based analysis we worked at the 30 min time step ( t),
while for storm events we chose the 10 min t.
The procedure for the in situ recording of by OBSs presents some relevant shortcomings,
especially those linked to chemical and biological fouling in some streams and brackish water
environments (Downing, 2006). Those factors caused the loss of 19% of the measurements
during our measurement period. When gaps were present in the record of , two different gap
filling methods were applied depending on the streamflow state. First, a linear regression
explaining as a function of streamflow (R2=0.62), fitted for the highest range of streamflow
(Q ! 0.15 m3 s
-1), was used to predict peak values of during three storm events (18/07/2009,
19/02/2010 and 25/02/2010). Second, as no clear relationship was found between streamflow
and for lower streamflow ranges, six different techniques were tested to replace the missing
values of in those ranges by: (1) the series mean, (2) the mean of surrounding points, (3) the
median of surrounding points, (4) a linear interpolation on surrounding points, (5) the linear
trend at each missing point and (6) a log-linear random simulation. Two evaluation criteria
were used to select the second technique as the most appropriate: (1) the Euclidean relative
distance to the mean and standard deviation of the original series and (2) a visual inspection
of the resulting gap-filled series. Table 6 summarizes the performances of the methods to fill
the missing values. After using each of these six techniques to estimate the annual sediment
yield, differences between -4% and 4% were found with respect to the chosen method (mean
of surrounding points).
6.3.3 Estimation of basin sediment yield and its uncertainty
The estimation of sediment loads being transported out of the basin through the Mejías creek,
was based on continuous t and Qt measurements, at time t. Qt was obtained from stage water
heights ht at the flume site, then evaluated in the rating curve of the flume to estimate Qt. A
138
calibration procedure was carried out to link t with suspended sediment concentration at time
t (St).
Table 6: Performance of the six gap-filling methods applied to the missing values of the original
turbidity record, with respect to two evaluation criteria: ED and VI.
Method ED (%) VI Y (t) Y (%)
(1) Series mean 11% fair 93.5 1%
(2) Mean of surrounding points 7% good 92.5 -
(3) Median of surrounding points 8% good 91.7 -1%
(4) Linear interpolation on surrounding points 10% poor 88.6 -4%
(5) Linear trend at each missing point 11% fair 96.5 4%
(6) Log-linear random simulation 5% poor 93.2 1%
ED: Euclidean relative distance to mean and standard deviation of the original turbidity series
VI: Visual inspection of the gap-filled turbidity series
Y: Total sediment yield according to each method
Y: Percentage difference in sediment yield with respect to the chosen method
*Bold entries highlight the two methods with best performance for each criterion: ED or VI
Fifty five water samples were collected at the flume site under different flow conditions
throughout the one-year measurement period, and then analyzed by gravimetric standard
methods in the laboratory, to obtain the respective Ssample values. The site-recorded sample
during sampling times were used to estimate the relationship vs S through both, linear and
power statistical regressions, whose goodness of fit and uncertainty functions around the
predicted values were compared. Figure 37 displays two regression curves that were fitted to
the set of water samples in order to predict the continuous St in the stream from t
measurements. The curve (a) is a power fit (R2=0.64), while curve (b) is a linear fit (R
2=0.48).
Although the 95% prediction interval (defined by the two discontinuous thin lines in each
graph) for the power fit looks very large for high values, and it is asymmetric (this latter
property making more difficult the analysis of uncertainty propagation), it has a much higher
determination coefficient (R2) than the linear fit and, most importantly, avoids negative values
for the confidence limits. Such limits (at the 95% level of confidence) enclose all sample
observations, which is not the case of the equivalent prediction interval for the lineal curve,
which excludes two samples in the high range of . On the other hand, the linear fit gives
lower estimation uncertainties for higher , but poses a problem on the much lower predictive
capacity for the mean response of suspended sediment concentrations.
139
S = 3.519 0.882
R2 = 0.638
0
500
1000
1500
0 100 200 300 400 500 (NTU)
S (mg
l!1)
S = 1.796
R2 = 0.477
0
500
1000
1500
0 100 200 300 400 500
(NTU)
S (mg
l!1)
Figure 37. Relationships between turbidity ( ) and suspended sediment concentration
(S), fitted by a) power and b) linear curves and denoted by the thick line in each graph.
Thin lines represent the 95% confidence bands about the regression line (dotted) and
for individual predictions (discontinuous). R2: determination coefficient.
Lewis (2003) found that power regression was less prone to extrapolation errors than
polynomial curves, and indicates two more additional advantages of this type of curves over
linear models: (1) predictions are always positive, and (2) residual variance is often less
dependent on turbidity. This last feature can improve estimation of the variance of the
sediment load. Therefore, we chose the most conservative alternative (the power regression),
in terms of a higher R2 and wider ranges of estimation uncertainty.
Then total sediment yield YT [M] over a period T [T], with time step t [T], was calculated by
integrating the product of Qt [L3 T
-1] and St [M L
-3] over time:
tSQYT
t
ttT !"#"1
(1)
a)
b)
140
Appendix A provides a detailed description of the calculation procedure of each of the terms:
ht, Qt, t, St and Yt, all at the time step t, and of C (the annual sediment yield over the
measurement period), along with their respective uncertainties, which were obtained
according to the “Guide to the expression of uncertainty in measurement” (GUM) by the Joint
Committee for Guides in Metrology (JCGM, 2008). Table 7 defines functional relationships f
between input and output variables, classifies the different types of uncertainty as (A) random
variability of individual measurements, (B) systematic error in the measurements or (C) errors
from analytical methods, and presents the equations or method to obtain it. Then, the
propagation of the standard uncertainties through the different steps and equations, noted u(·),
was combined in accordance to the law of propagation of uncertainty, to finally produce
standard uncertainties for the estimations of ht and Qt (water components) and of t, St, Yt and
C (sediment components). Such standard uncertainties can be used to calculate 95%
confidence intervals around the estimated time series.
6.3.4 Definition of storm events, baseflow separation and estimation of surface
runoff from roads
In the measurement period, a storm event was arbitrarily defined by an increase of 100% in
basin streamflow within a lag span of 1 h, i.e., by comparing streamflow at a given time with
its 6 previous values (the t is 10 min). Peakflows with values below a threshold of 0.2 m3 s
-1
were discarded for being considered small events, while other three events initially excluded
by the algorithm but with peakflows greater than 0.2 m3 s
-1 were included in the dataset. The
beginning of storm events was defined by three consecutive ts with R $ 0.2 mm just before
the main rainfall appears, while the end was determined by either the same condition (but just
after the rainfall event) or by three ts under a fully-baseflow recession, whichever condition
is met later in the time sequence.
The storm event hydrograph provides information of the basin response to a rainfall storm
event. Whereas during recession periods only baseflow is running in perennial streams (i.e.,
aquifer spring waters), storm periods include also the surface runoff component. To study the
basin sediment yield it is desirable to understand the behaviour of suspended sediment
concentrations over time and flow state, including the relative contribution by base and storm
flows.
141
Tab
le 7
. C
om
po
nen
ts o
f u
nce
rtai
nty
in
the
esti
mat
ion
of
ann
ual
cum
ula
tiv
e se
dim
ent
yie
ld a
nd
eq
uat
ion
s.
Mes
ura
nd
F
un
ctio
na
l
rela
tio
nsh
ip
f
(A)
Ra
nd
om
sta
nd
ard
un
cert
ain
ty
(B)
Sy
stem
ati
c
sta
nd
ard
un
cert
ain
ty
(C)
An
aly
tica
l
Sta
nd
ard
un
cert
ain
ty
Co
mb
ined
sta
nd
ard
un
cert
ain
ty
ht:
hei
gh
t o
f
wat
er a
t fl
um
e
at t
ime
t (m
)
- %
&%
&n
hs
hu
tA
t"
,
%
&3
,h
hu
Bt
!"
-
%&
%&
%&
B,
2
A,
22
tt
th
uh
uh
u'
"
Qt: st
ream
flo
w
at t
ime
t
(m3 s
-1)
Qt =
f Q
(h
t)
(rat
ing
cu
rve)
u(Q
t,A)
and
u
(Qt,
B):
p
ropag
atio
n
of
u(h
t,A)
an
d
u(h
t,B)
thro
ugh
rat
ing
cu
rve
eq.
(k:
un
cert
ain
ty t
yp
e, h
ere
A o
r B
):
%&
%&
kt
Q
kt
hu
hfQ
u,
,((
"
u(Q
t,C):
an
alyti
cal
un
cert
ain
ty
asso
ciat
ed
wit
h
the
rati
ng
cu
rve
eqn
: se
e
%&
%&
%&
B,
2
CA,
22
tt
tQ
uQ
uQ
u'
"'
(*
)
t:
turb
idit
y
at
tim
e t
(NT
U)
- %
&n
su
tA
t
"
)(
,
%&
3,
!"
Bt
u
- %
&%
&%
&B,
2
A,
22
tt
tu
uu
'
"
St:
susp
end
ed
sed
imen
t
con
cen
trat
ion
at t
ime
t (m
g l-1
)
St =
f S ( t
) (p
ower
re
gres
sion
)
u(S
t,A+
B):
pro
paga
tion
of
u(
t) (i
nclu
ding
unc
erta
inti
es t
ype
A a
nd B
) th
roug
h po
wer
reg
ress
ion
equa
tion
:
!
!
tS
tu
fS
u"
"##$
%B
A,
u(S
t,C):
pr
edic
tion
un
cert
aint
y of
the
pow
er
regr
essi
on,
whi
ch
is
a fu
ncti
on o
f th
e co
rrec
ted
RM
SE
: see
!
!
! &
%&
%$
C,2
BA,
22
tt
tS
uS
uS
u
(**)
Yt:
sedi
men
t yi
eld
at t
ime
t (k
g)
Yt =
f Y (
Qt,
St)
(ana
lyti
cal
equa
tion
) -
- -
!
!
!
!
!
!
(**)
2
22
2
tt
tt
tt
tt
tt
tt
SSu
QQu
S,Q
r
SSu
QQu
YYu
&
&&
%
' ()* +,
%' ()
* +,$
' ()* +,
C:
annu
al
sedi
men
t yi
eld
(kg)
C=
f C(Y
1,…
,YT)
(ana
lyti
cal
equa
tion
) -
- -
!
!
!
!
! &
$&
-
$&
&
..%
%$
T tt
tt
T tt
Yu
Yu
Y
Yu
Cu
21
1
1
22 2/
(*
**)
* E
rror
s A
and
C w
ere
calc
ulat
ed b
y th
e W
inF
lum
e so
ftw
are
as ±
3.6%
of
Qt,
at 9
5% le
vel o
f co
nfid
ence
. **
Ana
lytic
al u
ncer
tain
ty o
f S
t is
asym
met
ric
from
the
err
ors
of t
he p
ower
reg
ress
ion,
so
u(S
t) an
d u
(Yt)
are
calc
ulat
ed s
epar
atel
y fo
r up
per
and
low
er u
ncer
tain
ties.
r(Q
t, S
t) is
th
e co
rrel
atio
n be
twee
n Q
t and
St.
***
We
assu
med
tha
t Y
t is
an a
utor
egre
ssiv
e pr
oces
s of
ord
er o
ne. A
s a
resu
lt fr
om u
2 (St),
u(Y
t) is
asy
mm
etri
c, s
o th
e pr
opos
ed e
quat
ion
was
app
lied
sepa
rate
ly f
or u
pper
and
lo
wer
unc
erta
intie
s. !
1(Y
t) is
the
firs
t aut
ocor
rela
tion
of th
e ti
me
seri
es Y
t.
142
The separation of these two components can be done from the basin hydrograph using several
methods. In this study we chose a simple graphical procedure known as the constant slope
method (McCuen, 2005), which consists in connecting the start of the rising limb with the
inflection point on the recession limb of the hydrograph, hence assuming an instant response
of baseflow to the rainfall event. The inflection point was determined by visual inspection,
plotting the streamflow in a log-scale and finding the change in the recession slope for each
storm event. Then, the component above the line is surface runoff and that below is baseflow.
In order to give some insights on the source of surface runoff in the basin (SRb), an exercise
was performed by assuming that a fraction of the total rainfall falling on the roads of this
basin was running-off and being conducted through the ditches and gutters on the sides of the
roads to the basin outlet (Fig. 19c). This fraction was estimated as 65% by Rijsdijk et al.
(2007b) and above 80% by Ziegler et al. (2004) for unpaved roads in Indonesia and Thailand,
respectively. In this study we assumed a fraction of 80% to estimate this so-called potential
surface runoff from roads (SRr).
6.4 Results
The experimental results will be presented in three parts. First we introduce the water and
sediment dynamics at the storm-event time scale for both, the plot and the basin spatial scales.
Second we integrate these results for the one-year measurement period, to obtain the annual
water balance and sediment yield for the plots and the basin. Third, we compare the results at
temporal (storm events to annual) and spatial (plots to basin) scales.
6.4.1 Water and sediment yields during storm events
This subsection consists of three parts, (1) the analysis of water yield for a single storm event,
which pretends to explain the employed terms and analyses, (2) an overview or the main
properties of water yield over the selected storm events, and (3) the erosion/sediment yield
during individual and the entire set of storm events. Each of these sections presents the results
first at the basin scale, and then at plot scale.
A. Water yield during a single storm event
A detailed illustration of the hydrological measurements for a particular storm event
(05/06/2009) is presented at basin and plot scales in Fig. 38, for a t =10 min. The part (a)
143
shows the rainfall hyetograph of the storm with centroid at 16:10, while part (b) displays the
streamflow hydrograph with peakflow at 16:40, the baseflow (separated of total streamflow
according to the constant slope method), and the suspended sediment concentration. The
water balance of this event, at the basin scale, can be summarized as event R = 61 mm with
maximum intensity in 10 min (I10) equal to 47.3 mm h-1, producing an event streamflow depth
of Qd = 4.4 mm (7.2% of R), with peakflow streamflow rate (Qpr) of 2.5 mm h-1, occurring at
the time of concentration tc = 30 min. The duration of the event was 5.67 h, while surface
runoff in the basin (SRb) equalled 1.7 mm (3% of R) and was observed during a 1.67 h span.
Figure 38c presents the measurements of surface runoff in the two runoff plots: one under a
coffee agroforestry cover (AF), the other under coffee in monoculture (MC). The total surface
runoff during the storm event was SRAF =17x10-3 mm (agroforestry plot) and SRMC =20x10-3
mm (coffee monoculture plot), while peakflow rates were very similar in both plots (0.024
mm h-1, approximately).
B. Water yield during all the storm events
The proposed criterion allowed identifying 24 storm events, whose properties are presented in
Table 8. Event durations ranged from 3 to 12 h, event rainfall totals from 11 to 88 mm,
maximum I10 from 10 to 81 mm h-1 and Qpr reached values up to 5 mm h-1. The ratio between
baseflow and SRb was usually greater than one, indicating that even during storm events,
baseflow contributes more to streamflow (three times, in average) than SRb. This latter,
displayed as runoff coefficient (percentage of the event rainfall) presented a maximum value
around 7%, but could be as low as 0.7%. The potential surface runoff from roads (SRr) is
reported it in Table 8. For 71% of the storm events this quantity was large enough as to
account for all SRb, which is highlighted by the entry “>100%” in the percentage column (i.e.,
SRr as percentage of SRb). In agreement with the previous results, very low runoffs (in
comparison with SRb or SRr) were measured at the plot scale, with maximum depths around
5x10-2 mm per event in either of the two plots, and cumulative values over the 24 events equal
to 0.25 mm and 0.44 mm at the AF and MC plots, respectively. A correlation analysis
including all comparable variables in Table 8 only identified some significant relationships
between these plot scale measurements and the basin Qpr (Fig. 39). Then, the Qpr of a storm
event could be explained by SRAF with a much higher prediction capacity (R2=0.81) than by
SRMC (R2=0.52). The positive interceptions in both curves suggest that, even under very low
144
or null plot runoff conditions, peakflows in streamflow may exist (probably, as response of
non-plot rainfall interception and runoff).
0
2
4
6
8
10
14:30 15:30 16:30 17:30 18:30 19:30
R (mm
!t"1)
0.0
0.1
0.2
0.3
0.4
0.5
14:30 15:30 16:30 17:30 18:30 19:30
Q, B
(mm
!t"1)
0
250
500
750
1000
1250
S (mg
l"1)B
QS
0
0.001
0.002
0.003
0.004
0.005
14:30 15:30 16:30 17:30 18:30 19:30
Time (05/06/2009)
SR
(mm
!t"1)
SR AFSR MC
Figure 38 Water balance and sediment yield at basin and plot scales, for the storm event of 05/06/2009: a) rainfall R, b) streamflow Q, its baseflow B and suspended sediment concentration S and c) surface runoff SR for agroforestry AF and monoculture MC plots. t = 10 min.
a)
b)
c)
14
5
Tab
le 8
: H
ydro
logi
cal
and
sedi
men
t pr
oper
ties
of
the
24 s
torm
eve
nts
reco
rded
dur
ing
one
year
of
obse
rvat
ions
at
the
expe
rim
enta
l ba
sin.
R:
rain
fall
, I 1
0:
max
imum
int
ensi
ty i
n 10
min
, Q
pr:
stre
amfl
ow p
eak
rate
, B
: ba
sefl
ow,
SR
b:
surf
ace
runo
ff a
t ba
sin
scal
e, S
Rr:
surf
ace
runo
ff f
rom
all
the
bas
in r
oads
, S
: su
spen
ded
sedi
men
t co
ncen
trat
ion,
Q:
stre
amfl
ow,
Y:
even
t se
dim
ent
yiel
d, S
R:
surf
ace
runo
ff,
AF
: ag
rofo
rest
ry p
lot
and
MC
: m
onoc
ultu
re p
lot.
All
val
ues
from
dir
ect m
easu
rem
ents
are
giv
en in
rom
an ty
pe, w
hile
cal
cula
ted
or e
stim
ated
val
ues
are
pres
ente
d in
ital
ic ty
pe.
SR
b
SR
r S
R (
x10
-3m
m)
Plo
t er
osi
on
(g
) E
ven
t
nu
mb
er
Date
Ev
ent
du
rati
on
(h
r)
Ev
ent
R
(mm
)
Max
I10
(mm
h-1
)
Qpr
(mm
h-1
)
B
(mm
)(m
m)
(% R
)(m
m)
(% S
Rb)
Pea
k l
ag
S :
Q (
min
) Y
(t)
AF
M
C
AF
M
C
1 05
/06/
2009
5.
7 61
.0
47.3
2.
5 2
.8
1.7
3
%
1.9
>
100
%
0 1
.61
17
20
25
17
5 2
06/0
6/20
09
5.3
31.9
28
.1
0.8
2.4
0
.3
1%
1
.0
>100
%
0 0
.23
6
7 -
- 3
13/0
6/20
09
5.7
67.2
64
.9
2.8
3.8
2
.0
3%
2
.1
>100
%
-30
1.1
8
19
23
105
100
4 19
/06/
2009
6.
3 44
.5
48.1
1.
5 1
.7
1.1
3
%
1.4
>
100
%
-10
0.4
0
8 12
23
0 25
4 5
24/0
6/20
09
5.5
44.2
39
.5
1.3
1.6
1
.9
4%
1
.4
75
%
-10
0.5
5
8 13
-
- 6
29/0
6/20
09
4.3
52.7
66
.1
1.5
1.5
1
.4
3%
1
.7
>100
%
0 0
.49
3
11
117
404
7 03
-04/
07/2
009
8.3
50.1
42
.2
2.0
3.2
1
.7
3%
1
.6
90
%
0 0
.82
10
23
79
82
8
17-1
8/07
/200
98.
3 78
.5
80.5
5.
1 6
.4
4.8
6
%
2.5
5
1%
+
10
3.9
9
53
56
231
752
9 26
-27/
07/2
009
5.0
29.0
50
.7
1.3
2.6
1
.0
3%
0
.9
95
%
-10
0.5
6
11
22
21
122
10
14/0
8/20
09
3.2
16.1
32
.3
0.9
1.1
0
.6
3%
0
.5
91
%
0 0
.17
5
13
- -
11
15/0
8/20
09
4.2
32.9
29
.1
1.2
1.8
0
.8
2%
1
.0
>100
%
0 0
.25
6
16
- -
12
17/0
8/20
09
4.0
23.6
29
.5
1.7
1.7
1
.1
5%
0
.7
66
%
0 0
.33
19
50
-
- 13
19
/08/
2009
3.
5 21
.3
43.9
0.
8 1
.6
0.2
1
%
0.7
>
100
%
-10
0.2
0
2 27
12
0 24
4 14
20
/08/
2009
3.
3 19
.2
22.7
1.
2 1
.8
0.6
3
%
0.6
9
4%
-1
0 0
.24
10
12
32
10
3 15
11
/09/
2009
3.
7 39
.9
70.2
1.
8 1
.1
1.4
3
%
1.3
9
1%
-1
0 0
.73
1
24
195
308
16
24/0
9/20
09
4.3
27.4
40
.1
0.5
0.8
0
.3
1%
0
.9
>100
%
-10
0.1
3
6 20
-
- 17
24
/09/
2009
3.
3 18
.2
34.1
0.
9 0
.8
0.5
3
%
0.6
>
100
%
0 0
.16
5
17
243
234
18
02/1
1/20
09
11.5
80
.6
54.5
2.
5 3
.9
3.6
4
%
2.5
7
0%
0
2.1
5
14
2
7
- -
19
22/1
1/20
09
6.0
17.3
20
.3
0.7
2.7
0
.2
1%
0
.5
>100
%
0 0
.34
3
6
-
- 20
22
/11/
2009
4.
2 11
.3
9.8
0.8
2.3
0
.3
3%
0
.4
>100
%
0 0
.24
2
5
33
6 71
7 21
24
/11/
2009
6.
2 23
.8
21.7
0.
8 2
.0
0.6
2
%
0.7
>
100
%
0 0
.32
4
8
70
22
9 22
14
/12/
2009
8.
3 37
.1
13.7
0.
9 4
.3
1.6
4
%
1.2
7
1%
-4
0 0
.75
6
1
3
536
505
23
19/0
2/20
10
12.2
22
.2
16.4
1.
3 6
.8
1.6
7
%
0.7
4
4%
0
1.1
5
12
2
2
599
807
24
25/0
2/20
10
7.7
87.7
39
.4
2.7
4.4
3
.4
4%
2
.7
81
%
0 2
.20
1
5
29
40
97
T
ota
l
93
8
6
3
33
19
.2
245
444
2979
51
33
*P
lot s
urfa
ce r
unof
f in
bol
d-it
alic
type
wer
e es
tim
ated
by
linea
r re
gres
sion
s fr
om th
e ev
ent t
otal
rai
nfal
l, w
ith
R2 =
0.6
4 fo
r A
F a
nd R
2 = 0
.80
for
MC
.
146
C. Sediment yield during storm events
For the example of the event of 05/06/2009 an assessment of its sediment yield was made. As
suggested by Fig. 38b, the raise in suspended sediment concentration is directly linked to the
presence of SRb in the basin. The maximum concentration reached almost 1000 mg l-1 and the
total sediment yield given by Eq. (1) was 1.61 t. During this event the erosion amounted to 25
g and 175 g in the AF and MC plots, respectively.
y = 0.084x + 0.695
R2 = 0.805
y = 0.054x + 0.465
R2 = 0.592
0
1
2
3
4
5
6
0 10 20 30 40 50 60 70 80Plot surface runoff (x10
"3mm)
Qp
r (mm
h"1) AF
MC
Figure 39: Streamflow peak rate (Qpr) for 24 storm events, explained by plot surface runoff under coffee agroforestry (AF) and coffee monoculture (MC) systems.
The repetition of this procedure for the other 23 storm events under analysis (Table 8) lead to
the general conclusion that the peak value of sediment concentration typically arrived at the
same time as that of streamflow, or some 10 min earlier. The basin sediment yield showed
widely variable results ranging from 0.1 t to almost 4 t per event, and totalizing 19 t for the 24
events through the measurement period. From the correlation analysis some significant
relationships were identified. First, some basic rainfall/runoff patterns were identified and
linked to the sediment yield. For instance, the formation of SRb as a quadratic function of the
respective event rainfall R (R2=0.80), which can be appreciated in Fig. 40 as a bubble graph,
displaying the event sediment load (Y) as the bubble width (this value was attached as a label
for the events with greatest sediment yields). This graph makes clear the threefold correlation
between these water-volume related variables, while an analogous relationship on water flow
intensities (maximum I10 of rainfall against Qpr) was not clear (R2=0.49). It must be also
147
pointed out that one single event produced a major (21%) contribution to the annual event-
based sediment yield, while the six greatest events accounted for 64%. Secondly, sound
quadratic relationships were found to explain the event sediment yield as a function of the
event rainfall (R2=0.73), Qpr (R2=0.88) and SRb depth (R2=0.93). The dependence of the
sediment yield on this latter variable is presented in Fig. 41 and was verified by comparing
the baseflow separation graphs of the 24 storm events with their sediment concentration
curves (data not shown). Additional relationships were tested using the basin-averaged water
content in the top 1.6 m soil layer at the beginning of each storm event, but none was
significant.
Bubble width: Y (t)
1.181.61
2.15 2.20
3.99
1.15
y = 0.0003x2 + 0.0191x
R2 = 0.8038
0
1
2
3
4
5
6
0 20 40 60 80 100
Event R (mm)
SR
b (mm)
Figure 40: Basin surface runoff (SRb) for 24 storm events, as function of event total rainfall (R), and their link to sediment yield (Y) as bubble width.
y = 0.089x2 + 0.364x
R2 = 0.934
0
1
2
3
4
5
0 1 2 3 4 5SR b (mm)
Y (t)
Figure 41. Basin sediment yield (Y) for 24 storm events, according to the basin surface runoff (SRb).
148
Finally, the soil erosion weekly collected at the two plots (AF and MC) was classified by
pick-up date and associated to some of these 24 storm events, as reported in Table 8. Plot
erosion was not correlated to any other variable in Table 8, not even to the surface runoff on
each corresponding plot. The highest erosion rate was observed for the event on 19/02/2010,
which produced 0.6 kg at the AF plot and 0.8 kg at the MC plot. Though this event did not
report a particularly high rainfall, in terms of either total depth or maximum intensity, it was
the longest event in the measurement period (12.2 h). The total erosion over the 24 events
amounted to 3.0 kg in the AF plot and 5.1 kg in the MC plot.
6.4.2 Annual water and sediment yield
The next two subsections present annual figures of water and sediment yield. First, a general
water balance is presented, focusing on the annual water yield at basin and plot scales, in that
order. Then, we report the behaviour of the basin-scale sediment concentrations over the
measurement year, its integration to provide a total sediment yield for the basin, and the total
annual erosion collected at the experimental plots.
A. Annual water yield
A simple water balance was elaborated for the one-year measurement period (07/03/2009 to
06/03/2010) in the experimental basin. All annual-based analyses were performed on a time
step t = 30 min, because all measured and modelled quantities are available at this time step.
Total rainfall R added up to 2849 mm and total streamflow depth Qd was 1710 mm (60% of
R). Basin actual evapotranspiration measured at the flux tower was 751 mm (26% of R). The
remaining 14% of the rainfall could be leaving the basin through deep percolation or
subsurface outflow downstream of the basin outlet. The variations in basin streamflow over
the measurement period are displayed in Fig. 42 (left axis). The minimum, mean and
maximum streamflow values were 0.025 mm h-1, 0.195 mm h-1 and 4.88 mm h-1 (obtained at
time step t = 30 min). A partition of streamflow was made for the annual time series
(constant slope method) indicating an annual SRb = 55 mm (3% of Qd, 2% of R).
149
Figure 42 Streamflow Q and suspended sediment concentration S (black lines) in the experimental basin for the one-year measurement period. A 95% confidence interval (grey region) is displayed around the estimated values. t = 30 min.
150
At the plot scale, the surface runoff recorded during one year in the two plots summed SRAF =
0.5 mm and SRMC = 1.0 mm, which as a percentage of rainfall are 0.02% and 0.05%,
respectively. These very low values reflect the extremely high infiltration and drainage
capacities, typical of volcanic soils.
Figure 43 shows the annual cumulative water balance components, at basin and plot scales.
The pattern of rainfall is reflected in the accumulation of surface runoff, and also in
streamflow (but smoothed though). The evapotranspiration was distributed more evenly
throughout the year.
Figure 43 Annual cumulative water balance of the experimental basin, from 07/03/2009 to 06/03/2010. R: rainfall, Q: streamflow, AET: measured actual evapotranspiration, SRAF: Surface runoff at coffee agroforestry plot and SRMC: surface runoff at coffee monoculture plot.
B. Annual sediment yield and plot erosion
The behaviour of basin sediment concentration during the measurement period is presented in
Fig. 42 (right axis), and looks coherent with streamflow, especially during storm events. The
minimum, mean and maximum sediment concentrations are 1.91 mg l-1, 53.4 mg l-1 and 888
mg l-1, respectively. The annual sediment yield was calculated using T = 17,520 (1 year at 30
min time steps) in Eq. (1), which adds up to 92.5 t. For each time step, the combined standard
uncertainties from the different error types (random, systematic or analytic) were calculated
for streamflow, sediment concentration and annual sediment yield, according to the GUM
methodology (JCGM, 2008) as presented in Table 7; along with the respective 95%
151
confidence interval. Fig. 42 shows that the streamflow uncertainties are almost negligible over
time, but considerable high for sediment concentration, especially during storm events. This is
detailed for two large events (on 05/06/2009 and 02/11/2009) as inserted graphs in Fig. 42
(the first of these events was already presented in Fig. 38 and in Section 6.4.1. The sediment
accumulation process over time in the basin, and the correspondent estimated uncertainties
are displayed in Figure 44a.
Figure 44. a) Annual cumulative sediment yield in the experimental basin (black line) with 95% confidence interval (gray region) for the one-year measurement period. b) Annual cumulative erosion in the agroforestry (AF) and monoculture (MC) experimental plots.
The estimated uncertainty for the annual sediment yield considers, at each time step, the
probabilistic distribution of errors about the expected value of the sum of individual sediment
yields, reducing the relative uncertainty in comparison to the individual relative uncertainties
a)
b)
152
of either St or Yt, and providing estimation errors of -5% and 12% for the annual sediment
load (with 95% of confidence).
Accumulated erosion in the two experimental plots is presented in Fig. 44b. The collection of
eroded material started on 05/06/2009, leaving the two first months of our measurement
period uncovered. This fact does not cause a problem of representativeness because there
were no major storm events during the unrecorded period (see Fig. 42), being such storm
events the sources of surface runoff (and plot erosion, as explained above). Total annual
erosion summed up to 4.3 kg in the AF plot and 7.1 kg in the MC plot.
6.4.3 Upscaling from storm to annual scale and from plot to basin scale
The next two sub-sections associate the results presented above for different temporal (storm
and annual) and spatial (basin and plot) scales. First, we compare the water balance and the
water yields; second, we contrast the erosion and sediment yields.
A. Upscaling the water yield
A comparison between water yield at storm and annual scales is made in Table 9, which also
contrasts the surface runoff at the plot and basin scales. Around 33% of the annual rainfall fell
during the 24 main storm events selected for analysis. While this storms yielded only a 6% of
the total annual streamflow (both quantities including baseflow), they produced almost 60%
of the annual SRb. It must be noticed that the storm selection procedure identified the main
storms in terms of water and sediment yield, but a large amount of smaller rainfall events are
present over the year, contributing with 40% of the annual SRb. Similarly, at the plot scale the
chosen storms yielded almost 50% of the annual SRplot, either under an AF or MC scheme, a
percentage similar to that of the entire basin. However, weak linear relationships were found
between the surface runoff from each of the plots and SRb reflected by determination
coefficients of R2 = 0.61 (AF) and R2 = 0.45 (MC). These regressions were performed only on
the 24 main events.
Comparing the average surface runoff of the two plots with the annual estimate for the whole
basin, it accounted for 1% during the 24 chosen storm events and 4% over the whole year,
indicating the occurrence of three quarters of the plot runoff during smaller storm events
153
throughout the year, but making clear that the overall basin surface runoff is mostly (96%)
coming from sources other than the plots.
Table 9. Comparing water yield from storm events to annual scale, and from plot to basin scale. The storm summary was computed by adding the properties of the 24 main storm events during the one-year measurement period. R: rainfall, Qd: streamflow depth, SRb: basin surface runoff, SRAF: surface runoff at the agroforestry plot, SRMC: surface runoff at the monoculture plot, SR: surface runoff.
Time scale R* Qd* SRb* SRAF* SRMC* %SR Plot/Basin
24 storm events 938 96 33 0.25 0.49 1% Annual 2855 1709 55 0.50 1.01 4%
% Storm/Annual 33% 6% 59% 49% 48% -
*Values in mm.
B. Upscaling the sediment yield
While the 24 greatest storm events produced 19.2 t, the annual sediment yield was estimated
in 92.5 t. It means that only 21% of the annual sediment load was produced during the largest
peak events, while most of it is yielded by the long, sustained recessions. If we consider also
the smallest storm events (which contribute the remaining 40% to the basin surface runoff),
then all storm events produce 30.9 t which accounts for 33% of the annual basin sediment
yield.
Assuming that the entire experimental basin is covered in equal proportions by an
arrangement of homogeneous coffee plantations under AF and MC systems, and that our two
experimental plots are representative of the other plots in the basin, the total plot erosion
would be 4.9 t in the basin. This is only a 5% of annual basin sediment yield.
6.5 Discussion
The discussion is structure in five subsections: the first two present the time scaling of water
yields at the basin and plot scales, while the third section makes the same time scaling for
sediment yields and discusses the annual erosion in the experimental plots. The fourth
subsection copes with the questions of where the basin sediment is coming from and when,
154
going deeper into our spatio-temporal findings. In the last subsection the uncertainty in our
estimations is briefly analyzed.
6.5.1 Water yield of the experimental basin: upscaling in time
The study of 24 storm events revealed low runoff coefficients (surface runoff divided by total
rainfall R) in the experimental basin, varying between 1% and 7% of R. Such coefficients are
notably lower than those found during storm events in other coffee basins, for instance,
Afandi (2005) found runoff coefficients from 0.3% to 19% in a small (2 ha) high sloped
(55%) basin, mostly covered by coffee under AF and MC schemes, but also by other land
uses, in minor degree. Manik (2008) studied two coffee basins of 4 and 27 ha under an AF
system, with respective slopes of 29% and 26%, and reporting runoff coefficients between 5%
and 39% for individual events. In the same study, four coffee basins under MC were
followed, with areas varying from 3 to 20 ha and mean slopes between 20% and 46%,
yielding runoff coefficients between 1% and 74%. Baumann et al. (2008) worked in two
coffee MC basins (2.8 and 1.7 ha), one with and the other without individual terraces, with
mean slopes of 46% and 33%, respectively, and runoff coefficients of 7% - 44% (terraces)
and 2% - 50% (no terraces). All these basins of comparison were smaller (or much smaller)
than our experimental basin, increasing the variability of their behaviours. They also had
higher slopes, which might affect the magnitude of the runoff coefficient. However, the range
of our observed coefficients overlaps the ranges of the comparison basins.
The estimation of SRr by applying a runoff coefficient to the road area of the basin (Table 8)
seems to provide a close estimation of the SRb. That relationship yields a R2=0.74 and is
corroborated by the Fig. 45a. As expected, the slope coefficient is slightly above 1, indicating
that basin runoff tends to be higher than road runoff. The small difference between these two
terms should be explained by the runoff inputs from plots, and by the net input/output balance
from all road-runoff crossing the borders of the basin (Fig. 19c).
The annual surface runoff coefficient for our experimental basin was reported above as 2%,
which is a third part of the 7% measured by Afandi (2005) in the 2 ha coffee basin, and also
much lower than the mean coefficients found by Baumann et al. (2008) for coffee MC basins:
6% (no terraces) and 9% (individual terraces). A much higher mean runoff coefficient (35%)
was reported by Hakkim et al. (2004) for a 0.75 km2 basin under coffee MC.
155
y = 1.05x
R2 = 0.74
0
1
2
3
4
5
0 1 2 3 4 5
SR r (mm)
SR
b (mm)
y = 73.35x + 0.24
R2 = 0.56
0
1
2
3
4
5
0.00 0.02 0.04 0.06
SR plot (mm)S
R b
(mm)
Figure 45: Surface runoff in the basin SRb as a function of a) surface runoff on the roads SRr and b) surface runoff on the plots SRplot.
6.5.2 Runoff on the experimental plots: upscaling in time
The surface runoff measured at the plots (here called SRplot) is defined as the mean runoff on
the agroforestry (AF) and monoculture (MC) coffee plots. The distribution of shade trees in
the basin is relatively homogeneous so the assumption here is that our two experimental plots
are equally representative of the basin land cover. Figure 45b confirms that the influence of
SRplot on SRb is proportional but less obvious (R2=0.56). The interception coefficient of the
linear regression suggests that, even in the absence of plot runoff, the basin surface runoff
during storm events can reach values of 0.24 mm, which coincides with the fact that during
this sort of events, the road runoff was large enough as to account for all basin surface runoff.
Then, the link between the presence of surface runoff and the intensity of the basin response
is coherent with the relationship presented in Fig. 39 between plot surface runoff and the
value of the respective basin peak streamflow. In terms of magnitude, we observed very low
plot runoff on both experimental plots over these 24 storm events, showing maximum runoff
coefficients of 0.1% and 0.2% for AF and MC, respectively, during the event of 17/08/2009.
The pooled runoff coefficients for the 24 events are 0.03% (AF) and 0.05% (MC). These
values will be later compared to the annual runoff coefficients in both systems, and also to
seasonal and annual figures presented in other studies.
a) b)
156
The annual surface runoff coefficients in our experimental plots were found to be 0.02% and
0.05% of the incident rainfall under AF and MC coffee systems, respectively. These values
fall within the range reported by Afandi et al. (2002a) on coffee plots of 20×5 m size, with
seasonal runoff coefficients varying from 0% to 1.5% in the presence of Paspalum
conjugatum weed. The same author measured higher coefficients on natural weeds (0% - 9%)
or on clean-weeded ground (7% - 16%). Harmand et al. (2007) found annual runoff
coefficients of 2% and 3% on 1×1 m plots for coffee AF (E. deglupta shade trees) and coffee
MC, respectively. Siles (2007) and Cannavo et al. (2011) used similar plots to measure annual
runoff coefficients on coffee plantations, which varied between 3% and 6% for AF plots and
between 8% and 10% for MC plots. Ataroff and Monasterio (1997) observed coefficients
around 5% on AF and around 6% on MC, using 6×2 m plots (only coffee over 10 years old
considered here). Dariah et al. (2004) measured coefficients between 2% - 2.5% (AF,
Gliricidia sepium) and 2.3% (MC), on 15×8 m size plots. Finally, Verbist et al. (2010)
reported runoff coefficients of 4% - 7% under coffee AF and of 10% - 15% under coffee MC,
on 4×10 m plots.
6.5.3 Sediment yield of the experimental basin and erosion in the experimental plots: upscaling in time
The experimental basin yielded sediment loads over the storm events with rates from 1×10-3
to 45×10-3 t ha-1, expressed by land area unit. The average rate for the two experimental plots
(AF and MC) during storm events went from 5×10-4 to 70×10-4 t ha-1, implying lower rates by
one order of magnitude. The annual sediment yield of the basin amounted to 1.0 t ha-1 y-1.
According to a compilation of studies on basin sediment yield in southeast Asia (Bruijnzeel,
2004), this value falls within the range 0.2 - 1 t ha-1 y-1 that corresponds to the category of
forest on volcanic soils (there is no category for coffee agroforestry systems). However, our
measurement is very different of that of Afandi (2005), also on volcanic soils, ranging from
10 to 59 t ha-1 y-1 (under undisturbed conditions). This latter study was carried out in a 2 ha
micro-basin on a very steep slope (55%) with manually collected sediment samples,
comprising an experimental setup very different from ours.
The erosion annual rates that we measured at the experimental plots are 0.04 (AF plot) and
0.07 t ha-1 y-1 (MC plot). These values are the lowest compared to others found for coffee
plots in the literature. For instance, Rijsdijk et al. (1990) measured 0.15 - 0.5 t ha-1 y-1 on
157
Wischmeier plots, Ataroff and Monasterio (1997) observed total erosions of 1.2 t ha-1 y-1 (AF)
and 1.8 t ha-1 y-1 (MC) on 31% sloped plots of 2×6 m, Afandi et al. (2002a) reported erosions
of 4.8 - 23 t ha-1 y-1 (coffee, clean-weeded cover), 0 - 2.7 t ha-1 y-1 (coffee with Paspalum
conjugatum cover) and 0 - 1.5 t ha-1 y-1 (coffee with natural weeds cover) on 27% sloped plots
of 20×5 m area. Gómez-Gómez (2005) recorded 0.2 - 4 t ha-1 y-1 on 7×2 m plots with slopes
between 25% and 30%, while Dariah et al. (2004) observed erosions between 1.1 and 1.3 t ha-
1 y-1 (AF) and of 1.5 t ha-1 y-1 (MC), on 50% - 60% slopes. Hairiah et al. (2005) measured an
average rate of 1.6 t ha-1 y-1 under AF (Gliricidia sepium and Paraserianthes falcataria) and
9.7 t ha-1 y-1 under MC, for coffee systems over 10 years, on 10×4 m plots.
6.5.4 Where is the sediment coming from and when? A spatio!temporal upscaling
Two thirds of the sediment yield that we measured for one year in this coffee agroforestry
basin (1.0 t ha-1 y-1) were produced during streamflow recession periods. It means that sources
other than surface runoff must be delivering 66% of the suspended sediment production in the
basin. This is illustrated in Fig. 46, which presents (a) the distribution of the 30 min sediment
yield estimates, as well as (b) their sum, both by streamflow decile group. The sediment yield
in the 10th decile group (i.e., the group containing the 10% highest streamflows) is greatly
right-skewed because it includes all the peak sediment-events, reaching a maximum yield of
1.4 t in 30 min and producing 32.6 t (35% of the total basin sediment yield). Our findings
differ from those of other studies also supported on continuous turbidity records, which
emphasize the sediment yield dependency on large storm events. For instance, García-Ruiz et
al. (2008) reports that 70% of the measured sediment was yielded during 3 storm events, Lane
et al. (2006) also accounts 70% of the total yield as the result of 2 single events, and Gallart et
al. (1999) attributes 90% of the sediment yield to only 10% of the storm events.
In our experimental basin, the sediment yielded during all storms events amounted to 30.9 t,
but only 4.9 t could have come from plot sheetwash erosion, which means that over storm
periods only 16% of the basin sediment is coming from plots. The water and sediment yields
respond to the local geological conditions and the farming management practices, and seemed
to be exceptionally low compared to the results of other studies, as discussed above. The tree
planting along some plot boundaries and the maintenance of riparian vegetation buffers in this
158
coffee farm, the latter proving to be very efficient according to Afandi et al. (2002b), might
lower the erosion delivery from the plots even more.
Figure 46: a) Distribution and b) sum of the sediment yield estimates (at 30 min time step), by streamflow decile group. (For the left graph: the vertical scale truncs the largest values of sediment yield in the 10th decile group, the box size represents the interquartile range, the whiskers extend to the most extreme data point which is no more than the interquartile range, circles are values beyond whiskers limits).
Such a low plot delivery to the streams entails other event-based sources of sediment and
transport mechanisms, such as subsurface runoff return at river channel sides and through
road lateral cuts, accelerated soil abrasion from the river bed/banks (perhaps intensified by
Hortonian infiltration excess on riparian zones), or abrasion from footpaths, unpaved roads
and its lateral gutters. Either the appropriate ground cover by (1) coffee canopies, (2) litter
from shade trees and from the seasonal pruning of the coffee plants and (3) free growing of
selected natural weeds, as well as the high soil infiltration and drainage capacities in the basin
(Gómez-Delgado et al., 2011a), determine that no landslides or gully erosion mechanisms are
present in this basin. The hypothesis that most of surface runoff (and hence sediment yield)
comes from roads during storm events seems reasonable from the field observations over two
years, and the simple estimation of road runoffs presented in Table 8 and Fig. 45a. These
hypothetical functions in our experimental basin seem coherent with several reports about the
a) b)
159
role of hillside trails, paths, settlements and roads in tropical regions (Ziegler et al., 2000;
Bruijnzeel, 2004; Ziegler et al., 2004; Rijsdijk, 2005; Sidle et al., 2006; Rijsdijk et al.,
2007b).
Now looking at the recession periods, they surprisingly contributed with most of the basin
sediment yield. Sediment yields are lower during low flows, but acts as a sustained flux over
time, so the total yield of the first 9 decile groups doubles that of the highest decile, this latter
containing all the storm events (Fig. 46b). Then, permanent soil abrasion from the river bed or
river banks is possible, as well as contributions from aquifer springs. Hence, as suggested by
Bruijnzeel (2004), more attention should be paid to underlying geological controls of basin
hydrological behaviour when analysing the effect of land use on water flows or sediment
production. Site effects may be more important than some surface particularities (Verbist et
al., 2010), and some processes like riparian slumping seems to be highly linked to the nature
of volcanic substrates (Rijsdijk et al., 1990).
6.5.5 Uncertainty in sediment estimations
The reliability of the estimations of suspended sediment concentration during low flows is
considered acceptable when the time step estimations are aggregated to produce long-run
totals (probabilistic reduction of the cumulative relative error). The random and systematic
standard uncertainties of the suspended sediment measurements were at least two orders of
magnitude lower than those from analytical calculations, in special from turbidity -
concentration relationships, as stated by Wass and Leeks (1999). It means that simplified
uncertainty estimations based in this error type only might be enough for rapid assessments of
sediment yield uncertainties (Lewis, 1996; Brasington and Richards, 2000; Sun et al., 2001;
Lane et al., 2006). Concerning the calibration of the OBS device, to survey the sediment flux
at the outlet of the basin, it must be noticed that the turbidity - concentration curve presented a
low degree of linearity between and sediment concentration (Figure 37), in opposition to the
findings of other studies using similar devices (Lewis, 1996; Lewis and Eads, 2001; Downing,
2006). This suggests that the particle properties of the sediment may not have been constant
throughout the sampling period (Wass and Leeks, 1999).
160
6.6 Conclusions
The spatio-temporal scaling approach, from plot to basin and from storm event to annual
scales, seems to provide good explanation capacities to trace the sources of soil erosion, and
to make hypothesis about the main sediment delivery processes, in a volcanic coffee-
agroforestry basin of Costa Rica.
By measuring the major hydrological processes, the water erosion and the sediment yields, we
found very low surface runoff and erosion rates at the plot scale, but also at the basin scale.
From the scaling exercises, we found discrepancies between the meters at each of these two
scales, which were acceptably explained by the intercepting and transporting effect of the
road network throughout the basin.
Recession periods supplied 66% of the annual sediment yield, contradicting many studies that
have measured or assumed most of the annual load being yielded during a few storm events.
Hence, only one third of the sediment produced in our basin was generated from storms, and
from this only the 16% could have come from plot erosion. That is why, although the already
well documented beneficial effects of agroforestry practices were evidenced (shaded coffee
plots produced a half of the erosion in comparison to monoculture coffee plots), the roads,
footpaths and stream channels play a much more important role on surface runoff and
sediment yield to river system. But of much greater importance is the recession sediment
yield, so more attention should be paid to the improvement of the available techniques for the
observation of suspended sediment during low flows, to the analysis of the stream channel
dynamics and to hydrogeological investigations.
The follow-up of the sediment yield by means of an optical backscatter sensor provided an
estimation of the annual sediment yield of the basin with acceptable uncertainties, this latter
calculated through a standard uncertainty approach that we newly applied to sediment load
calculations in the short- and long-terms.
List of symbols and abbreviations
AF agroforestry system / agroforestry plot C annual sediment yield at the basin outlet (t) h water stage height at the flume (cm) I10 maximum rainfall intensity in 10 min during a particular storm event (mm h-1)
161
MC monoculture system / monoculture plot Q streamflow (m s-1 or mm t
-1, with t either 10 or 30 min) Qd streamflow depth: total streamflow volume of a given period, divided by total basin
area (mm) Qpr peakflow rate in the streamflow during a storm event (mm h-1) R rainfall (mm) S suspended sediment concentration (mg l-1) SRb surface runoff on the basin (mm) SRAF surface runoff on the coffee agroforestry experimental plot (mm) SRMC surface runoff on the coffee monoculture experimental plot (mm) SRplot surface runoff on the coffee experimental plots (mm) SRr surface runoff on the roads (mm) T time step integration for the calculation of sediment loads (number of time steps) Y sediment yield during a given time span (kg or t) t time step (min) streamflow turbidity (NTU)
Appendix A: Calculation of annual cumulative sediment yield and its
uncertainty
A.1. Annual cumulative sediment yield
The sediment yield Yt of a time step t (min) is given by:
tSQY ttt ! (A.1)
where Qt (m3 s-1) is the basin streamflow at time t and St (mg l-1) is the respective suspended
sediment concentration (mg l-1). A conversion factor was applied to obtain the sediment yield
in kilograms (kg) or tonnes (t).
As explained in section 6.3.3, the total sediment yield YT over a period T was calculated using
Eqn. (1). If the time T is chosen to be the one-year measurement period, then YT is equal to the
annual cumulative sediment yield C (t):
"!
!y
t
tt tSQC1
1
(A.2)
The annual estimation were carried out at the time step t = 30 min, because all measured
quantities are available at this time step throughout the year of interest. For each time t, Qt
was obtained from a rating curve of the form:
# $ctt bhaQ %! (A.3)
where a, b, and c are the constant rating coefficients estimated for the hydraulic flume and ht
is the stage height of the streamflow (m), recorded at each time t by the pressure transducer.
162
The value of St is obtained from the power regression equation introduced in section 6.3.3 and
Fig. 37, i.e.:
&'( ttS ! (A.4)
where ! and " are the amplitude and exponent parameters of the power regression (estimated
through ordinary least squares method) and t is the turbidity (NTU) at time t as recorded bu
the optical backscatter sensor (OBS).
A.2. Definitions and equations for the uncertainty calculation
To calculate the uncertainty of our measured and calculated values (ht, t, Qt, St, C) we
followed the terminology suggested by the guide “Evaluation of measurement data - Guide to
the expression of uncertainty in measurement”, developed by the Working Group 1 of the
Joint Committee for Guides in Metrology (JCGM, 2008). We adopted some basic definitions
from that report to describe the components of uncertainty in Table 7. These terms were
adapted to our time series analysis, with variable inputs over time t, as follows:
Measurand: particular quantity M subject to measurement at time t.
Functional relationship f: relationship between measurand !t and input quantities "i,t on
which !t depends, and between output estimate #t and input estimates $i,t on which #t
depends, at time t.
Standard uncertainty [u(#t)]: uncertainty of the result of a measurement #t expressed as a
standard deviation, at time t.
Random standard uncertainty: standard uncertainty obtained by taking the positive square
root of the statistically evaluated variance, at time t.
Systematic standard uncertainty: evaluation of a non-statistical equivalent standard deviation,
at time t.
Analytical standard uncertainty: defined in this article as the standard uncertainty from the
application of particular analytical operations or models, such as statistical regression
analyses.
Combined standard uncertainty [uc(#t)]: standard uncertainty of the result of a measurement
#t, when that result is obtained from the values of a number of other quantities, at time t. It is
the estimated standard deviation associated with the result and is equal to the positive square
163
root of the combined variance obtained from all variance and covariance components,
however evaluated, using the law of propagation of uncertainty.
Law of propagation of uncertainty: combination of the uncertainties of the input quantities #i,t,
taken equal to the standard deviations of the probability distributions of the #i,t, to give the
uncertainty of the output quantity #t if that uncertainty is taken equal to the standard deviation
of the probability distribution of #t. This is:
# $ # $ # $ # $ # $" ""!
)
! %! *
*
*
*%++
,
-../
0
*
*!
n
i
n
i
n
ij
tjtiji
ji
ti
i
tc uurff
uf
u1
1
1 1,,,
2
2
2 ,2 111111
11
2 222 (A.5)
where: f# is the functional relationship between the output estimate #t and the ith input
estimate $i, u2($i,t) is the variance of each input $i,t at time t, and r($i, $j) is the correlation
coefficient of two input estimates $i and $j.
Relative combined standard uncertainty: if #t is of the form nji p
tn
p
tj
p
tit ,,, 1112 ! and the
exponents pi are known positive or negative numbers having negligible uncertainties, the
combined variance can be expressed as:
# $ # $ # $ # $ # $""")
! %!! 334
5
667
8
334
5
667
8%
334
5
667
8!3
4
567
8 1
1 1 ,
,
,
,
1
2
,
,
2
,2n
i
n
ij tj
tjj
ti
tii
ji
n
i ti
tii
t
tcupup
rupu
1
1
1
111
1
1
22
(A.6)
To handle an accumulation # of the values $t of a time series of the form "!
!T
t
t
1
12 , we
propose the next variation for the Eqn (A.5):
# $ # $ 9 :# $ # $ # $" " "!
)
! %!)*
*
*
*%++
,
-../
0
*
*!
T
t
T
ti
T
titj
tjtittitj
tjti
t
t
c uuff
uf
u1
1
1 1
2
2
2 2 111;11
11
2 222 (A.7)
The partial derivatives # $ 12 !** tf 12 inside of the left summation term of Eq. (A.5) and at
each time step t. The same applies to tif 12 ** and tjf 12 ** . The value %[tj-ti]($t) is the
correspondent autocorrelation of order tj-ti for the time series #t. Hence, if we assume that #t
is an autoregressive process of order one, we can simplify Eqn (A.7) as follows:
# $ # $ # $ # $ # $" "! !
)%!T
t
T
t
ttttc uuuu1 2
1122 2 111;12 (A.8)
where %1($t) is the first autocorrelation of the time series $t.
164
Variance from independent observations: when the best available estimate of the expected
value of a quantity q that varies randomly, and for which n independent observations qk have
been obtained under the same conditions of measurement, is the arithmetic mean q of the n
observations, then the experimental variance of the observations, which estimates the variance
&2 of the probability distribution of q, is given by the experimental standard deviation s
2(qk),
and the best estimate of the variance of the mean &2( q ) = &2/n is given by:
# $ # $n
qsqs k
22 ! (A.9)
Variance according the uniform distribution: it may be possible to estimate bounds (upper
and lower limits) for an input estimate "i,t, in particular, to state that “the probability that the
value of "i,t lies within the interval $ to + for all practical purposes is equal to one and the
probability that "i,t lies outside this interval is essentially zero”. If there is no specific
knowledge about the possible values of "i,t within the interval, one can only assume that it is
equally probable for "i,t to lie anywhere within it (a uniform distribution of possible values).
Then $i,t , the expectation or expected value of "i,t, is the midpoint of the interval, $i,t = ( $ +
+)/2, with associated variance eaqual to u2($i,t) = ( + – $)
2 / 12. If the difference between
the bounds, + – $, is equals 2 , then this equation becomes:
# $ 32,
2 !tiu 1 (A.10)
A.3. Calculation of the uncertainty of annual cumulative sediment
yield
Uncertainty in the measurement of height ht of water surface at the flume, at time t
The random uncertainty, also called uncertainty type A, comes from the random variability of
the measurements of ht at each t, as recorded by the pressure transducer. Then, according to
Eqn (A.8), at every time step t the standard uncertainty is equal to the standard error:
# $ # $ nhshu tAt !, (A.11)
where s(ht) is obtained at every time step t from the datalogger record (3 sampling
observations of ht per minute, n = 90 sampling observations per t = 30 min). The systematic
(type B) standard uncertainty was estimated from to the maximum error in the visual
measurement of the water level, h = 0.001 m, and from Eqn (A.9):
165
# $ 3, hhu Bt ! (A.12)
Then, the combined standard uncertainty of height ht at time t is the square root of:
# $ # $ # $B,2
A,22
ttt huhuhu %! (A.13)
Uncertainty in the estimation of streamflow Qt at the flume site, at time t
Given the functional relationship fQ defined in Eqn (A.3), the random and systematic
uncertainties are calculated, for time t, according to Eqn (A.5) as:
# $ # $ # $ # $kt
c
tkt
Q
kt hubhcahuh
fQu ,
1,,
)%!*
*! (A.14)
where k is the uncertainty type: A or B. In addition, the analytical uncertainty u(Qt,C) is
associated to the statistical fitting of the Eqn (A.3). Both uncertainties A and C were
calculated by the WinFlme software as ±3.6% of Qt, at 95% level of confidence, while uncertainty
B was calculated from Eqns (A.12) and (A.14). The squares of these terms add up to produce the total
variance in the Qt estimation:
# $ # $ # $B,2
CA,22
ttt QuQuQu %! % (A.15)
whose square root is the uncertainty in the estimation of streamflow Qt at time t.
Uncertainty in the measurement of turbidity t at the flume, at time t
Uncertainty A is estimated from Eqn (A.8) and from the standard deviation s( t) obtained at
every time step t from the datalogger record (3 sampling observations of t per minute, n = 90
sampling observations per t = 30 min):
# $ nsu tAt '' !)( , (A.16)
The uncertainty B is calculated from assuming an error in the reading from the turbidimeter
= 0.001 NTU (it is an electronic device), and from Eqn (A.9):
# $ 3, '' !Btu (A.17)
Then, the total uncertainty in turbidity t at time t, is the square root of:
# $ # $ # $B,2
A,22
ttt uuu ''' %! (A.18)
166
Uncertainty in the estimation of suspended sediment concentration St, at time t
From the functional relationship fS defined in Eqn (A.4A.3) and from Eqn (A.5), the
propagation of random and systematic uncertainties from turbidity measurements are
calculated, for time t, according to:
# $ # $ # $tttS
t uuf
Su ''&(''
& 1BA,
)% !
*
*! (A.19)
In addition, the analytical uncertainty type C u(St,C) is associated to the statistical fitting of the
Eqn (A.3) and taken equal to the difference between the regressed value of sediment
concentration tS and each of the two bounds LtS ,ˆ and UtS ,
ˆ (lower, upper) defined by one
standard error (the corrected RMSE of the power regression) about tS . Given that all tS values
and the bounds [ UtLt SS ,,ˆ,ˆ ] origin from a exponential transformation, they report asymmetrical
upper/lower uncertainties u(St,C)+ and u(St,C)$ that will be treater separately from now on. The
square root of the sum of squares of the standard uncertainties is the total uncertainty in the
estimation of St (separately for the asymmetrical upper and lower uncertainties):
# $ # $ # $)%) %! C,
2BA,
22ttt SuSuSu (A.20)
# $ # $ # $%%% %! C,
2BA,
22ttt SuSuSu (A.21)
Uncertainty in the estimation of sediment yield Yt at the flume, at time t
The uncertainty of Yt is derivated from Eqn (A.6) to propagate the uncertainties of Qt and St
throughout the product described in the Eqn. (A.1). The uncertainty of Qt is obtained from
Eqn (A.15). Again, the lower and upper uncertainties of Yt are computed separately, as
functions of Eqns (A.20) and (A.21):
# $ # $ # $ # $ # $ # $t
t
t
ttt
t
t
t
t
t
t
S
Su
Q
QuSQr
S
Su
Q
Qu
Y
Yu ))) %34
567
8%3
4
567
8!3
4
567
8,2
222
(A.22)
# $ # $ # $ # $ # $ # $t
t
t
ttt
t
t
t
t
t
t
S
Su
Q
QuSQr
S
Su
Q
Qu
Y
Yu %%% %34
567
8%3
4
567
8!3
4
567
8,2
222
(A.23)
167
Uncertainty in the estimation of annual sediment yield C at the flume site
The annual cumulative sediment yield from Eqn (A.2) is obtained adding up all the terms in
the sediment yield time series Yt over time t. Eqn (A.8) is used to obtain separately the
respective uncertainties from lower and upper Yt uncertainties:
# $ # $ # $ # $ # $" "! !
))))) %!T
t
T
t
tttt YuYuYYuCu1 2
1122 2 ; (A.24)
# $ # $ # $ # $ # $" "! !
%%)%% %!T
t
T
t
tttt YuYuYYuCu1 2
1122 2 ; (A.25)
Acknowledgements
This work was funded by the Centre de Coopération Internationale en Recherche
Agronomique pour le Développement (CIRAD, France), the European project CAFNET
(EuropAid/121998/C/G) and the Cafetalera Aquiares farm (http://www.cafeaquiares.com).
CATIE (Centro Agronómico Tropical de Investigación y Enseñanza) provided collaboration
and infrastructure. We gratefully thank the Cafetalera Aquiares farm for allowing us to install
the experimental basin and for providing us research facilities. Alexis Pérez carried out
fieldwork and laboratory analysis of soils, Álvaro Barquero and his family contributed with
the field work. Field trips and discussions with José Alberto Zúñiga, Lars Gottschalk, Irina
Krasovskaia, Sadí Laporte and Rafael Chacón provided useful insight. Gustavo Calvo and the
Chemical Laboratory of ICE kindly backed the analysis of sediment samples. Federico
Gómez-Delgado was supported by a CIRAD fellowship during his PhD.
References
Afandi, 2005. Runoff and Sediment Yield from Coffee Micro-Catchment Under Humid
Tropical Climate of Lampung, Indonesia. In: Agus, F., Noordwijk, M.v. (Eds.), Alternatives
to Slash and Burn in Indonesia: Facilitating the development of agroforestry systems: Phase 3
Synthesis and Summary report. World Agroforestry Centre, Sindang Barang, Bogor,
Indonesia.
Afandi, Manik, T.K., Rosadi, B., Utomo, M., Senge, M., Adachi, T., Oki, Y., 2002a. Soil
Erosion under Coffee Trees with Different Weed Managements in Humid Tropical Hilly Area
168
of Lampung, South Sumatra, Indonesia. Journal of the Japanese Society of Soil Physics 91, 3-
14.
Afandi, Rosadi, B., Maryanto, Nurarifani, Utomo, M., Senge, M., Adachi, T., 2002b.
Sediment Yield from Various Land Use Practices in a Hilly Tropical Area of Lampung
Region, South Sumatra, Indonesia. Journal of the Japanese Society of Soil Physics 91, 25-38.
Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration: Guidelines for
computing crop water requirements. Food and Agriculture Organization, Rome.
Ataroff, M., Monasterio, M., 1997. Soil erosion under different management of coffee
plantations in the Venezuelan Andes. Soil Technology 11, 95-108.
Aylward, B., 2004. Land use, hydrological function and economic valuation. The joint
UNESCO International Hydrological Programme (IHP) - International Union of Forestry
Research Organizations (IUFRO) symposium and workshop, Forest - water - people in the
humid tropics: past, present and future hydrological research for integrated land and water
management, Universiti Kebangsaan Malaysia, 30 July-4 August 2000. Cambridge University
Press, pp. 99-120.
Baumann, J., Rodríguez Morales, J.A., Arellano Monterrosas, J.L., 2008. The effect of
rainfall, slope gradient and soil texture on hydrological processes in a tropical watershed. 15th
International Congress of ISCO. International Soil Conservation Organization, Budapest,
Hungary.
Brasington, J., Richards, K., 2000. Turbidity and suspended sediment dynamics in small
catchments in the Nepal Middle Hills. pp. 2559-2574.
Bruijnzeel, L.A., 2004. Hydrological functions of tropical forests: not seeing the soil for the
trees? Agriculture, Ecosystems & Environment 104, 185-228.
Cannavo, P., Sansoulet, J., Harmand, J.M., Siles, P., Dreyer, E., Vaast, P., 2011. Agroforestry
associating coffee and Inga densiflora results in complementarity for water uptake and
decreases deep drainage in Costa Rica. Agriculture, Ecosystems & Environment 140, 1-13.
Clifford, N.J., Richards, K.S., Brown, R.A., Lane, S.N., 1995. Laboratory and field
assessment of an infrared turbidity probe and its response to particle size and variation in
169
suspended sediment concentration. Hydrological sciences journal = Journal des sciences
hydrologiques. 40, 771-791.
Chappell, N.A., McKenna, P., Bidin, K., Douglas, I., Walsh, R.P.D., 1999. Parsimonious
modelling of water and suspended-sediment flux from nested-catchments affected by
selective tropical forestry.
Dabney, S.M., Locke, M.A., Steinriede Jr, R.W., 2006. Turbidity sensors track sediment
concentrations in runoff from agricultural fields. Federal Interagency Sedimentation
Conference Proceedings, Nevada, USA.
Dariah, A., Agus, F., Arsyad, S., Sudarsono, dan-Maswar, 2004. Erosi dan aliran permukaan
pada lahan pertanian berbasis tanaman kopi di Sumberjaya, Lampung Barat. Agrivita 26, 52-
60.
de Vente, J., Poesen, J., 2005. Predicting soil erosion and sediment yield at the basin scale:
Scale issues and semi-quantitative models. Earth-Science Reviews 71, 95-125.
Downing, J., 2006. Twenty-five years with OBS sensors: The good, the bad, and the ugly.
Continental Shelf Research 26, 2299-2318.
Foster, I.D.L., Millington, R., Grew, R.G., 1992. The impact of particle size controls on
stream turbidity measurement; some implications for suspended sediment yield estimation. In:
IAHS (Ed.), Erosion and Sediment Transport Monitoring Programmes in River Basins. IAHS,
Oslo, Norway, pp. 51-62.
Gallart, F., Llorens, P., Latron, J., Regüés, D., 1999. Hydrological processes and their
seasonal controls in a small Mediterranean mountain catchment in the Pyrenees. Hydrol.
Earth Syst. Sci. 6, 527-537.
García-Ruiz, J.M., Regüés, D., Alvera, B., Lana-Renault, N., Serrano-Muela, P., Nadal-
Romero, E., Navas, A., Latron, J., Martí-Bono, C., Arnáez, J., 2008. Flood generation and
sediment transport in experimental catchments affected by land use changes in the central
Pyrenees. Journal of Hydrology 356, 245-260.
Gilroy, E.J., Hirsch, R.M., Cohn, T.A., 1990. Mean square error of regression-based
constituent transport estimates. Water Resour. Res. 26, 2069-2077.
170
Gippel, C.J., 1989. The use of turbidimeters in suspended sediment research. Hydrobiologia
176-177, 465-480.
Gippel, C.J., 1995. Potential of turbidity monitoring for measuring the transport of suspended
solids in streams. pp. 83-97.
Gómez-Delgado, F., Roupsard, O., Moussa, R., le Maire, G., Taugourdeau, S., Bonnefond,
J.M., Pérez, A., van Oijen, M., Vaast, P., Rapidel, B., Voltz, M., Imbach, P., Harmand, J.M.,
2010. Modelling the hydrological behaviour of a coffee agroforestry basin in Costa Rica.
Hydrol. Earth Syst. Sci. Discuss. 7, 3015-3071.
Gómez-Gómez, R., 2005. Efecto del control de malezas con paraquat y glifosato sobre la
erosión y pérdida de nutrimentos del suelo en cafeto. Agronomía Mesoamericana 16, 77-87.
Hairiah, K., Suprayogo, D., Widianto, Prayogo, C., 2005. Trees that produce mulch layers
which reduce run-off and soil loss in coffee multistrata systems. Alternatives to slash and
burn in Indonesia: facilitating the development of agroforestry systems: phase 3 synthesis and
summary report. World Agroforestry Centre - ICRAF, SEA Regional Office, Bogor,
Indonesia, pp. 9-30.
Harmand, J.-M., Ávila, H., Dambrine, E., Skiba, U., de Miguel, S., Renderos, R., Oliver, R.,
Jiménez, F., Beer, J., 2007. Nitrogen dynamics and soil nitrate retention in a Coffea arabica -
Eucalyptus deglupta agroforestry system in Southern Costa Rica. Biogeochemistry 85, 125-
139.
Jansson, M.B., 1992. Turbidimeter measurements in a tropical river, Costa Rica. In: IAHS
(Ed.), Erosion and Sediment Transport Monitoring Programmes in River Basins. IAHS, Oslo,
Norway, pp. 71-78.
Jansson, M.B., Erlingsson, U., 2000. Measurement and quantification of a sedimentation
budget for a reservoir with regular flushing. Regul. Rivers-Res. Manage. 16, 279-306.
Jaubert, M., 2001. River basin management: a new challenge for ICE. Tecnología ICE 11,
100-111.
JCGM, 2008. Evaluation of measurement data - Guide to the expression of uncertainty in
measurement. BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP, OIML, Sèvres, France.
171
Lana-Renault, N., Regues, D., Marti-Bono, C., Begueria, S., Latron, J., Nadal, E., Serrano, P.,
Garcia-Ruiz, J.M., 2007. Temporal variability in the relationships between precipitation,
discharge and suspended sediment concentration in a small Mediterranean mountain
catchment. Nordic hydrology 38, 139.
Lane, P.N.J., Sheridan, G.J., Noske, P.J., 2006. Changes in sediment loads and discharge from
small mountain catchments following wildfire in south eastern Australia. Journal of
Hydrology 331, 495-510.
Lenzi, M.A., Marchi, L., 2000. Suspended sediment load during floods in a small stream of
the Dolomites (northeastern Italy). CATENA 39, 267-282.
Lewis, J., 1996. Turbidity-Controlled Suspended Sediment Sampling for Runoff-Event Load
Estimation. Water Resour. Res. 32, 2299-2310.
Lewis, J., 2003. Turbidity-controlled sampling for suspended sediment load estimation. In:
IAHS (Ed.), Erosion and sediment transport measurement in rivers: technological and
methodological advances. IAHS, Oslo, Norway, pp. 13-20.
Lewis, J., Eads, R., 2001. Turbidity threshold sampling for suspended sediment load
estimation. 7th Federal Interagency Sedimentation Conference, Nevada, USA, pp. III110 -
III117.
Manik, T.K., 2008. Hydrological consequences of converting forestland to coffee plantations
and other agriculture crops on Sumber Jaya watershed, West Lampung, Indonesia. PhD
Thesis. National University of Singapore, Singapore, p. 281.
McCuen, R.H., 2005. Hydrologic analysis and design. Pearson Prentice Hall, Upper Saddle
River, NJ, USA.
MINAE-RECOPE, 1991. Mapa geológico de Costa Rica, Scale 1:200000. San José, Costa
Rica.
Minella, J.P.G., Merten, G.H., Reichert, J.M., Clarke, R.T., 2008. Estimating suspended
sediment concentrations from turbidity measurements and the calibration problem. pp. 1819-
1830.
Mora-Chinchilla, R., 2000. Geomorfología de la cuenca del Río Turrialba (unpublished).
Universidad de Costa Rica, San José, Costa Rica.
172
Pagiola, S., 2008. Payments for environmental services in Costa Rica. Ecological Economics
65, 712-724.
Peel, M.C., Finlayson, B.L., McMahon, T.A., 2007. Updated world map of the Köppen-
Geiger climate classification. Hydrol. Earth Syst. Sci. 11, 1633-1644.
Raclot, D., Le Bissonnais, Y., Louchart, X., Andrieux, P., Moussa, R., Voltz, M., 2009. Soil
tillage and scale effects on erosion from fields to catchment in a Mediterranean vineyard area.
Agriculture, Ecosystems & Environment 134, 201-210.
Rijsdijk, A., 2005. Evaluating sediment sources and delivery in a tropical volcanic watershed.
IAHS publication 292, 16-23.
Rijsdijk, A., Bruijnzeel, L.A., Bremmer, C.N., Kaatee, E.G., S., C.K., 1990. Sediment sources
and their evaluation in the Upper Konto watershed, East Java, Indonesia. In: IAHS (Ed.),
Proceedings of the Fiji Symposium, June 1990. IAHS, Suva, Fiji, p. 391.
Rijsdijk, A., Bruijnzeel, L.A., Prins, T.M., 2007a. Sediment yield from gullies, riparian mass
wasting and bank erosion in the Upper Konto catchment, East Java, Indonesia.
Geomorphology 87, 38-52.
Rijsdijk, A., Sampurno Bruijnzeel, L.A., Sutoto, C.K., 2007b. Runoff and sediment yield
from rural roads, trails and settlements in the upper Konto catchment, East Java, Indonesia.
Geomorphology 87, 28-37.
Roupsard, O., Bonnefond, J.-M., Irvine, M., Berbigier, P., Nouvellon, Y., Dauzat, J., Taga, S.,
Hamel, O., Jourdan, C., Saint-André, L., Mialet-Serra, I., Labouisse, J.-P., Epron, D., Joffre,
R., Braconnier, S., Rouzière, A., Navarro, M., Bouillet, J.-P., 2006. Partitioning energy and
evapo-transpiration above and below a tropical palm canopy. Agricultural and Forest
Meteorology 139, 252-268.
Sajeena, S., Nandakumar, V., Hakkim, V.M.A., 2004. Runoff variation due to landuse change
in small watersheds of Western Ghats. Indian Journal of Environmental Protection 3, 99-105.
Schoellhamer, D.H., Wright, S.A., 2003. Continuous measurement of suspended-sediment
discharge in rivers by use of optical backscatterance sensors. IAHS publication 283, 1 - 9.
Seeger, M., Errea, M.P., BeguerIa, S., Arnáez, J., MartI, C., GarcIa-Ruiz, J.M., 2004.
Catchment soil moisture and rainfall characteristics as determinant factors for
173
discharge/suspended sediment hysteretic loops in a small headwater catchment in the Spanish
pyrenees. Journal of Hydrology 288, 299-311.
SEPSA, 2009. Boletín Estadístico Agropecuario 19. Secretaría Ejecutiva de Planificación
Sectorial Agropecuaria, San José, Costa Rica.
Sidle, R.C., Ziegler, A.D., Negishi, J.N., Nik, A.R., Siew, R., Turkelboom, F., 2006. Erosion
processes in steep terrain--Truths, myths, and uncertainties related to forest management in
Southeast Asia. Forest Ecology and Management 224, 199-225.
Siles, P., 2007. Hydrological processes (water use and balance) in a coffee (Coffea arabica L.)
monoculture and a coffee plantation shaded by Inga densiflora in Costa Rica. Ecole Doctorale
RP2E (Ressources Procédés Produits Environnement). Université Henri Poincaré, Nancy-I,
Nancy, France, p. 153.
Sun, H., Cornish, P.S., Daniell, T.M., 2001. Turbidity-based erosion estimation in a
catchment in South Australia. Journal of Hydrology 253, 227-238.
Vanacker, V., Molina, A., Govers, G., Poesen, J., Deckers, J., 2007. Spatial variation of
suspended sediment concentrations in a tropical Andean river system: The Paute River,
southern Ecuador. Geomorphology 87, 53-67.
Vega, F.E., Rosenquist, E., 2001. The coffee berry borer and coffee research at the United
States Department of Agriculture. World Coffee Conference, London, UK.
Verbist, B., Poesen, J., van Noordwijk, M., Widianto, Suprayogo, D., Agus, F., Deckers, J.,
2010. Factors affecting soil loss at plot scale and sediment yield at catchment scale in a
tropical volcanic agroforestry landscape. CATENA 80, 34-46.
Verbist, B.J.P., Widayati, A., van Noordwijk, M., 2003. The link between land and water
prediction of sediment sources in a previously forested watershed in Lampung, Sumatra,
Indonesia. Modsim 2003: International Congress on Modelling and Simulation, Vols 1-4 -
Vol 1: Natural Systems, Pt 1; Vol 2: Natural Systems, Pt 2; Vol 3: Socio-Economic Systems;
Vol 4: General Systems. Univ Western Australia, Nedlands, pp. 560-565.
Wahl, T.L., Clemmens, A.J., Replogle, J.A., Bos, M.G., 2000. WinFlume - Windows-based
software for the design of long-throated measuring flumes. Fourth Decennial National
Irrigation Symposium. American Society of Agricultural Engineers, Phoenix, Arizona.
174
Walling, D.E., 1977. Assessing Accuracy of Suspended Sediment Rating Curves for a Small
Basin. Water Resources Research 13, 530-538.
Walling, D.E., Webb, B.W., 1982. Sediment availability and the prediction of storm-period
sediment yields. IAHS publication 137, 327 - 337.
Walling, D.E., Webb, B.W., Woodward, J.C., 1992. Some sampling considerations in the
design of effective strategies for monitoring sediment-associated transport. In: IAHS (Ed.),
Erosion and Sediment Transport Monitoring Programmes in River Basins. IAHS, Oslo,
Norway, pp. 279-288.
Wass, P.D., Leeks, G.J.L., 1999. Suspended sediment fluxes in the Humber catchment, UK.
pp. 935-953.
Wass, P.D., Marks, S.D., Finch, J.W., Leeks, G.J.L., Ingram, J.K., 1997. Monitoring and
preliminary interpretation of in-river turbidity and remote sensed imagery for suspended
sediment transport studies in the Humber catchment. Science of The Total Environment 194-
195, 263-283.
Webb, B.W., Walling, D.E., 1982. The magnitude and frequency characteristics of fluvial
transport in a devon drainage basin and some geomorphological implications. CATENA 9, 9-
23.
Wunder, S., Engel, S., Pagiola, S., 2008. Taking stock: A comparative analysis of payments
for environmental services programs in developed and developing countries. Ecological
Economics 65, 834-852.
Ziegler, A.D., Giambelluca, T.W., Sutherland, R.A., Nullet, M.A., Yarnasarn, S., Pinthong, J.,
Preechapanya, P., Jaiaree, S., 2004. Toward understanding the cumulative impacts of roads in
upland agricultural watersheds of northern Thailand. Agriculture, Ecosystems & Environment
104, 145-158.
Ziegler, A.D., Sutherland, R.A., Giambelluca, T.W., 2000. Runoff generation and sediment
production on unpaved roads, footpaths and agricultural land surfaces in northern Thailand.
Earth Surface Processes and Landforms 25, 519-534.
175
CHAPTER 7
Improving water and sediment peakflow simulations from an
eco!hydrological model, by including the effect of direct road!
runoff: continuous versus event!based validation strategies15
!"!#$%&'()*!+,-!./0"& !"!#!$%&'(')*$+,-./ *0 !1!$+,2)*$3,-// 0$
$45+67!$83+$91,:;,&/!$<=>?@A@$3,BC.)&&')*!$<* B1)$
"$549!$49DE9!$F@@>G$; B$H,/I!$4,/C $+'1 $
1$46J59!$KFK@!$>@L@F$J-**' &" !$4,/C $+'1 $
0$5D+6!$83+$M5;6N!$G$E& 1)$O' & !$>?@A@$3,BC.)&&')*!$<* B1)$
#4,**)/.,B0'B2$ -CP,*Q$
)=R '&Q$<S,R)T7U'1)V2,V1*$
J)&).P,B)Q$WL@AX$GL$LY$GL$ZK$
< [Q$WL@AX$GL$LY$G@$?L$
Abstract
A lumped eco-hydrological model called Hydro-SVAT (Gómez-Delgado et al., 2011a)
performed well and was validated at the annual time scale for modelling streamflow, but
underestimated peakflows during some large storm events, which could affect significantly
the annual estimation of sediment yield. It was hypothesized that not only plots, but also roads
could contribute to surface runoff generation during storm events. The latter source was
introduced in the revised version of the model. In this work we compared four alternative
model structures and parameterization schemes that were validated by two different
approaches (continuous-time and event-based) and evaluated using four different criteria
15 Draft paper in preparation (not ready for submission yet):
Gómez-Delgado, F., Roupsard, O., Moussa, R., 2011. Improving water and sediment peakflow simulations by
the design of a semi-distributed eco-hydrological model including the effect of direct road-runoff: continuous
versus event-based validation strategies. In prep.
176
(Nash-Sutcliffe, C2M, errors on volume and peakflow). The introduction of roads as a
secondary source of surface runoff did not significantly improve the performances of the
initial model for simulating large peakflows. A new hypothesis was proposed but not tested
yet: undetermined impermeable areas other than roads (like foot pathways) could significantly
increase the maximum rates of surface runoff over large storm events.
Key words: Model, parameterization, validation, storm event, road, surface runoff,
baseflow, sediment yield, coffee agroforestry$
7.1 Introduction
Hydrological models can be calibrated and validated using time-series of streamflow, and
inverted in order to generate estimates of instant stocks and fluxes of water, and more
specifically to infer the partitioning between infiltration and surface runoff, which is of
paramount importance for assessing Hydrological Environmental Services. Although the
annual performances of such models can be very good for predicting the total amount of
streamflow and give reliable estimates of the water balance partitioning, it is sometimes
difficult to achieve simultaneously faithful predictions at both, annual and per-event time
scales. Nevertheless, a correct prediction of peakflow for any class of storm (from small to
large) is necessary for achieving accurate predictions of the annual sediment yield basing on a
hydrological model. In particular, it is necessary to simulate very well the largest peakflow
events that usually carry the largest concentrations of sediments.
In a previous work, Gómez-Delgado et al. (2011a) built, calibrated and validated a lumped
eco-hydrological model called Hydro-SVAT that performed well on an annual base and for a
coffee agroforestry microbasin of Costa Rica, but stressed that peakflows were
underestimated, which would likely affect the assessment of annual sediment yield, and
suspected that roads could contribute significantly as an important source of surface runoff
during storm events (Gómez-Delgado et al., 2011b).
Several attempts were performed here, simultaneously aiming at minimizing the number of
parameters of the model and at predicting large peakflow events more accurately, introducing
roads as a second source of surface runoff, in complement to plots. Although some attempts
177
have been made to produce general operational protocols for model parameterization,
calibration, and validation (Anderson and Woessner, 1992; Refsgaard, 1997; Henriksen et al.,
2003), no consensus on a “universal” methodology has been achieved in hydrology research
in general, and in rainfall-streamflow modelling in particular. As such, an infinite number of
techniques can be combined through these three modelling steps, making very difficult the
definition of an optimum procedure to select the model structure, the objective function(s) to
calibrate and the validation criteria.
Whatever combination of parameterization, calibration and validation techniques, some basic
modelling principles should be looked at. In the particular case of conceptual models, whose
aim is to provide a simplified representation of the physical system, the election and quantity
of process-parameters (non-measurable) during the model building/parameterization stage
should depend more on the physical processes to be described, than on any rule or predefined
protocol concerning model parsimony. Then, instead of parameterizing a model by
minimizing the number of parameters to calibrate, more importance should be given to the
physical meaning of the process-parameters throughout the calibration process, according to
the model purposes and the desired output variables. On the other hand, over-parameterization
must also be avoided (Perrin, 2000; Mathevet, 2005), and some general rules of thumb could
guide the parameterization process. For instance, for daily scale lumped rainfall-streamflow
models, between three and five parameters might be sufficient to achieve an adequate
accuracy/simplicity balance (Perrin et al., 2001). Sorooshian and Gupta (1983) showed that
the inadequacies in the structure of the models are primarily responsible for the inability to
obtain unique and conceptually realistic calibrated parameter sets, caused by poor properties
of the optimization response surface. Beven (1993; 2006) has widely discussed the concept of
equifinality of models and parameter sets, stating that errors in model structure, boundary
conditions and observed data can lead to an equal reproduction of observed variables by many
different combinations of parameter sets (in particular when a single criteria is employed to
assess the model performance).
Looking at the calibration stage, either single-objective (Diskin and Simon, 1977; Pickup,
1977; Sorooshian and Gupta, 1983; Duan et al., 1992) or multi-objective (Lindström, 1997;
Yapo et al., 1998; Madsen, 2000; Moussa and Chahinian, 2009; Efstratiadis and
Koutsoyiannis, 2010) automated methods have been widely used, the latter being developed
178
mainly during the last decade. With this huge mosaic of methods at hand (with their
respective advantages and limitations), the conceptual meaning of the process-parameters is
still a relevant principle to consider. In the case of single-objective methods (those that
concern us in this study) a common problem is the existence of multi-modal response
surfaces, which can cause that the parametric point of convergence is just a local optimum,
and it varies according to the given starting values (Pickup, 1977). However, a local optimum
could be satisfactory if the fitted parameter values exhibit conceptual coherency, in opposition
to a global optimum with superior performance, but based on strange, meaningless or
theoretically inconsistent parameter values that might produce incoherent predictions under
changing conditions or model extrapolations.
Concerning the validation step, two main approaches have been proposed by Willmott et al.
(1985): (1) “scientific”: the extent to which the model's behaviour is consistent with
prevailing scientific theory, and (2) “operational”: centred on model accuracy, i.e., the extent
to which model-predicted events approach a corresponding set of independently obtained,
reliable observations (usually measured), and precision, i.e., the degree to which model-
predicted values approach a linear function of the reliable observations approaches. The first
approach demands that the model behaviour is examined for each individual sub-system of
the basin for which the model is intended to provide realistic descriptions (Beldring, 2002).
This sort of analysis has already been conducted for the experimental basin under study
(Gómez-Delgado et al., 2011a). On the other hand, the second approach, dealing with the
operational evaluation of precision and accuracy of the models, often provides the most
tangible means of establishing model credibility (Willmott et al., 1985) and is the one that this
paper will cope with.
This study aims to compare four model structures and parameterization alternatives, defined
either by different model structures, or by the reduction of the parametric space to optimize.
As both overall (general shape, volume) and storm event properties (idem, plus peakflow) are
of interest in the study region, two time-based validation approaches were tested: the first
defined on a continuous-time basis (summary measures calculated over all time steps in the
study period), and the second defined specifically for each storm event (measures calculated
over hydrograph storms only).
179
Three main considerations in our analysis are:
(1) Parameterization step: the time-dependent sensitivity analysis performed by Gómez-
Delgado et al (2011a) was used to reduce the initial model presented in that publication (10
parameters) to a simpler one (7 parameters), as suggested by Pickup (1977). However, as such
procedure was carried out on one particular study period and the simplified model is here
applied in a different time interval, the generality of such method will be tested.
(2) Calibration step: although we base this procedure on a single-objective automated method,
we applied a so-called “directed starting point” procedure to deal with the problems of multi-
modal parametric spaces or equifinality, aiming to control the interrelations between the
parameter values before and after the model calibration process. This pretends to maintain the
execution of all models within a realistic parameter space for comparison purposes, not
expecting to reach the “perfect” model calibration (which is intended, for instance, by using
global search methods).
(3) Validation step: on the “operational” approach mentioned above, four performance
measures will be explored under a continuous-time assessment (evaluating all time steps in
the study period), while five summary measures will be studied under an event-based
assessment (using only those time steps during storm events). Cumulative frequency
distributions are also tested, in order to better interpret the previously mentioned performance
or summary measures mentioned above.
After completing these steps, we conclude on the efficiency of the new double-source surface
runoff model (including road effect) to predict streamflow accurately at both, annual and
storm event scales.
7.2 Study site: location, soil and climate
The area of interest is located in Reventazón river basin, in the Central-Caribbean region of
Costa Rica (Fig. 10a,b). It lies on the slope of the Turrialba volcano (central volcanic
mountain range of the country) and drains to the Caribbean Sea. The Aquiares coffee farm is
one of the largest in Costa Rica (6.6 km2), “Rainforest Alliance TM” certified, 15 km from
CATIE (Centro Agronómico Tropical de Investigación y Enseñanza). Within the Aquiares
farm, we selected the Mejías creek micro-basin (Fig. 10c) for the “Coffee-Flux” experiment,
180
which has been described with more details by Gómez-Delgado et al. (2011a). The “Coffee-
Flux” experimental basin and instrument layout was designed to trace the main water balance
components and sediment yield, employing spatially representative methods (Fig. 10c). It is
part of the FLUXNET network for the monitoring of greenhouse gases of terrestrial
ecosystems.
The basin is placed between the coordinates -83º44’39” and -83º43’35” (West longitude), and
between 9º56’8” and 9º56’35” (North latitude) and is homogeneously planted with coffee
(Coffea arabica L., var Caturra) on bare soil, shaded by free-growing tall Erythrina
poeppigiana trees. The initial planting density for coffee was 6,300 plants ha-1
, with a current
age >30 years, 20% canopy openness and 2.5 m canopy height. It is intensively managed and
selectively pruned (20% per year, around March). Shade trees have a density of 12.8 trees ha-
1, with 12.3% canopy cover and 20 m canopy height. The experimental basin has an area of
0.9 km2, an elevation range from 1,020 up to 1,280 m.a.s.l. and a mean slope of 20%.
Permanent streams extend along 5.6 km, implying a drainage density of 6.2 km km-2
. The
average slope of the main stream is 11%.
According to the classification by Mora-Chinchilla (2000), the experimental basin is located
along a 1.3 km wide strip of volcanic avalanche deposits, characterized by chaotic deposits of
blocks immersed in a matrix of medium-to-coarse sand, which is the product of the collapse
of the south-eastern slope of Turrialba volcano’s ancient crater. The general classification
given by the geological map of Costa Rica (MINAE-RECOPE, 1991) describes the general
stratigraphy as shallow intrusive volcanic rocks, and the particular region as proximal facies
of modern volcanic rocks (Quaternary), with presence of lava flows, agglomerates, lahars and
ashes. Soils belong to the order of andisols according to the USDA soil taxonomy, which are
soils developing from volcanic ejecta, under weathering and mineral transformation
processes, very stable, with high organic matter content and biological activity and very large
infiltration capacities.
A network of unpaved roads is present in the basin, amounting to a total length of 9.9 km and
with an average width between 3.5 and 4.5 m. Then, roads comprise an area of around 4.5%
that of the entire basin.
The climate is tropical humid with no dry season and strongly influenced by the climatic
conditions in the Caribbean hillside, according to Köppen-Geiger classification (Peel et al.,
181
2007). The mean annual rainfall in the study region for the period 1973-2009 was estimated
as 3014 mm at the Aquiares farm station. At the experimental basin the rainfall in the one-
year measurement period (2849 mm) was close to the annual mean, but showed a monthly
deviation of ±100 mm around the historical regime. Mean monthly net radiation ranged in
2009 from 5.7 to 13.0 MJ m-2
d-1
, air temperature from 17.0 to 20.8 °C, relative humidity
from 83 to 91%, windspeed at 2 m high from 0.4 to 1.6 m s-1
and potential evapotranspiration
(Allen et al., 1998) from 1.7 to 3.8 mm d-1
.
7.3 Materials and Methods
7.3.1 The experimental setup
Rainfall (R): it was monitored at 3 m above ground in the middle of 3 profiles of the basin and
in between the two experimental shaded and non-shaded plots, using three lab-calibrated
ARG100 tipping-bucket (R.M. Young, MI, USA) connected to CR800 dataloggers (Campbell
Scientific, Shepshed, UK), and integrated every 10 min. Other climate variables were logged
on top of an eddy-flux tower (Gómez-Delgado et al., 2011a).
Streamflow (Q): A long-throated steel flume (length: 3.9 m; width: 2.8 m; height: 1.2 m) was
home-built to measure the streamflow at the outlet of the experimental basin, with recording
capacity up to 3 m3 s
-1, the maximum estimated discharge for the study period from an
intensity-duration-frequency analysis. The flume was equipped with a PDCR-1830 pressure
transducer (Campbell Scientific) to record water head at gauge point (20 s, 10 and 30 min
integration), while the rating curve was calculated considering the geometric and hydraulic
properties of the flume using WinFlume software (Wahl et al., 2000). A validation of the
rating curve was made successfully using the salt dilution method as well as a pygmy current
meter. The water balance closure of the basin was reported by Gómez-Delgado et al. (2011a),
indicating that 89% of the cumulated rainfall was explained by the sum of streamflow (flume)
and evapotranspiration (eddy-covariance tower). This is suggesting that most of the surface,
subsurface and baseflow generated by the basin, together with sediment, was actually
captured at the level of this flume.
182
7.3.2 Data set and hydrological models
This work is based on the data of a previous study (Gómez-Delgado et al., 2011b). The
streamflow was measured during a one-year period (07/03/2009 - 06/03/2010) at the outlet of
the Aquiares experimental basin, at 30 minutes time step.
Four alternative models have been developed and calibrated in order to compare different
ways of conceptualizing this coffee agroforestry basin. Figure 47 presents them in the next
order:
(1) Model M1: is a simple two-reservoir model (with one surface-regulating compartment),
calibrated over 7 parameters and explained in detail in Moussa et al. (2007a) and Moussa and
Chahinian (2009). (Fig. 47a).
(2) Model M2: is the five-reservoir Hydro-SVAT model based on 10 parameters, developed
and explained in detail in Gómez-Delgado et al. (2011a) (Fig. 47b).
(3) Model M3: is a simplification of M2 through a sensitivity analysis, reducing the number of
parameters to optimize to 7, but keeping the five-reservoir model structure. A complete
explanation of this model can be found in Gómez-Delgado et al. (2011a) (Fig. 47b).
(4) Model M4: is a model based on M3, but including the effect of the road network in the
generation of surface runoff, as explained in the next section (7.3.3). It has 8 parameters to
optimize. (Fig. 47c).
Studying these four schemes we pretend to identify the effects of both, the model structure
and the parameterization strategy, on the modelled streamflow. The latter is addressed by
varying the amount of optimization parameters (from 7 to 10) while the former looks at three
model structures (2 and 5 reservoirs, and the addition of a road interception component).
7.3.3 The road!interception routine
This routine was developed to be coupled to the simplified Hydro-SVAT model M3 described
above. It is based on the hypothesis that a much larger surface runoff fraction is observed on
the unpaved roads of the experimental basin (Fig. 10c) than on the hillslopes. Then, the
percentage of the rainfall running-off on roads is calculated as:
183
Fig
ure
47
. F
ou
r d
iffe
ren
t m
od
el s
tru
ctu
res
to t
est:
(a)
sim
ple
mo
del
: M
1,
(b)
Hy
dro
-SV
AT
mo
del
in
tw
o d
iffe
ren
t v
ersi
on
s: M
2 (
10
par
amet
ers)
an
d M
3 (
7
par
amet
ers)
, an
d (
c) H
yd
ro-S
VA
T w
ith
ro
ad-i
nte
rcep
tio
n m
od
el:
M4.
184
! ! !tA
AtRtIn
b
rr "# (69)
where Inr(t): road interception during the time step t [L], R(t): rainfall during the time step t
[L], Ar: area of roads [L2], Ab: basin area [L
2] and (t): road-runoff factor at time t
[dimensionless].
The road-runoff factor (t) has been estimated by some authors from field measurements. For
instance, Rijsdijk et al. (2007b) reported this fraction as 65% of the rainfall, while Ziegler et
al. (2004) estimated this value above 80% for unpaved roads in Indonesia and Thailand,
respectively. The new model that we propose is based on the assumption that this fraction
varies in proportion to rainfall intensity, so that the more intense the rainfall, the higher the
fraction of it falling on roads that is converted in surface runoff. Hence, the next quadratic
model was designed to calculate the road-runoff factor:
!
! !
!$$
%
$$
&
'
(
)
#"
r
m
r
m
m
r
k
RtRif
k
RtRif
R
tRk
t2
2
2
2
1
(70)
where kr: calibration factor [dimensionless] and Rm: the maximum rainfall that fell during a
time step t, in the entire modelling period [L].
The calculation of this factor is illustrated by Fig. 48. The principle is that within a given
range, the factor (t) is a quadratic function of rainfall intensity R(t), up to certain threshold
beyond which all rainfall falling on roads is converted to surface runoff.
Figure 48. Road-runoff factor as a
function of the rainfal intensity.
185
Then, the interception by roads Inr is subtracted from the incident rainfall, which is then used
as input to the traditional Hydro-SVAT model. The Inr component is then added to the surface
runoff component of the model that is later translated to the basin outlet by the routing
procedure explained in Gómez-Delgado et al. (2011a).
7.3.4 Model calibration procedure
The same calibration procedure was used for the four models tested. The Nelder-Mead
simplex algorithm (Nelder and Mead, 1965) was applied as described in Lagarias et al.
(1998), starting from very similar sets of initial parameter values for each model as presented
in Table 10. This table was built trying to match all analogue or conceptually equivalent
parameters, and then initializing them under a “directed starting point approach”, that aims to
produce coherent interrelations between the model at all times. This is a direct search method
for multidimensional unconstrained minimization.
Table 10. Initial parameter values for the calibration of the four models tested.
Model Model Parameter description
Param. M1 Param. M2 M3 M4
Maximum storage in the soil-reservoir (m) Sm 0.05 - - - -
Coefficient defining threshold in the soil reservoir above
which infilltration reaches field capacity (dimensionless)
CSm 0.25 - - - -
Calibration factor for road-runoff factor (dimensionless) - - kr - - 20
Vertical/lateral split coefficient to divide outputs from
non-saturated reservoirs into non-saturated runoff and
vertical flows (dimensionless)
0.234 0.234 0.234 0.234
Discharge rate for surface reservoir (s-1) - - kB 1×10-3 1×10-3 1×10-3
Infiltration rate at field capacity (m s-1) fc 1×10-5fc 1×10-5 - -
Coefficient to calculate the maximum infiltration capacity
from field capacity (dimensionless)
Cfc 10 ! 10 - -
Discharge coefficient, total outputs from non-saturated
root reservoir (s-1)
- - kC 1×10-4 1×10-4 1×10-4
Discharge coefficient for total outputs from non-saturated
non-root reservoir (s-1)
- - kD 1×10-6 - -
Threshold level in the aquifer reservoir, above which a
shallow-aquifer outlet is found (m)
- - EX 0.5 0.5 0.5
Discharge coefficient for baseflow from shallow aquifer
reservoir (s-1) - - kE2 1×10-6 1×10-6 1×10-6
Discharge coefficient for baseflow from deep aquifer
reservoir (s-1)
k 1×10-7 kE1 1×10-7 1×10-7 1×10-7
Discharge coefficient for deep percolation from the
aquifer reservoir (s-1)
kE3 1×10-8 kE3 1×10-8 1×10-8 1×10-8
186
The value 1-NS, where NS is the Nash-Sutcliffe efficiency coefficient (Nash and Sutcliffe,
1970) was the objective function to minimize, and termination tolerances equal to 1×10-4
were
set for both, the function value and the estimated local (multivariate) minimum. For
calibration purposes, the Nash-Sutcliffe criterion was evaluated on a continuous time basis
(i.e., over the entire record of measured and modelled time series).
7.3.5 Model evaluation procedure
Two types of analyses of model performance were made: evaluation on continuous time, and
evaluation on storm events. The first type is calculated over the measured and modelled time
series for all the time steps (which in our particular case amounts to 17520 time steps at 30
min resolution), while the second type is obtained only over those time steps belonging to 24
main storm events that were identified in a previous work (Gómez-Delgado et al., 2011b) and
for the same data set (amounting to 575 time steps for this study).
A. Evaluation on continuous time
The model evaluation on continuous time is based on the next four criteria:
(1) Nash-Sutcliffe efficiency coefficient (NS) [dimensionless] (Nash and Sutcliffe, 1970),
which measures the overall agreement of the shape of measured and observed hydrographs:
! !* +
!* +,
,
#
#
-
--#
T
t
Tmea
T
t
modmea
QtQ
tQtQ
1
2
1
2
1NS (71)
where Qmea(t) [L3 T
-1] is the measured streamflow at the time step t [T], Qmod(t) [L
3 T
-1] is the
modelled streamflow at the time step t and TQ [L3 T
-1] is the mean measured streamflow
during the evaluation period T [T]. The NS is a normalized measure in the interval ]-!,1] that
compares the error generated by a particular model with the errors of the benchmark model
(denoted TQ ) that uses as its prediction the (constant) mean value of the observed target
( ! TQtQ # ). This means that: NS equal to 1 indicates perfect model performance (Qmod(t) =
Qmea(t) for all t); NS equal to 0 indicates that the model is, on average, performing only as
well as the mean target value used for prediction; and negative NS indicates an altogether
questionable model choice.
187
(2) Bounded NS efficiency coefficient (C2M) [dimensionless] (Mathevet, 2005), also assessing
the agreement in the shapes of the hydrographs:
! !* +
!* +
! !* +
!* +
1
1
2
1
2
1
2
1
2
2M 11C
-
#
#
#
#
$$.
$$/
0
$$%
$$&
'
-
-1
$$.
$$/
0
$$%
$$&
'
-
--#
,
,
,
,T
t
Tmea
T
t
modmea
T
t
Tmea
T
t
modmea
QtQ
tQtQ
QtQ
tQtQ
(72)
The C2M is a normalized measure related to the NS, but varies in the interval ]-1, 1]. For NS =
1 we have C2M = 1, for NS = 0 we have C2M = 0 and for NS = -! we have C2M = -1.
(3) Global volume error (Vg) [L3] (Moussa and Chahinian, 2009), which quantifies the
average difference between measured and modelled streamflow volumes:
! !, ,# #
-#T
t
T
t
modmea dttQdttQ1 1
gV (73)
where dt [T] is the time step.
(4) Relative volume error (Vr) [dimensionless] (Moussa and Chahinian, 2009), also measuring
the volumetric difference in streamflow, but in relative terms:
! !
!,
,,
#
##
-#
T
t
mea
T
t
mod
T
t
mea
dttQ
dttQdttQ
1
11
rV (74)
B. Evaluation on storm events
The next six criteria were evaluated on a storm event basis:
(1) Event-based NS coefficient NS(i) [dimensionless] the Eq. 71 is applied to each of the i =
1, 2, …, N individual storms, producing an event-series that contains N values of NS(i). The
mean of these NS(i) coefficients over the N events (____
NS) is an overall indicator of the model
efficiency to reproduce the storm events for the modelling period:
!
N
iN
i
,## 1
NS
NS (75)
188
(2) Event-based bounded NS coefficient C2M(i) [dimensionless]: the application of Eq. 72 to
each of the storm events produces an event-series of N values of C2M(i). We define 2MC as the
mean value of C2M(i):
!
N
iN
i
,## 1
2M
2M
C
C (76)
Considering that the C2M(i) is a bounded function varying between -1 and 1 (Mathevet, 2005),
its mean over the event-series ( 2MC ) is considered a better overall indicator of storm-based
model efficiency along the modelling period with respect to the mean of NS(i), which is
unbounded on the negative side. Other techniques like the comparison of event-based
cumulative distribution functions are also better defined for the C2M(i).
(3) Event-based global volume error Vg(i) [L3]: it is obtained by evaluating the Eq. 73 for
each of the N storm events. An overall storm-based indicator for the entire modelling period is
given by the mean ( gV ) of the individual global volume errors.
!
N
iN
i
,## 1
g
g
V
V (77)
(4) Event-based relative volume error Vr(i) [dimensionless]: it is the result of the application
of Eq. 74 to each storm event under analysis. The mean ( rV ) of the N relative volume errors
in the event-series is a performance indicator of the model with respect to the storm
streamflow volume during the modelling period.
!
N
iN
i
,## 1
r
r
V
V (78)
(5) Global peakflow noted Pg(i) [L3 T
-1] (Moussa and Chahinian, 2009), which measures the
agreement between the measured and modelled peakflow rates:
! ! !iQiQi mod,pmea,p -#gP (79)
where Qp,mea(i) and Qp,mod(i) are respectively the measured and modelled streamflow peak-
rates during the storm event i. The mean ( gP ) of the N individual Pg(i) values provides an
189
overall absolute indicator of the accuracy of the model to reproduce the measured peakflows
over the modelling period.
!
N
iN
i
,## 1
g
g
P
P (80)
(6) Relative peakflow Pr(i) [dimensionless] (Moussa and Chahinian, 2009), quantifying the
relative difference in peakflows:
! ! !
!iQ
iQiQi
mea,p
mod,pmea,p -#rP (81)
The mean ( rP ) of the N storm values of Pr(i) provides the overall indicator of the relative
accuracy of the model to reproduce peakflows over the modelling period.
!
N
iN
i
,## 1
r
r
P
P (82)
7.4 Results and discussion
This section is subdivided into three subsections, addressing first the model performance on
continuous time, second the performance on storm events, and finally the partition of
streamflow according to the each tested model.
7.4.1 Model performance on continuous time
Once an optimal set of parameters was defined by the convergence of the Nelder-Mead
algorithm (usually between 3000 and 4000 model runs), the measured and modelled
streamflows were compared by calculating the four evaluation criteria given in the section
7.3.5A, whose results are given in Table 11. The model with the poorer performance
reproducing the shape of the hydrograph was M3 (the Hydro-SVAT simplified through
sensitivity analysis) with a NS = 0.76, C2M = 0.61 (see Fig. 49c). For every model, the
volumetric errors were around 1% or less, which makes useless any attempt to improve this.
The corresponded volumetric errors (Vg = 0.99 m3 and Vr = 1.1%) were only slightly minor
than those of M1 (the model with the simplest structure), which produce almost the same
190
precision in the hydrograph shape (Fig. 49a). From M3, the introduction of a simple road-
interception routine in M4 (at the cost of introducing one more parameter to calibrate)
increased the predictive capacity of the model and produced noticeably better shape
performances through the overall time series, as well as the minimum volumetric errors
between all models (Fig. 49d). This suggests the possibility that the hypotheses about road-
interception are correct, and that taking them into account has a remarkable impact on the
water balance estimations.
Table 11 Performance of the four models tested on continuous time. NS: Nash-
Sutcliffe efficiency coefficient, C2M: bounded Nash-Sutcliffe, Vg: global volume
error, Vr: relative volume error.
Model Evaluation criterion
No. Reserv. No. Param. NS C2M Vg (m3) Vr (%)
M1 3 7 0.77 0.63 1.10 1.2
M2 5 10 0.86 0.76 0.70 0.81
M3 5 7 0.76 0.61 0.99 1.1
M4 5 8 0.83 0.71 0.24 0.28
Finally, the original Hydro-SVAT model M2 (ten parameters) made the best approach to the
measured hydrograph shape (Fig. 49b), producing the highest NS = 0.86 and C2M = 0.76 and
the slope closest to 1 when comparing measured and modelled streamflows (right side of the
figure).
From the typical display of model results as time-series, given in Fig. 49 (left side), some
general conclusions can be drawn. For instance, it seems that the lowest part of the recessions
have not been properly represented by any of the models (or else, that the measurements have
a bias in this streamflow range). This is especially clear in the driest months (January-
February 2010), where all Hydro-SVAT models were clearly inaccurate (M2, M3 and M4) ,
suggesting that the highest discharge path of in the aquifer reservoir (presented as QE2 in Fig.
47b,c) might be unnecessary. In contrast, the simple model M1 appears to be ineffective to
model mid-range streamflow values (0.02 - 0.05 m3 s
-1). Despite of these shortcomings,
aquifer recharge/discharge processes that are of primary importance in this experimental basin
(Gómez-Delgado et al., 2011a) seem to be well represented by all the modelling schemes.
191
Figure 49. Performance of the four models to predict streamflow (displayed in logarithmic
scale) on continuous time, both as time series (left side) and as the measured-modelled plot
(right side), for: (a) M1: simple model, (b) M2: Hydro-SVAT model 10 parameters, (c) M3:
Hydro-SVAT model 7 parameters and (d) M4: Hydro-SVAT with road-interception routine.
a)
b)
c)
d)
192
The measured-modelled plots (right side) reflect that all the models tend to underestimate the
actual streamflow, although M2 is, again, clearly superior to the other models.
The cumulative distribution functions of the measured and modelled streamflow are presented
in Fig. 50. While the Hydro-SVAT M2 follows the measured streamflow distribution in most
of its ranges, only over-performing on the highest values (the peakflows), M1 seems to stay
away from it in all ranges, over-performing the mid-range values and under-performing the
extremes (check the zoomed boxes).
Figure 50. Performance of the four models tested on continuous time, in comparison to the measured
streamflow, as a cumulative distribution function.
193
7.4.2 Model performance on storm events
The same streamflow time series from the four presented models were compared to the
observed one, using the six evaluation criteria given in section 7.3.5B. This analysis aims to
characterize the ability of the models to reproduce the peakflows, even if the calibration
criterion was targeting the overall hydrograph shape, also including recession periods. The six
validation criteria were evaluated over 24 individual storm events previously identified
(Gómez-Delgado et al., 2011b) and the results are displayed in Fig. 51. The NS and C2M
coefficients (Fig. 51a,b) systematically show that models M2 and M4 reproduce peakflow
shapes in a more accurate way than the other two models and reaching maximum efficiencies
of NS=0.99 and C2M=0.97 (model M2, storm event #8). The mean values over the 24 storms
are provided in Table 12. The fact that the NS coefficient (mean of NS(i), i = 1, 2,…, 24
events) is higher for M1 than for M4 (against the visual evidence observed in Fig. 51a) reveals
the flaw of using the NS index to calculate summary measures on groups of individual events.
In this particular case, since M4 produced one largely negative NS value (event #20, NS(20) =
-5.3), the respective mean was reduced from the more representative value of NS = 0.44
(over 23 events) to NS = 0.20 (now including the event #20). By contrast, the bounded NS
exhibited a minimum C2M = -0.73 (event #20), which affected to a much lesser extent the
overall mean ( 2MC = 0.40). Then, 2MC outperforms NS , making clear that M4 reproduces the
hydrograph shape in a better fashion than M1 and confirming the trends in Fig. 51a,b.
The NS(i) and C2M(i) are clearly correlated (Pearson correlation coefficient = 0.89), and
similar patterns were also observed between global and relative volumes (correlation = 0.89)
and between global and relative peakflow rates (correlation = 0.76). Although the correlations
between the volume and peakflow measures are not significantly different from zero (two-
tailed, 0.01 level), they do not have a clear relationship. In contrast, a significant correlation
was detected between C2M(i) and Vg(i) and between C2M(i) and Vr(i), which means that the
overall shape of the hydrograph is linked to the volumetric performance of the models, and
great attention must be paid to any weighting methods involving these variables, in the frame
of multi-objective calibration procedures. Even more, when rank-order association measures
are tested (here the Spearman’s rho), almost all global and relative criteria have inter-
correlations significantly different from zero.
194
2
1
0
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24NS
M1
M2
M3
M4
1
0.5
0
0.5
1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
C2
M
M1
M2
M3
M4
0
50
100
150
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Vg
M1
M2
M3
M4
0
0.25
0.5
0.75
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Vr
M1
M2
M3
M4
0
0.25
0.5
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Pg
M1
M2
M3
M4
0
0.25
0.5
0.75
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Pr
M1
M2
M3
M4
Figure 51. Model performance on storm events, according to six criteria: (a) Nash-Sutcliffe NS, (b)
bounded Nash-Sutcliffe C2M, (c) global volume error Vg, (d) relative volume error Vr, (e) global peak
error Pg, (f) relative peak error Pr.
a)
b)
c)
d)
e)
f)
Event number
195
Table 12. Mean values of the model performances evaluated on 24 storm events,
according to six evaluation criteria. NS : event-mean Nash-Sutcliffe efficienty
coefficient, 2MC :event-mean bounded Nash-Sutcliffe, gV: event-mean global
volume error, rV : event-mean relative volume error, gP : event-mean global peak
error and rP : event-mean relative peak error.
Evaluation criterion
Model #Res #Par NS 2MC gV (m3) rV (%) gP (m3 s-1) rP (%)
M1 3 7 0.32 0.33 37.1 18 0.13 0.32
M2 5 10 0.55 0.52 29.2 15 0.07 0.19
M3 5 7 -0.02 0.19 48.3 23 0.11 0.32
M4 5 8 0.20 0.40 42.7 19 0.08 0.24
The two volume criteria presented the same coefficient of variation (CV = 0.84), suggesting
that any of these measures (global or relative) could be used indifferently to evaluate the
volume performance of the models. In the case of peakflow assessments, the global method
presented higher variability (CV = 0.98) than the relative (CV = 0.70). Low NS and C2M
coefficients, and large volume and peak-rate errors were detected in 4 out of these 24 storm
events (with the numbers 16, 19, 20 and 23). An additional analysis of these events could help
to detect deficiencies in the models and/or in the measured streamflow records.
A boxplot representation of the event-based model evaluations is shown in Fig. 52. It can be
seen at a glance that M2 and M4 have the highest efficiencies (Fig. 52a,b) and the lowest peak-
rate errors (Fig. 52e,f), while M1 and M2 best reproduce the observed streamflow volumes
(Fig. 52c,d). In order to assess to what extent the summary measure that we have proposed to
assess the overall model performance (the mean of the individual storms) is affected by
outlier events, this value has been added to the boxplots. Then, in addition to the already
discussed incoherence between the mean (given by the × mark) and the median values (the
thick line inside the box) affecting the NS criterion of M4 (caused by one highly negative
outlier, not displayed), and properly overcome by C2M, similar ranking problems are face to
classify M1 according to Vg and Vr criteria. Pg and Pr seem to be less sensitive to the
appearance of extreme values.
196
-1.0
0.0
1.0
Model
NS
1 2 3 4
-1.0
0.0
1.0
Model
C2M
1 2 3 4
040
80
120
Model
Vg
1 2 3 4
0.0
0.2
0.4
0.6
Model
Vr
1 2 3 4
0.00
0.15
0.30
Model
Pg
1 2 3 4
0.0
0.4
0.8
Model
Pr
1 2 3 4
Figure 52. Performance boxplots according to the four models tested on storm events for: (a) Nash-Sutcliffe: NS(i), (b) bounded Nash-Sutcliffe: C2M(i), (c) global volume error: Vg(i), (d) relative volume error: Vr(i), (e) global peakflow rate error: Pg(i), (f) relative peakflow rate error: Pr(i). Some outliers are not shown given the chosen range for the vertical axis. The box size represents the interquartile range, the whiskers extend to the most extreme data point, which is no more than one interquartile range, circles are values beyond whiskers limits. The × mark indicates the mean value.
From this analysis for summary performance measures, three main ideas can be stated. First,
that hydrograph shape performances are better assessed by the C2M criteria, which is less
sensitive to outliers than the NS. In addition, volume measures (global and relative) seem
completely redundant, while some differences were noticed in peak-rate measures. Second,
that all the measures here evaluated (and commonly used for calibration and validation in
hydrological modelling) can be highly correlated and should not be considered as independent
a) b)
c) d)
e) f)
197
when multi-objective calibration is carried out. Third, that a mean/median comparison can be
useful to assess the quality of the mean as an event-based summary measure.
The use of performance cumulative distributions (Fig. 53) is proposed as an alternative
method for model comparison, providing an outlook of models behaviour throughout the
storm events.
2
1
0
1
0 0.5 1Cumulative frecuency
NS
M1M2M3M4
1
0
1
0 0.5 1Cumulative frecuency
C2M
M1M2M3M4
0
20
40
60
80
100
120
140
0 0.5 1Cumulative frecuency
Vg
M1M2M3M4
0
0.4
0.8
0 0.5 1Cumulative frecuency
Vr
M1M2M3M4
0
0.2
0.4
0 0.5 1Cumulative frecuency
Pg
M1M2M3M4
0
0.4
0.8
0 0.5 1Cumulative!frecuency
Pr
M1M2M3M4
Figure 53. Performance cumulative distributions for the four models tested on storm events: (a) Nash-Sutcliffe coefficient NS(i), (b) bounded Nash-Sutcliffe C2M(i), (c) global volume error Vg(i), (d) relative volume error Vr(i) (e) global peak error Pg(i)and (f) relative peak error Pr(i). The horizontal scale refers to the cumulative frequency of storm events: the frequency 0.5 indicates a half of the events, while 1 represents all them.
a) b)
c) d)
e) f)
198
Hence, C2M (Fig. 53b) provides a clearer vision of the performance distributions in
comparison to NS (Fig. 53a), and shows the instability of that half of the events with the
lowest C2M values. M2 is definitely the model that performs best the hydrograph shapes, while
M4 behaves similarly but is much less precise in the most “complex” events (those with
frequencies between 0 and 1). Both global and relative volume curves (Fig. 53c,d) basically
provide the same information, indicating that M1 produces the highest errors and M4 the
lowest. In contrast, global and relative peak curves (Fig. 53e,f) behave differently, the latter
one indicating that the models with lowest overall errors (M2 and M4) also produced the
highest relative peak errors.
7.4.3 Streamflow partition
A final analysis presented in Fig. 54 consists in a visual inspection of the modelled
hydrographs for some storm events, and the comparison of the streamflow partition (surface
runoff/baseflow) made by the models, with respect to a manual partition. The first four sub-
figures (a, b, c, d) describe the storm event occurred on 24/06/2009, according to the four
studied models, some of them with very good performances. It seems clear that M2 and M4
(Fig. 54b,d) approached the observed streamflow in a much more accurate fashion than M1
and M2, and that the three chosen storm event measures (NSi(i), Pr(i), Vr(i)) coincide with the
visual information in the graphs. The baseflow that was extracted from the measured
streamflow trough the constant slope method (details in Gómez-Delgado et al., 2011b) is
indicated by the gray dotted line. Then, for this first storm event only the original Hydro-
SVAT (M2) and (in a minor degree) its simplified version M3 simulated with some success
the aquifer recharge process, and then, the baseflow component. In addition, M2 also
reproduced well the shape of the falling limb. A less successful simulation was achieved in
Fig. 54e,f,g,h; corresponding to the storm event of 02/11/2009. Visual inspection clearly
indicates an important lack of accuracy, the NS coefficients are lower than those of the
previous event, while relative peak-rate errors were notably higher (and similar for all the four
models). The fact that the volume errors are smaller than those in the previously studied event
is misleading, because in this particular case the general trend of all models to overestimate
the streamflow (demonstrated in Fig. 49) is fairly well compensated by the volume deficiency
produced by the shortage in the reproduction of the peak-rate. Again, M2 was the only model
producing comparable baseflow to those estimated in the measured time series.
199
0.0
0.1
0.2
0.3
0.4
16:00 18:30 21:00 23:30 02:00
Q (
m3 s
-1)
Q!mea
Baseflow
QBF+SR!M1
QBF!M1
0.0
0.1
0.2
0.3
0.4
16:00 18:30 21:00 23:30 02:00
Q (
m3 s
-1)
Q!mea
Baseflow
QBF+SR!M2
QBF!M2
0.0
0.1
0.2
0.3
0.4
16:00 18:30 21:00 23:30 02:00
Q (
m3 s
-1)
Q!mea
Baseflow
QBF+SR!M3
QBF!M3
0.0
0.1
0.2
0.3
0.4
16:00 18:30 21:00 23:30 02:00
Q (
m3 s
-1)
Q!mea
Baseflow
QBF+SR!M4
QBF!M4
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
11:00 14:30 18:00 21:30 01:00
Q (
m3 s
-1)
Q!mea
Baseflow
QBF+SR!M1
QBF!M1
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
11:00 14:30 18:00 21:30 01:00
Q (
m3 s
-1)
Q!mea
Baseflow
QBF+SR!M2
QBF!M2
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
11:00 14:30 18:00 21:30 01:00
Q (
m3 s
-1)
Q!mea
Baseflow
QBF+SR!M3
QBF!M3
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
11:00 14:30 18:00 21:30 01:00
Q (
m3 s
-1)
Q!mea
Baseflow
QBF+SR!M4
QBF!M4
Figure 54. Modelled streamflow and its partition over two representative storm events. Event of 24/06/2009: (a) M1: simple model, (b) M2: Hydro-SVAT model 10 parameters, (c) M3: Hydro-SVAT model 7 parameters, and (d) M4: Hydro-SVAT with road interception. Event of 02/11/2009: (e) M1, (f) M2, (g) M3, (h) M4. Qmea: measured streamflow; Baseflow: manually separated baseflow; QBF: modelled baseflow; QSR: modelled surface runoff; QBF + QSR: total modelled streamflow.
a)
c)
e)
b)
d)
f)
g) h)
NS = 0.80 Pr = 0.25 Vr = 0.12
NS = 0.93 Pr = 0.13 Vr = 0.02
NS = 0.71 Pr = 0.24 Vr = 0.10
NS = 0.90 Pr = 0.09 Vr = 0.11
NS = 0.61 Pr = 0.53 Vr = 0.08
NS = 0.73 Pr = 0.44 Vr = 0.03
NS = 0.51 Pr = 0.51 Vr = 0.06
NS = 0.75 Pr = 0.44 Vr = 0.00
200
7.5 Conclusions
This study aimed to improve an existent eco-hydrological model adapted for several purposes.
Four topics were treated: i) the role of roads on runoff genesis, given that they can directly
impact peakflows and, consequently, sediment transfer in farmed coffee-basins; ii) the effect
of the model structure and parameterization procedures (like model simplification) on its
flexibility and efficiency (iii) the application of a “directed-calibration” approach; and (iv) the
employ of multi-criteria validation techniques at both continuous-time and storm event
periods.
7.5.1 On the effect of roads in model performance
Once we adapt the structure of Hydro-SVAT as a semi-distributed model, seeking an accurate
simulation of peakflows over storm events, and taking into account the two sources of surface
runoff (plots and roads), the analysis showed that the sole addition of a road interception
routine for improving the simulations of surface runoff did not significantly improve the
performances of the model proposed by Gómez-Delgado et al. (2011a). Two proposals can be
made to refine this further: to build M4 (semi-distributed) directly from M2 (original Hydro-
SVAT) in order to keep the good performances of M2 during recessions after peakflows and
its proper emulation of the aquifer behaviour. A second possibility is to make variable the
percentage of impermeable areas (roads plus footpaths, having the latter unknown
dimensions), maintaining the routine that we have developed to take into account the effect of
rainfall intensity.
When a successful peakflow routine is developed, the next step will be the development of a
sediment yield routine to couple into Hydro-SVAT.
7.5.2 On the model structure and the reduction of parameters
The study of four models lead us to an unexpected conclusion: the best model performances
on streamflow were achieved by the model that was created first: the 10 parameter Hydro-
SVAT (M2). The simple three-reservoir model used as reference (M1) and the simplified
version of M2 (M3) presented limited conditions to reproduce the high resolution (30 minutes
time step) streamflow series employed in this study. These results confirm that simple models
are not always as appropriate as complex ones, especially if several properties of complex
201
systems want to be predicted (not only the streamflow hydrograph shape). Standard
assessments considering the relationship between model efficiency and number of parameters
can provide useful information in the model evaluation process.
The parameter reduction analysis developed to produce a simplified model M3 out of the
original Hydro-SVAT (M2), and based on a sensitivity analysis was successful when M3 was
evaluated using the original data set used for the evaluation of the sensitivity (Gómez-
Delgado et al., 2011a), but produced poor results in the calibration under the new conditions
of this work (the study period is different). This seems to indicate that a model simplified
according to a sensitivity analysis can be successful under the particular conditions, but may
reduce model capabilities that could be required under different conditions. Better results may
be obtained if this procedure is applied using a split-sample approach.
7.5.3 On the “directed” model calibration
The proposition of a “directed starting point” method for single-objective calibration (based
on the overall NS coefficient) seemed to work appropriately, effectively driving the
calibration runs of all the models to remain within conceptually feasible parametric spaces.
Because of this, model comparisons were greatly facilitated. The potential use of multi-start
algorithms, all with feasible and coherent between models starting values, could be tested in
order to produce more general results.
7.5.4 On the two time scale (continuous!time, storm event) multi!criteria
validation
It is common for hydrologists to model continuous long term periods for water resources
management, and to model individual storm events for flood forecasting and protection
schemes, or for simulating sediment and pollutant transfer. The first important challenge that
awaits the modeller in this task is to develop or to choose a rainfall-streamflow model, and to
calibrate a set of parameters that can simulate accurately and simultaneously a continuous
period, but also a number of storm events.
We have proposed the revision of the model performance at both, the continuous-time and the
storm event periods. This approach appeared very useful to have overall validations of both
the hydrograph reproduction, and the model performance during storm events. The method
202
can be especially useful when peakflow properties and effects (as the mentioned above) in
addition to the usual water balance properties, require to be investigated through a modelling
work.
Concerning the multi-criteria approach, it was evidenced that the hydrograph shape efficiency
measure C2M (the bounded NS coefficient) behaved in a much more stable way than the
classical NS. It was also much more efficient to calculate summary statistics (e.g. the mean of
a group of storm events), which extend its capabilities from the use for which it was created,
i.e., the summary analysis of models’ performance in large amounts of basins (Mathevet,
2005). However, the main finding dealing with the criteria studied here (especially the NS
coefficient and its bounded version, as well as the global and relative volume errors) was the
evident lack of independence. Although global and relative peak-errors were slightly less
correlated, they are not completely independent. These results represent a serious problem
when these criteria are used as objective functions in multi-objective model calibration
procedures that rely on weighting routines. The use of principal component analysis could be
useful for better approaching the theoretical conditions that those techniques demand.
References
Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration: Guidelines for
computing crop water requirements. Food and Agriculture Organization, Rome.
Anderson, M.P., Woessner, W.W., 1992. The role of the postaudit in model validation.
Advances in Water Resources 15, 167-173.
Beldring, S., 2002. Multi-criteria validation of a precipitation-runoff model. Journal of
Hydrology 257, 189-211.
Beven, K., 1993. Prophecy, reality and uncertainty in distributed hydrological modelling.
Advances in Water Resources 16, 41-51.
Beven, K., 2006. A manifesto for the equifinality thesis. Journal of Hydrology 320, 18-36.
Diskin, M.H., Simon, E., 1977. A procedure for the selection of objective functions for
hydrologic simulation models. Journal of Hydrology 34, 129-149.
203
Duan, Q., Sorooshian, S., Gupta, V., 1992. Effective and efficient global optimization for
conceptual rainfall-runoff models. Water Resour. Res. 28, 1015-1031.
Efstratiadis, A., Koutsoyiannis, D., 2010. One decade of multi-objective calibration
approaches in hydrological modelling: a review. Hydrological Sciences Journal 55, 58-78.
Gómez-Delgado, F., Roupsard, O., le Maire, G., Taugourdeau, S., Pérez, A., van Oijen, M.,
Vaast, P., Rapidel, B., Harmand, J.M., Voltz, M., Bonnefond, J.M., Imbach, P., Moussa, R.,
2011a. Modelling the hydrological behaviour of a coffee agroforestry basin in Costa Rica.
Hydrol. Earth Syst. Sci. 15, 369-392.
Gómez-Delgado, F., Roupsard, O., Moussa, R., 2011b. Water and sediment yield in a coffee
agroforestry system at various spatio-temporal scales: from plot to basin and from event to
annual scales. Agriculture, Ecosystems & Environment, in prep.
Henriksen, H.J., Troldborg, L., Nyegaard, P., Sonnenborg, T.O., Refsgaard, J.C., Madsen, B.,
2003. Methodology for construction, calibration and validation of a national hydrological
model for Denmark. Journal of Hydrology 280, 52-71.
Lagarias, J.C., Reeds, J.A., Wright, M.H., Wright, P.E., 1998. Convergence Properties of the
Nelder-Mead Simplex Method in Low Dimensions. SIAM Journal on Optimization 9, 112-
147.
Lindström, G., 1997. A Simple Automatic Calibration Routine for the HBV Model. Nordic
Hydrology 28, 153-168.
Madsen, H., 2000. Automatic calibration of a conceptual rainfall-runoff model using multiple
objectives. Journal of Hydrology 235, 276-288.
Mathevet, T., 2005. Quels modèles pluie-debit globaux au pas de temps horaire ?
Développements empiriques et comparaison de modèles sur un large échantillon de bassins
versants. PhD Thesis, ENGREF, Paris, p. 354.
MINAE-RECOPE, 1991. Mapa geológico de Costa Rica, Scale 1:200000. San José, Costa
Rica.
Mora-Chinchilla, R., 2000. Geomorfología de la cuenca del Río Turrialba (unpublished).
Universidad de Costa Rica, San José, Costa Rica.
204
Moussa, R., Chahinian, N., 2009. Comparison of different multi-objective calibration criteria
using a conceptual rainfall-runoff model of flood events. Hydrol. Earth Syst. Sci. 13, 519-535.
Moussa, R., Chahinian, N., Bocquillon, C., 2007. Distributed hydrological modelling of a
Mediterranean mountainous catchment - Model construction and multi-site validation. Journal
of Hydrology 337, 35-51.
Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models part I - A
discussion of principles. Journal of Hydrology 10, 282-290.
Nelder, J.A., Mead, R., 1965. A Simplex Method for Function Minimization. The Computer
Journal 7, 308-313.
Peel, M.C., Finlayson, B.L., McMahon, T.A., 2007. Updated world map of the Köppen-
Geiger climate classification. Hydrol. Earth Syst. Sci. 11, 1633-1644.
Perrin, C., 2000. Vers une amélioration d’un modèle global pluie-débit au travers d’une
approche comparative. PhD Thesis, INPG-Cemagref, Paris.
Perrin, C., Michel, C., Andréassian, V., 2001. Does a large number of parameters enhance
model performance? Comparative assessment of common catchment model structures on 429
catchments. Journal of Hydrology 242, 275-301.
Pickup, G., 1977. Testing the efficiency of algorithms and strategies for automatic calibration
of rainfall-runoff models. Hydrological Sciences Bulletin 22, 257-274.
Refsgaard, J.C., 1997. Parameterisation, calibration and validation of distributed hydrological
models. Journal of Hydrology 198, 69-97.
Rijsdijk, A., Sampurno Bruijnzeel, L.A., Sutoto, C.K., 2007. Runoff and sediment yield from
rural roads, trails and settlements in the upper Konto catchment, East Java, Indonesia.
Geomorphology 87, 28-37.
Sorooshian, S., Gupta, V.K., 1983. Automatic calibration of conceptual rainfall-runoff
models: The question of parameter observability and uniqueness. Water Resour. Res. 19, 260-
268.
205
Wahl, T.L., Clemmens, A.J., Replogle, J.A., Bos, M.G., 2000. WinFlume - Windows-based
software for the design of long-throated measuring flumes. Fourth Decennial National
Irrigation Symposium. American Society of Agricultural Engineers, Phoenix, Arizona.
Willmott, C.J., Ackleson, S.G., Davis, R.E., Feddema, J.J., Klink, K.M., Legates, D.R.,
O'Donnell, J., Rowe, C.M., 1985. Statistics for the Evaluation and Comparison of Models. J.
Geophys. Res. 90, 8995-9005.
Yapo, P.O., Gupta, H.V., Sorooshian, S., 1998. Multi-objective global optimization for
hydrologic models. Journal of Hydrology 204, 83-97.
Ziegler, A.D., Giambelluca, T.W., Sutherland, R.A., Nullet, M.A., Yarnasarn, S., Pinthong, J.,
Preechapanya, P., Jaiaree, S., 2004. Toward understanding the cumulative impacts of roads in
upland agricultural watersheds of northern Thailand. Agriculture, Ecosystems & Environment
104, 145-158.
$
206
207
CHAPTER 8
General conclusions and perspectives
This work intended to study the general context and current state of the Hydrological
Environmental Services (HES) and the hypotheses on which they are being promoted and
implemented in Costa Rica, emphasizing the role of coffee producers as HES providers, and
of hydropower producers as HES users. Then, the study provided an outlook on
ecophysiological, hydrological and sedimentological processes and models, as well as the
correspondent bibliographic review, which is centred on tropical, coffee-based agroforestry
systems. The new Aquiares coffee experimental basin, created to provide the necessary data
to support the analyses of the thesis, was then described in detail, followed by three paper-
based chapters. The first was dealing with the development of a new model called “Hydro-
SVAT” to investigate the eco-hydrological behaviour of the experimental basin and closing
the water balance. The second described the measurements of water and sediment yield in the
basin, comparing its behaviour from plot to basin scale, and during storm events in
comparison to recession times. The third paper (still in process) aimed the improvement of
water and sediment peakflow simulations from the Hydro-SVAT model, by introducing a
road-runoff routine, and being assessed at both, storm-event and continuous-time scales.
8.1 On hydrological environmental services, hydropower and coffee
From the introductory chapter we conclude on the importance that HES have recently gained
in Costa Rica, backed by a particular tradition on sustainable development and having
hydropower and agricultural/coffee production as two main socio-economical activities in the
historical context. A remark was made in the sense that HES can be provided by either
currently existing ecosystems (like pristine forest basins), or by directed measures driven
through Basin Management Programs (BMP). In both cases, objective quantification has to be
done. We have proposed that observational/experimental methods, along with modelling
efforts, constitute the most reliable way to achieve such quantifications, and it can also be the
base for the necessary monitoring programs, followed by general periodical evaluations of the
objectives of any BMP or HES-targeted projects. The inattention of quantitative approaches
208
might lead to the promotion of unfair, myth-based and/or unsustainable socio-economical
agreements for the payments of HES. For the sake of an example, the misuse of the popular
conception that trees are “water factories” (when they should rather be seen as water pumps,
extracting water from the soil to sustain their vital processes) can lead to large replanting
schemes directly competing with the water yield in a hydro-power drainage basin.
Although reforestation or agroforestry practices have proven to be effective to reduce plot
erosion and basin-scale sediment yield under certain conditions, this latter objective
demonstrated to have a complex response to such practices, being critically affected by other
factors besides land use (e.g. climate, soil/lithology, topography, management, infrastructure
such as roads and scale-dominant erosion processes).
8.2 On processes and models: ecophysiology, hydrology and
sedimentology
This chapter (on processes and models) gave the bases to understand the biophysical links
between ecophysiological, hydrological and sedimentological approaches, which are
indispensable to evaluate those HES dealing with water/sediment yields from coffee
agroforestry basins. From the perspective of the provider of HES, most management practices
to implement in agricultural fields interfere with the ecophysiology of the plantation,
impacting major processes on both, the hydrology and the sedimentology patterns (for
instance, by affecting evapotranspiration, infiltration and erosion). But the variables of
interest for most of HES beneficiaries will be perceived and must be studied under the typical
hydrological and sedimentological approaches and scales: water and sediment yield at basin
scale. Then, an equilibrium must be achieved at either the measurement (e.g. variables to
measure, spatio-temporal resolution of the observations) or the modelling stages (e.g.
balanced representation of processes across different scientific fields, proper
conceptualization of their links).
In addition, an important element, source of confusion between researchers working in
different fields, was found to be the operative definition of erosion and sediment yield. While
the first term is commonly employed to identify soil loss processes at the plot scale (which
can be seen at larger scales like a “potential” sheet soil loss), the second term is more
accepted for basin scale assessments (considering the “actual” soil loss from the system after
209
depositional processes and mass wasting). Although area-based upscaling procedures have
been developed to link plot erosion to basin sediment yield, its use should be merely
indicative, as many other critical factors demand the most serious consideration.
8.3 On the bibliographic review on ecophysiological, hydrological and
sedimentological studies, with focus on tropical and coffee!based
systems
From a bibliographic review on ecophysiological and hydrological studies it was observed
that these fields back two main approaches for measuring, modelling and closing a water
balance, the engineering (hydrological) approach aiming to understand surface and return
flow at basin scales, and the agronomical (ecophysiological) approach, more interested in
ecosystem water balance at plot scale and on the effects of the local microclimatology on
plant physiological responses. However, when both fields are of equal interest in a research,
an interdisciplinary multi-scale work is desirable, respecting the importance of the more
visible biophysical system, but also that of the subsurface/hidden components. Therefore,
some gross modelling assumptions found at the thematic borders of those approaches, can be
effectively replaced by much more robust combined methods.
The revision of erosion and sedimentological studies highlighted the problem of upscaling
plot erosion to basin sediment yield. Concerning erosion, the many research antecedents in
coffee plots presented a large variability in the results, making clear that these processes are
highly site-dependent, and that universal estimations must be treated carefully. The only study
found on sediment yield in coffee basins reveals a lack of knowledge in this particular
context. Reviewing sedimentological studies in other types of basins indicated that the most
accurate and current technique to quantify sediment fluxes is the continuous measurement of
turbidity, performing adequately during both storm events and recession times, but
demanding careful in situ calibration. Great deficiencies were observed in the treatment of
uncertainties on the estimations of sediment concentration and yield. Finally, it was widely
demonstrated in this literature review that it is conceptually incorrect the simple upscaling of
plot erosion measurements or estimations through a direct integration of the individual results
intending to provide basin scale estimates.
210
A methodology for the assessment of HES was then proposed, basing on eco-hydrological
and sedimentological approaches and relying on the combination of experimental and
modelling approaches. The target variables had to be clearly defined, being in this study the
aquifer recharge and the water and sediment yields (all three at basin scale). The first two
demanded the detailed closure of the water balance (which required modelling work to
determine the non-measurable components), while the third was addressed through the
continuous determination of the sediment flow, considering all its methodological and
technical uncertainties.
8.4 On the study site, experimental setup and data
In order to follow the methodological proposal presented in the previous section, a completely
new experimental site was built, selecting and equipping a coffee agroforestry basin (1 km2)
on the hillslopes of the Turrialba volcano, characterized by permeable andisols. The field
work proved to be highly effective, providing a unique opportunity to observe directly and
simultaneously several eco-hydrological and sedimentological variables, flow paths and stores
under this tropical environment (e.g. rainfall, climate, streamflow, soil water content,
evapotranspiration, leaf area index, water table level, turbidity and plot erosion). In addition,
field observation allowed the understanding of other processes, which helped to explain
relevant properties and patterns in the data (e.g. high infiltrations, low plot erosion, high
surface runoff on roads), and hence to improve the model parameterization and building.
8.5 On modelling the hydrological behaviour of a coffee agroforestry basin
in Costa Rica
This paper-based chapter (Gómez-Delgado et al., 2011a) presented the successful water
balance closure and partition in the experimental site, combining measuring and modelling
approaches. The general behaviour of this coffee agroforestry basin over one year of
observations can be summarized by the fact that 91% of the rainfall (R) was infiltrated
through the highly permeable andisol, 64% of R was measured as streamflow, 25% of R was
measured as evapotranspiration, no major seasonal water stock variation in the soil was
detected as well as a high contribution of the aquifer to the streamflow as baseflow.
211
A new lumped eco-hydrological model called “Hydro-SVAT” was developed. This model
required to couple hydrological and ecophysiological (SVAT) routines and was calibrated
using the streamflow at outlet of the basin. It was validated by independent and direct
measurements of evapotranspiration, soil water content and water table level. This allowed
the estimation of the deep percolation as being around 11% of R. The precision of the model
was judged fair enough according to an uncertainty analysis, while a novel time-varying
sensitivity analysis permitted to investigate the source of such uncertainty. The first
parameterization of the model was considered adequate, though model simplification could be
attempted, by fixing the two or three less sensitive parameters.
The conceptual eco-hydrological “Hydro-SVAT” model makes it possible to assess different
conditions of climate, land cover, soils and hydrogeology, only demanding rainfall and
streamflow data. A validation in a new coffee-agroforestry basin will be attempted (ongoing
PhD of Mario Villatoro in the basin of Llano Bonito in Tarrazú, Costa Rica). The magnitudes
of the already calibrated parameters can be used as starting values to conduct new calibrations
in sites under similar conditions, in order to produce realistic estimates of actual
evapotranspiration, soil water content and aquifer recharge, which are variables rarely
measured in basins of developing countries.
The experimental coffee-agroforestry basin on andisoils proved to be very conservative in
terms of infiltration, aquifer recharge, regulation of the streamflow and reduction of the
surface contaminant transport. The present evaluation is a good basis for negotiating the
payment for environmental services from end-users of this “blue” water, to be opposed to the
“green water” transpired by plants of “brown water” running off (next chapter). However, one
should remain conscious about some potential inherent vulnerabilities (e.g. contaminant
transport capacity to aquifers) for which no data is available yet.
8.6 On water and sediment yield in a coffee agroforestry system at various
spatio!temporal scales: from plot to basin and from event to annual
scale
This chapter is based on a paper to be submitted in November 2010 (Gómez-Delgado et al.,
2011b, in prep). Basing on the experimental data, a spatio-temporal scaling approach, from
plot to basin and from storm event to annual scales, allowed to trace the main sources of soil
212
erosion in the experimental basin, and to propose hypothesis about the main sediment delivery
processes. Over one year, very low surface runoff and erosion rates were measured at the plot
and the basin scales. Discrepancies were observed in the total yields between these two scales,
the basin showing much more surface runoff and more sediment yield than an estimation
based on the plots, which was interpreted as the effect of roads on surface runoff, that of plot-
road contacts on erosion, and also other sources of sediment yield such as river banks and
bed. The streamflow-recession periods supplied 66% of the annual sediment yield,
contradicting many studies that have measured or assumed most of the annual load being
yielded during a few storm events. Hence, only one third of the sediment measured at the
outlet of the basin was generated by storms, and from this storm-sediment only 16% could
have been produced as plot erosion, according to measurements. Even though beneficial
effects of agroforestry were confirmed here (shaded coffee plots produced half the erosion in
comparison to monoculture coffee plots), the roads played a much more important role on
surface runoff and sediment yield over events. In addition, most of the sediment yield was not
produced over events, but during recessions. Given the importance of recession-time sediment
yield, attention should be paid to the observation of suspended sediment during low flows, to
the analysis of the stream channel dynamics and to the need for additional hydrogeological
investigations.
The optical backscatter sensor provided an estimation of the annual sediment yield of the
basin with acceptable uncertainties, this latter calculated through a standard uncertainty
approach that was newly applied to sediment load calculations. These results confirmed that
the experimental coffee-agroforestry basin is very conservative in terms of annual sediment
yield, which did not exceeded 1 t ha-1 yr
-1. The evaluation here presented seems to be a good
base for negotiating the payment of environmental services. It might appear superfluous to try
to reduce such low sediment yields. However, if it were decided to do so, it would be
recommendable to stabilize the river banks and bed, and to protect the contact between roads
and plots, rather than just planting more trees by default.
It would be worth comparing the sediment yields and the sources of surface runoff and
erosion identified here, using similar approaches in other coffee basins, and in particular, on
different types of soils. It is also recommended to build a coupled ecophysiological-
hydrological-erosion model from these experimental results (next chapter).
213
8.7 On improving water and sediment peakflow simulations through the
designing a semi!distributed eco!hydrological model including the
effect of direct road!runoff: continuous versus event!based validation
strategies
The lumped eco-hydrological model Hydro-SVAT presented before, representing the basin as
single unit, performed well to reproduce the streamflow regime on an annual base, but we
stressed that peakflows over storm events (mostly surface runoff) were underestimated, which
would certainly affect any further attempt to model the basin sediment yield on a per event or
annual basis.
In order to take into account this effect, we first inferred that roads could contribute
significantly to surface runoff over storm events, according to the surface runoff and erosion
work conducted at plot and basin scales. The need to conceptualize them as a different
hydrologic unit in the basin, in addition to plots, was then pointed out. This conducted to a
revised semi-distributed version of the first lumped model, including two types of surface
units, the plots and the roads. The sole addition of a road-interception routine did not improve
the simulations of surface runoff significantly in comparison to the simplified version of
Hydro-SVAT so far. Two proposals can be made towards achieving a successful model: to
add the road-runoff routine directly to the original ten-parameter model (in order to take
advantage of its appropriate representation of the aquifer dynamics), and/or to make variable
the proportion of impermeable areas other than roads (like footpaths inside the plots) whose
dimensions remain unknown. The work is still in process, and the desired result (i.e. a semi-
distributed model that can accurately simulate both the annual hydrograph and the peakflows,
but with a reasonably low level of complexity) appeals a deep re-conceptualization of the
current ensemble.
Second, considering that Hydro-SVAT already had ten parameters in its original version, it
appeared necessary to explore the possibilities of building a simplified version. A
simplification of the original ten-parameter Hydro-SVAT model was assessed in terms of its
flexibility and efficiency. The parameter reduction analysis based on sensitivity analysis
produced a new model that performed well on the same data used for the sensitivity analysis,
but that showed deficiencies when used on a new data set. However, a significant reduction in
model capabilities may occur when sensitivity analysis is used to reduce the amount of
214
parameters of a model, basing on a single data set. Better results may be obtained if this
procedure is applied using a split-sample approach.
A “directed starting point” method for single-objective calibration (based on the overall Nash-
Sutcliffe coefficient: NS) seemed to work appropriately, effectively driving the calibration
runs of all the models to remain within conceptually feasible parametric spaces. An additional
use of multi-start algorithms could be tested along with this “directed-calibration” in order to
produce more general results.
Continuous-time and event-based validation techniques were compared in order to provide
feedback to the model parameterization step. The hydrograph shape efficiency measure C2M
(the bounded NS coefficient) used in the validation step was a much more stable criterion
than the classical NS, providing more reliable summary statistics (e.g. the mean of a group of
storm events). The lack of independence of some criteria studied here (NS, C2M, global and
relative volume errors, and global and relative peak errors) presents a serious problem for
their use as objective functions in multi-objective model calibration procedures that rely on
weighting routines. The use of principal component analysis could be useful for better
approaching the theoretical conditions demanded by those techniques.
Finally, the comparison of four models (some including the effect of roads on the generation
of surface runoff, others simplifying the amount of model reservoirs and parameters) led us to
an unexpected conclusion: noticeable best model performances on streamflow and baseflow
partition were achieved by the model that was created first: the ten-parameter Hydro-SVAT,
which is more complex than the other models tested. In this ten-parameter model, however,
the road routine to refine the simulation of surface runoff was not tested.
These very preliminary results confirmed that simple models are not always accurate,
especially if several properties of complex systems want to be predicted (not only the
streamflow hydrograph shape but also single peak events, required to model sediment yield).
Standard assessments considering the relationship between model efficiency and number of
parameters, can provide useful information in the model selection process.
215
8.8 Perspectives
The methodology proposed in this thesis to evaluate possible HES from coffee agroforestry
basins in Costa Rica, is expected to bring a new quantification approach into the discussions
reported in the up-to-date bibliographical review presented in the first chapter of this
document. It is expected that this methodology can be used as a basis for future design,
negotiation and implementation schemes for the payment of HES, and also for the post-
evaluation of the projects (Basin Management Programmes).
However, additional field work could be carried out in order to improve the precision of the
biophysical methods. For instance, additional studies on hillslope and road surface runoff
might help to gain much more knowledge on the influence of these elements in the
hydrological response during storm events. It would also provide valuable information of the
possible water gains and losses through the roads crossing the borders of the basin.
Although the measured runoff and erosion on the experimental plots were not as great as they
had been thought to be at the beginning of this project, a set of replications of shaded and
non-shaded plots should be added to the experimental setup (especially at the foot of the
hillslopes, in conditions where the soil is closer to saturation), if a generalization of these
results has to be done (ongoing PhD by Laura Benegas).
Data on streamflow composition (baseflow, non-saturated flow and surface runoff) would
bring better understanding on the groundwater dynamics in the basin. Moreover, it would be
meaningful to investigate the sources of suspended sediment outputs either over storm events
or, more importantly, during recession times, which transported two thirds of the annual
sediment load. Therefore, an analysis on the origins of sediment in the river is of major
interest, by measuring how much of it comes from the river banks or the river bed, but also
from roads and from the contact between roads and plots.
Concerning the Hydro-SVAT model, further observations to determine any seasonal
variations in soil and ecophysiological properties driven by agricultural practices (for
instance, the effect of pruning on throughfall, and that of weeding on infiltration), could help
to refine the assumptions of the model, as well as its simulation capacities.
The implementation of infiltration and field capacity tests on the soils of the basin (ongoing
PhD by Laura Benegas, previous studies by MSc Rintaro Kinoshita) would provide actual
216
values for the saturated hydraulic conductivity and the soil saturation and field capacities,
which could noticeably improve the conceptual realism of the model, reducing, at the same
time, the number of parameters. The repeated execution of such tests all over the basin would
enable the refinement of the current Hydro-SVAT in the shape of a semi- or fully-distributed
model.
Finally, since the storm events produce one third of the annual sediment yield in the
experimental basin, the purpose of improving the simulation of streamflow peak rates during
storm events is worthwhile. Although road-runoff routines were implemented for those
periods, the behaviour of the model remains inaccurate, and other hypotheses should be
explored. For example, the temporal variation in soil properties during storm events could be
investigated in order to test how much the inclusion of this effect can improve model
simulations. Other variants in the semi-distributed model presented in the last technical
chapter of this thesis can be also examined.
The development of a sediment yield routine to be coupled to Hydro-SVAT is also a missing
task at present, which first demands to solve the problem of accurate peakflow simulation.
Based on the observed behaviour of the sediment concentrations in time, a stochastic model
could be useful to represent sediment flow during recession times, while conceptual or
empirical approaches reproducing the hysteretic behaviour of the streamflow - sediment
concentration curve, might provide an appropriate framework for the modelling of sediment
peaks during storm events.
At last, it is much desired the validation of the Hydro-SVAT model in other coffee basins
(ongoing PhD by Mario Villatoro). The water balance according to this model can be
compared to that of a plot to multi-plot coffee-agroforestry model (van Oijen et al., 2010b),
contributing with the validation of the latter one, or else, in the refining of some of its
hypotheses concerning the balance of infiltration, surface runoff, evapotranspiration,
subsurface runoff, deep drainage, percolation, etc. (ongoing PhD by Louise Meylan).
Considering that Hydro-SVAT is fundamentally a generic model for the water balance
partitioning at plot and basin scales, it is suggested to develop some applications under
different forest basins (e.g. Eucalyptus plantations in Brazil: Eucflux project, secondary
forests in Congo: EU-Climafrica project).
217
References
Abbott, M.B., Bathurst, J.C., Cunge, J.A., O'Connell, P.E., Rasmussen, J., 1986. An
introduction to the European Hydrological System - Systeme Hydrologique Europeen, "SHE",
1: History and philosophy of a physically-based, distributed modelling system. Journal of
Hydrology 87, 45-59.
Afandi, 2005. Runoff and Sediment Yield from Coffee Micro-Catchment Under Humid
Tropical Climate of Lampung, Indonesia. In: Agus, F., Noordwijk, M.v. (Eds.), Alternatives
to Slash and Burn in Indonesia: Facilitating the development of agroforestry systems: Phase 3
Synthesis and Summary report. World Agroforestry Centre, Sindang Barang, Bogor,
Indonesia.
Afandi, Manik, T.K., Rosadi, B., Utomo, M., Senge, M., Adachi, T., Oki, Y., 2002a. Soil
Erosion under Coffee Trees with Different Weed Managements in Humid Tropical Hilly Area
of Lampung, South Sumatra, Indonesia. Journal of the Japanese Society of Soil Physics 91, 3-
14.
Afandi, Rosadi, B., Maryanto, Nurarifani, Utomo, M., Senge, M., Adachi, T., 2002b.
Sediment Yield from Various Land Use Practices in a Hilly Tropical Area of Lampung
Region, South Sumatra, Indonesia. Journal of the Japanese Society of Soil Physics 91, 25-38.
Agarwal, A., Dickinson, W.T., 1991. Effect of texture, flow, and slope on interrlll sediment
transport. Transactions of the ASABE 34, 1726-1731.
Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration: Guidelines for
computing crop water requirements. Food and Agriculture Organization, Rome.
Amthor, J.S., 1994. Plant respiration responses to the environment and their effect on the
carbon balance. In: Wilkinson., R.E. (Ed.), Plant-Environment Interactions. Marcel Dekker,
Inc., New York, USA, pp. 501-554.
Anderson, E.P., Pringle, C.M., Rojas, M., 2006. Transforming tropical rivers: an
environmental perspective on hydropower development in Costa Rica. pp. 679-693.
Anderson, M.P., Woessner, W.W., 1992. The role of the postaudit in model validation.
Advances in Water Resources 15, 167-173.
218
Andréassian, V., 2004. Waters and forests: from historical controversy to scientific debate.
Journal of Hydrology 291, 1-27.
Arnold, J.G., Allen, P.M., 1996. Estimating hydrologic budgets for three Illinois watersheds.
Journal of Hydrology 176, 57-77.
Arnold, J.G., Srinivasan, R., Muttiah, R.S., Allen, P.M., 1999. Continental scale simulation of
the hydrologic balance. JAWRA Journal of the American Water Resources Association 35,
1037-1051.
Ataroff, M., Monasterio, M., 1997. Soil erosion under different management of coffee
plantations in the Venezuelan Andes. Soil Technology 11, 95-108.
Ávila, H., Harmand, J.M., Dambrine, E., Jiménez, F., Beer, J., Oliver, R., 2004. Dinámica del
nitrógeno en el sistema agroforestal Coffea arabica con Eucalyptus deglupta en la Zona Sur
de Costa Rica. Agroforestería en las Américas 41-42, 83-91.
Aylward, B., 2004. Land use, hydrological function and economic valuation. The joint
UNESCO International Hydrological Programme (IHP) - International Union of Forestry
Research Organizations (IUFRO) symposium and workshop, Forest - water - people in the
humid tropics: past, present and future hydrological research for integrated land and water
management, Universiti Kebangsaan Malaysia, 30 July-4 August 2000. Cambridge University
Press, pp. 99-120.
Baldocchi, D., Meyers, T., 1998. On using eco-physiological, micrometeorological and
biogeochemical theory to evaluate carbon dioxide, water vapor and trace gas fluxes over
vegetation: a perspective. Agr. Forest Meteorol. 90, 1-25.
Baldocchi, D.D., Harley, P.C., 1995. Scaling carbon dioxide and water vapour exchange from
leaf to canopy in a deciduous forest. II. Model testing and application. Plant, Cell &
Environment 18, 1157-1173.
Baumann, J., Rodríguez Morales, J.A., Arellano Monterrosas, J.L., 2008. The effect of
rainfall, slope gradient and soil texture on hydrological processes in a tropical watershed. 15th
International Congress of ISCO. International Soil Conservation Organization, Budapest,
Hungary.
219
Beer, J., Muschler, R., Kass, D., Somarriba, E., 1998. Shade management in coffee and cacao
plantations. Agroforestry Systems 38, 139-164.
Begin, Z.B., Schumm, S.A., 1979. Instability of Alluvial Valley Floors: A Method for its
Assessment. Transactions of the ASAE 22, 347-350.
Beldring, S., 2002. Multi-criteria validation of a precipitation-runoff model. Journal of
Hydrology 257, 189-211.
Bergström, S., 1995. The HBV model. In: Singh, V.P. (Ed.), Computer Models of Watershed
Hydrology. Water Resources Publications, Colorado, USA.
Beven, K., 1993. Prophecy, reality and uncertainty in distributed hydrological modelling.
Advances in Water Resources 16, 41-51.
Beven, K., 2006. A manifesto for the equifinality thesis. Journal of Hydrology 320, 18-36.
Beven, K.J., 1996. Response to comments on ‘A discussion of distributed hydrological
modelling’ by J.C. Refsgaard et al. In: Abbott, M.B., Refsgaard, J.C. (Eds.), Distributed
Hydrological Modelling. Kluwer Academic, Dordrecht, The Netherlands, pp. 289-295.
Beylich, A.A., Sandberg, O., 2005. Geomorphic Effects of the Extreme Rainfall Event of 20–
21 July, 2004 in the Latnjavagge Catchment, Northern Swedish Lapland. Geografiska
Annaler: Series A, Physical Geography 87, 409-419.
Bormann, H., 2006. Impact of spatial data resolution on simulated catchment water balances
and model performance of the multi-scale TOPLATS model. Hydrol. Earth Syst. Sci. 10, 165-
179.
Bosch, J.M., Hewlett, J.D., 1982. A review of catchment experiments to determine the effect
of vegetation changes on water yield and evapotranspiration. Journal of Hydrology 55, 3-23.
Boulet, G., Chehbouni, A., Braud, I., Vauclin, M., Haverkamp, R., Zammit, C., 2000. A
simple water and energy balance model designed for regionalization and remote sensing data
utilization. Agricultural and Forest Meteorology 105, 117-132.
Brasington, J., Richards, K., 2000. Turbidity and suspended sediment dynamics in small
catchments in the Nepal Middle Hills. pp. 2559-2574.
220
Braud, I., Dantas-Antonino, A.C., Vauclin, M., Thony, J.L., Ruelle, P., 1995. A simple soil-
plant-atmosphere transfer model (SiSPAT) development and field verification. Journal of
Hydrology 166, 213-250.
Bréda, N., Granier, A., 1996. Intra- and interannual variations of transpiration, leaf area index
and radial growth of a sessile oak stand (Quercus petraea). Ann. For. Sci. 53, 521-536.
Brooks, K.N., Ffolliott, P.F., Gregersen, H.M., DeBano, L.F., 1997. Hydrology and the
management of watersheds. 2nd ed., Iowa State University Press, Iowa, USA.
Bruijnzeel, L.A., 1990. Hydrology of moist tropical forests and effects of conversion: a state
of knowledge review. UNESCO, Division of Water Sciences, International Hydrological
Programme, Paris.
Bruijnzeel, L.A., 2004. Hydrological functions of tropical forests: not seeing the soil for the
trees? Agriculture, Ecosystems & Environment 104, 185-228.
Cannavo, P., Sansoulet, J., Harmand, J.M., Siles, P., Dreyer, E., Vaast, P., 2011. Agroforestry
associating coffee and Inga densiflora results in complementarity for water uptake and
decreases deep drainage in Costa Rica. Agriculture, Ecosystems & Environment 140, 1-13.
Cannell, M.G.R., 1999. Environmental impacts of forest monocultures: water use,
acidification, wildlife conservation, and carbon storage. New Forests 17, 239-262.
Castro, F., Montes, E., Raine, M., 2004. Centroamérica, la Crisis Cafetalera: Efectos y
Estrategias Para Hacerle Frente. Latin America and Caribbean Region Sustainable
Development working paper No. 23. The World Bank, p. 60.
Cattan, P., Cabidoche, Y.M., Lacas, J.G., Voltz, M., 2006. Effects of tillage and mulching on
runoff under banana (Musa spp.) on a tropical Andosol. Soil Till. Res. 86, 38-51.
Cattan, P., Ruy, S.M., Cabidoche, Y.M., Findeling, A., Desbois, P., Charlier, J.B., 2009.
Effect on runoff of rainfall redistribution by the impluvium-shaped canopy of banana
cultivated on an Andosol with a high infiltration rate. J. Hydrol. 368, 251-261.
Cattan, P., Voltz, M., Cabidoche, Y.M., Lacas, J.G., Sansoulet, J., 2007. Spatial and temporal
variations in percolation fluxes in a tropical Andosol influenced by banana cropping patterns.
J. Hydrol. 335, 157-169.
221
Ceballos, A., Schnabel, S., 1998. Hydrological behaviour of a small catchment in the dehesa
landuse system (Extremadura, SW Spain). Journal of Hydrology 210, 146-160.
Clifford, N.J., Richards, K.S., Brown, R.A., Lane, S.N., 1995. Laboratory and field
assessment of an infrared turbidity probe and its response to particle size and variation in
suspended sediment concentration. Hydrological sciences journal = Journal des sciences
hydrologiques. 40, 771-791.
Cook, P.G., Hatton, T.J., Pidsley, D., Herczeg, A.L., Held, A., O'Grady, A., Eamus, D., 1998.
Water balance of a tropical woodland ecosystem, Northern Australia: a combination of micro-
meteorological, soil physical and groundwater chemical approaches. Journal of Hydrology
210, 161-177.
Cormary, Y., Guilbot, A., 1969. Relations pluie-débit sur le bassin de la Sioule. Rapport
D.G.R.S.T. No. 30. Rapport D.G.R.S.T. No. 30, Université des Sciences et Techniques,
Montpellier, France.
Crawford, N.H., Linsley, R.K., 1966. Digital simulation in hydrology: Stanford watershed
model IV. Technical Report 39, Department of Civil Engineering, Stanford University,
California, USA, p. 210.
Cruse, R.M., Larson, W.E., 1977. Effect of Soil Shear Strength on Soil Detachment due to
Raindrop Impact. Soil Sci. Soc. Am. J. 41, 777-781.
Chahinian, N., Moussa, R., Andrieux, P., Voltz, M., 2005. Comparison of infiltration models
to simulate flood events at the field scale. Journal of Hydrology 306, 191-214.
Chahinian, N., Moussa, R., Andrieux, P., Voltz, M., 2006. Accounting for temporal variation
in soil hydrological properties when simulating surface runoff on tilled plots. Journal of
Hydrology 326, 135-152.
Chappell, N.A., McKenna, P., Bidin, K., Douglas, I., Walsh, R.P.D., 1999. Parsimonious
modelling of water and suspended-sediment flux from nested-catchments affected by
selective tropical forestry.
Charlier, J.B., Cattan, P., Moussa, R., Voltz, M., 2008. Hydrological behaviour and modelling
of a volcanic tropical cultivated catchment. Hydrol. Process. 22, 4355-4370.
222
Charlier, J.B., Moussa, R., Cattan, P., Cabidoche, Y.M., Voltz, M., 2009. Modelling runoff at
the plot scale taking into account rainfall partitioning by vegetation: application to stemflow
of banana (Musa spp.) plant. Hydrol. Earth Syst. Sci. 13, 2151-2168.
Chomitz, K.M., Brenes, E., Constantino, L., 1999. Financing environmental services: the
Costa Rican experience and its implications. The Science of The Total Environment 240, 157-
169.
Chomitz, K.M., Kumari, K., 1998. The Domestic Benefits of Tropical Forests: A Critical
Review. pp. 13-35.
Dabney, S.M., Locke, M.A., Steinriede Jr, R.W., 2006. Turbidity sensors track sediment
concentrations in runoff from agricultural fields. Federal Interagency Sedimentation
Conference Proceedings, Nevada, USA.
Dariah, A., Agus, F., Arsyad, S., Sudarsono, dan-Maswar, 2004. Erosi dan aliran permukaan
pada lahan pertanian berbasis tanaman kopi di Sumberjaya, Lampung Barat. Agrivita 26, 52-
60.
Dauzat, J., Rapidel, B., Berger, A., 2001. Simulation of leaf transpiration and sap flow in
virtual plants: model description and application to a coffee plantation in Costa Rica.
Agricultural and Forest Meteorology 109, 143-160.
Dawes, W.R., Zhang, L., Hatton, T.J., Reece, P.H., Beale, G.T.H., Packer, I., 1997.
Evaluation of a distributed parameter ecohydrological model (TOPOG_IRM) on a small
cropping rotation catchment. Journal of Hydrology 191, 64-86.
de Pury, D.G.G., Farquhar, G.D., 1997. Simple scaling of photosynthesis from leaves to
canopies without the errors of big-leaf models. Plant, Cell & Environment 20, 537-557.
de Reffye, P., Claude, E., Jean, F., Marc, J., Claude, P., 1988. Plant models faithful to
botanical structure and development. ACM SIGGRAPH Computer Graphics 22, 151-158.
De Roo, A.P.J., Offermans, R.J.E., 1995. LISEM: a physically-based hydrological and soil
erosion model for basin-scale water and sediment management. IAHS publication, Modelling
and Management of Sustainable Basin-scale Water Resource Systems (Proceedings of a
Boulder Symposium).
223
de Vente, J., Poesen, J., 2005. Predicting soil erosion and sediment yield at the basin scale:
Scale issues and semi-quantitative models. Earth-Science Reviews 71, 95-125.
de Vente, J., Poesen, J., Arabkhedri, M., Verstraeten, G., 2007. The sediment delivery
problem revisited. Progress in Physical Geography 31, 155-178.
Diskin, M.H., Nazimov, N., 1995. Linear reservoir with feedback regulated inlet as a model
for the infiltration process. Journal of Hydrology 172, 313-330.
Diskin, M.H., Simon, E., 1977. A procedure for the selection of objective functions for
hydrologic simulation models. Journal of Hydrology 34, 129-149.
Dolman, A.J., van der Molen, M.K., ter Maat, H.W., Hutjes, R.W.A., 2004. The Effects of
Forests on Mesoscale Atmospheric Processes. In: Mencuccini, M., Grace, J., Moncrieff, J.,
McNaughton, K.G. (Eds.), Forests at the Land-Atmosphere Interface. CABI Publishing,
Wallingford, UK.
Donigan, A., Bicknell, B., Imhoff, J.C., 1995. Hydrological simulation program – Fortran
(HSPF). In: Singh, V.P. (Ed.), Computer Models of Watershed Hydrology. Water Resources
Publications, Colorado, USA.
Doorenbos, J., Pruitt, W.O., 1977. Crop water requirements. Irrigation and Drainage, Paper
24. Food and Agriculture Organization of the United Nations, Rome, Italy, p. 179.
Dorel, M., Roger-Estrade, J., Manichon, H., Delvaux, B., 2000. Porosity and soil water
properties of Caribbean volcanic ash soils. Soil Use Manage. 16, 133-140.
Douglas, I., Bidin, K., Balamurugam, G., Chappell, N.A., Walsh, R.P.D., Greer, T., Sinun,
W., 1999. The role of extreme events in the impacts of selective tropical forestry on erosion
during harvesting and recovery phases at Danum Valley, Sabah.
Downing, J., 2006. Twenty-five years with OBS sensors: The good, the bad, and the ugly.
Continental Shelf Research 26, 2299-2318.
Droogers, P., Van Loon, A., Immerzeel, W.W., 2008. Quantifying the impact of model
inaccuracy in climate change impact assessment studies using an agro-hydrological model.
Hydrol. Earth Syst. Sci. 12, 669-678.
224
Duan, Q., Sorooshian, S., Gupta, V., 1992. Effective and efficient global optimization for
conceptual rainfall-runoff models. Water Resour. Res. 28, 1015-1031.
Dufrêne, E., Davi, H., François, C., Maire, G.l., Dantec, V.L., Granier, A., 2005. Modelling
carbon and water cycles in a beech forest: Part I: Model description and uncertainty analysis
on modelled NEE. Ecological Modelling 185, 407-436.
Eagleson, P.S., 1978. Climate, soil, and vegetation: 3. A simplified model of soil moisture
movement in the liquid phase. Water Resour. Res. 14, 722-730.
Efstratiadis, A., Koutsoyiannis, D., 2010. One decade of multi-objective calibration
approaches in hydrological modelling: a review. Hydrological Sciences Journal 55, 58-78.
Everaert, W., 1991. Empirical relations for the sediment transport capacity of interrill flow.
Earth Surface Processes and Landforms 16, 513-532.
Falge, E., Baldocchi, D., Olson, R., Anthoni, P., Aubinet, M., Bernhofer, C., Burba, G.,
Ceulemans, R., Clement, R., Dolman, H., Granier, A., Gross, P., Grünwald, T., Hollinger, D.,
Jensen, N.-O., Katul, G., Keronen, P., Kowalski, A., Ta Lai, C., Law, B.E., Meyers, T.,
Moncrieff, J., Moors, E., William Munger, J., Pilegaard, K., Rannik, Ü., Rebmann, C.,
Suyker, A., Tenhunen, J., Tu, K., Verma, S., Vesala, T., Wilson, K., Wofsy, S., 2001. Gap
filling strategies for long term energy flux data sets. Agr. Forest Meteorol. 107, 71-77.
Fallas, J., 2002. Agua, bosques y servicios ambientales. Seminario Internacional Servicios
Hidrológicos de los Ecosistemas Forestales. FONAFIFO, San Jose, Costa Rica.
Farley, K.A., Jobbágy, E.G., Jackson, R.B., 2005. Effects of afforestation on water yield: a
global synthesis with implications for policy. Global Change Biology 11, 1565-1576.
Fleming, G., 1975. Computer Simulation Techniques in Hydrology. Environmental Science
Series, Elsevier, New York, USA.
Fortin, J.P., Moussa, R., Bocquillon, C., Villeneuve, J.P., 1995. Hydrotel, un modèle
hydrologique distribué pouvant bénéficier des données fournies par la télédétection et les
systèmes d’information géographique. Rev. Sci. Eau / J. Water Sci. 8, 97-124.
Foster, G.R., 1986. Understanding ephemeral gully erosion. Soil Conservation: Assessing the
National Research Inventory. National Research Council, Board on Agriculture, National
Academy Press, Washington, USA, pp. 90-118.
225
Foster, G.R., Lane, L.J., Nowlin, J.D., Laflen, J.M., Young, R.A., 1981. Estimating erosion
and sediment yield on field-sized areas. Transactions of the ASABE 24, 1253-1263.
Foster, G.R., Yoder, D.C., Weesies, G.A., Toy, T.J., 2001. The Design Philosophy Behind
RUSLE2: Evolution of an Empirical Model. In: Ascough II, J.C., Flanagan, D.C. (Eds.), Soil
Erosion Research for the 21st Century, Proc. Int. Symp. Honolulu. ASAE, pp. 95-98.
Foster, I.D.L., Millington, R., Grew, R.G., 1992. The impact of particle size controls on
stream turbidity measurement; some implications for suspended sediment yield estimation. In:
IAHS (Ed.), Erosion and Sediment Transport Monitoring Programmes in River Basins. IAHS,
Oslo, Norway, pp. 51-62.
Frey, H.C., Patil, S.R., 2002. Identification and Review of Sensitivity Analysis Methods. Risk
Anal. 22, 553-578.
Fujieda, M., Kudoh, T., de Cicco, V., de Calvarcho, J.L., 1997. Hydrological processes at two
subtropical forest catchments: the Serra do Mar, São Paulo, Brazil. Journal of Hydrology 196,
26-46.
Gallart, F., Llorens, P., Latron, J., Regüés, D., 1999. Hydrological processes and their
seasonal controls in a small Mediterranean mountain catchment in the Pyrenees. Hydrol.
Earth Syst. Sci. 6, 527-537.
García-Ruiz, J.M., Regüés, D., Alvera, B., Lana-Renault, N., Serrano-Muela, P., Nadal-
Romero, E., Navas, A., Latron, J., Martí-Bono, C., Arnáez, J., 2008. Flood generation and
sediment transport in experimental catchments affected by land use changes in the central
Pyrenees. Journal of Hydrology 356, 245-260.
García, A., Sainz, A., Revilla, J.A., Álvarez, C., Juanes, J.A., Puente, A., 2008. Surface water
resources assessment in scarcely gauged basins in the north of Spain. J. Hydrol. 356, 312-326.
Genereux, D.P., Jordan, M.T., Carbonell, D., 2005. A paired-watershed budget study to
quantify interbasin groundwater flow in a lowland rain forest, Costa Rica. Water Resour. Res.
41, W04011.
Ghadiri, H., Payne, D., 1981. Raindrop impact stress. Journal of Soil Science 32, 41-49.
Gilroy, E.J., Hirsch, R.M., Cohn, T.A., 1990. Mean square error of regression-based
constituent transport estimates. Water Resour. Res. 26, 2069-2077.
226
Gippel, C.J., 1989. The use of turbidimeters in suspended sediment research. Hydrobiologia
176-177, 465-480.
Gippel, C.J., 1995. Potential of turbidity monitoring for measuring the transport of suspended
solids in streams. pp. 83-97.
Gómez-Delgado, F., Roupsard, O., le Maire, G., Taugourdeau, S., Pérez, A., van Oijen, M.,
Vaast, P., Rapidel, B., Harmand, J.M., Voltz, M., Bonnefond, J.M., Imbach, P., Moussa, R.,
2011a. Modelling the hydrological behaviour of a coffee agroforestry basin in Costa Rica.
Hydrol. Earth Syst. Sci. 15, 369-392.
Gómez-Delgado, F., Roupsard, O., Moussa, R., 2011b. Water and sediment yield in a coffee
agroforestry system at various spatio-temporal scales: from plot to basin and from event to
annual scales. Agriculture, Ecosystems & Environment, in prep.
Gómez-Gómez, R., 2005. Efecto del control de malezas con paraquat y glifosato sobre la
erosión y pérdida de nutrimentos del suelo en cafeto. Agronomía Mesoamericana 16, 77-87.
Govers, G., Giménez, R., Van Oost, K., 2007. Rill erosion: Exploring the relationship
between experiments, modelling and field observations. Earth-Science Reviews 84, 87-102.
Granier, A., Biron, P., Lemoine, D., 2000. Water balance, transpiration and canopy
conductance in two beech stands. Agricultural and Forest Meteorology 100, 291-308.
Granier, A., Breda, N., Biron, P., Villette, S., 1999. A lumped water balance model to
evaluate duration and intensity of drought constraints in forest stands. Ecological Modelling
116, 269-283.
Green, W.H., Ampt, G.A., 1911. Studies on soil physics, 1. The flow of air and water through
soils. Journal of Agricultural Science 4, 1-24.
Grier, C.G., Running, S.W., 1977. Leaf Area of Mature Northwestern Coniferous Forests:
Relation to Site Water Balance. Ecology 58, 893-899.
Guillemette, F., Plamondon, A.P., Prévost, M., Lévesque, D., 2005. Rainfall generated
stormflow response to clearcutting a boreal forest: peak flow comparison with 50 world-wide
basin studies. Journal of Hydrology 302, 137-153.
227
Gutiérrez, M.V., Meinzer, F.C., Grantz, D.A., 1994. Regulation of transpiration in coffee
hedgerows: covariation of environmental variables and apparent responses of stomata to wind
and humidity. Plant Cell Environ. 17, 1305-1313.
Hairiah, K., Suprayogo, D., Widianto, Prayogo, C., 2005. Trees that produce mulch layers
which reduce run-off and soil loss in coffee multistrata systems. Alternatives to slash and
burn in Indonesia: facilitating the development of agroforestry systems: phase 3 synthesis and
summary report. World Agroforestry Centre - ICRAF, SEA Regional Office, Bogor,
Indonesia, pp. 9-30.
Hakkim, V.M.A., Nandakumar, V., Sajeena, S., 2004. Runoff variation due to landuse change
in small watersheds of Western Ghats. Indian Journal of Environmental Protection 3, 99-105.
Hamby, D.M., 1994. A review of techniques for parameter sensitivity analysis of
environmental models. Environ. Monit. Assess. 32, 135-154.
Hamby, D.M., 1995. A Comparison of Sensitivity Analysis Techniques. Health Phys. 68,
195-204.
Harmand, J.-M., Ávila, H., Dambrine, E., Skiba, U., de Miguel, S., Renderos, R., Oliver, R.,
Jiménez, F., Beer, J., 2007. Nitrogen dynamics and soil nitrate retention in a Coffea arabica -
Eucalyptus deglupta agroforestry system in Southern Costa Rica. Biogeochemistry 85, 125-
139.
Havnø, K., Madsen, M.N., Dørge, J., 1995. MIKE 11 – a generalized river modelling
package. In: Singh, V.P. (Ed.), Computer Models of Watershed Hydrology. Water Resources
Publications, Colorado, USA.
Hayami, S., 1951. On the propagation of flood waves. Disaster Prev. Res. Inst. Bull. 1, 1-16.
Helton, J.C., 1999. Uncertainty and sensitivity analysis in performance assessment for the
Waste Isolation Pilot Plant. Computer Physics Communications 117, 156-180.
Henriksen, H.J., Troldborg, L., Nyegaard, P., Sonnenborg, T.O., Refsgaard, J.C., Madsen, B.,
2003. Methodology for construction, calibration and validation of a national hydrological
model for Denmark. Journal of Hydrology 280, 52-71.
228
Hodnett, M.G., Tomasella, J., 2002. Marked differences between van Genuchten soil water-
retention parameters for temperate and tropical soils: a new water-retention pedo-transfer
functions developed for tropical soils. Geoderma 108, 155-180.
Hoffmann, W.A., Jackson, R.B., 2000. Vegetation-Climate Feedbacks in the Conversion of
Tropical Savanna to Grassland. Journal of Climate 13, 1593-1602.
Holmes, J.W., Sinclair, J.A., 1986. Water Yield from Some Afforested Catchments in
Victoria. In: Barton, A.I.o.E., Australia (Ed.), Hydrology and Water Resources Symposium
1986: River Basin Management. Institution of Engineers Griffith University, Brisbane,
Brisbane, Qld., pp. 214-218.
Horton, R.E., 1933. The role of infiltration in the hydrologic cycle. Transactions, American
Geophysical Union 14, 446-460.
ICE, 2009. Plan de Expansión de la Generación Eléctrica, Período 2010–2021. Instituto
Costarricense de Electricidad, San José, Costa Rica.
Ilstedt, U., Malmer, A., Verbeeten, E., Murdiyarso, D., 2007. The effect of afforestation on
water infiltration in the tropics: A systematic review and meta-analysis. Forest Ecology and
Management 251, 45-51.
Imbach, A.C., Fassbender, H.W., Beer, J., Borel, R., Bonnemann, A., 1989. Sistemas
agroforestales de café (Coffea arabica) con laurel (Cordia alliodora) y café con poró
(Erythrina poeppigiana) en Turrialba, Costa Rica. VI. Balances hídricos e ingreso con lluvias
y lixiviación de elementos nutritivos. Turrialba 39, 400-414.
Imbach, P., Molina, L., Locatelli, B., Roupsard, O., Ciais, P., Corrales, L., Mahe, G., 2010.
Climatology-based regional modelling of potential vegetation and average annual long-term
runoff for Mesoamerica. Hydrol. Earth Syst. Sci. 7, 1801-1817.
Ines, A.V.M., Droogers, P., 2002. Inverse modelling in estimating soil hydraulic functions: a
Genetic Algorithm approach. Hydrol. Earth Syst. Sci. 6, 49-65.
Jackson, R.B., Carpenter, S.R., Dahm, C.N., McKnight, D.M., Naiman, R.J., Postel, S.L.,
Running, S.W., 2001. Water in a changing world. Ecological Applications 11, 1027-1045.
229
Jackson, R.B., Jobbagy, E.G., Avissar, R., Roy, S.B., Barrett, D.J., Cook, C.W., Farley, K.A.,
le Maitre, D.C., McCarl, B.A., Murray, B.C., 2005. Trading Water for Carbon with Biological
Carbon Sequestration. Science 310, 1944-1947.
Jansson, M.B., 1992. Turbidimeter measurements in a tropical river, Costa Rica. In: IAHS
(Ed.), Erosion and Sediment Transport Monitoring Programmes in River Basins. IAHS, Oslo,
Norway, pp. 71-78.
Jansson, M.B., Erlingsson, U., 2000. Measurement and quantification of a sedimentation
budget for a reservoir with regular flushing. Regul. Rivers-Res. Manage. 16, 279-306.
Jaubert, M., 2001. River basin management: a new challenge for ICE. Tecnología ICE 11,
100-111.
JCGM, 2008. Evaluation of measurement data - Guide to the expression of uncertainty in
measurement. BIPM, IEC, IFCC, ILAC, ISO, IUPAC, IUPAP, OIML, Sèvres, France.
Jiménez, F., 1986. Balance hídrico con énfasis en percolación de dos sistemas agroforestales:
café-poró y café-laurel, en Turrialba, Costa Rica. MSc Thesis. CATIE, Turrialba, Costa Rica,
p. 104.
Joffre, R., Rambal, S., 1993. How Tree Cover Influences the Water Balance of Mediterranean
Rangelands. Ecology 74, 570-582.
Jothityangkoon, C., Sivapalan, M., Farmer, D.L., 2001. Process controls of water balance
variability in a large semi-arid catchment: downward approach to hydrological model
development. Journal of Hydrology 254, 174-198.
Kaimowitz, D., 2004. Useful Myths and Intractable Truths: The Politics of the Link Between
Forests and Water in Central America. In: Bonell, M., Bruijnzeel, L.A. (Eds.), Forests, water
and people in the humid tropics: past, present, and future hydrological research for integrated
land and water management. Cambridge University Press, Cambridge, UK, pp. 86-98.
Kinner, D.A., Stallard, R.F., 2004. Identifying storm flow pathways in a rainforest catchment
using hydrological and geochemical modelling. Hydrol. Process. 18, 2851-2875.
Lagarias, J.C., Reeds, J.A., Wright, M.H., Wright, P.E., 1998. Convergence Properties of the
Nelder-Mead Simplex Method in Low Dimensions. SIAM Journal on Optimization 9, 112-
147.
230
Lana-Renault, N., Regues, D., Marti-Bono, C., Begueria, S., Latron, J., Nadal, E., Serrano, P.,
Garcia-Ruiz, J.M., 2007. Temporal variability in the relationships between precipitation,
discharge and suspended sediment concentration in a small Mediterranean mountain
catchment. Nordic hydrology 38, 139.
Lane, P.N.J., Sheridan, G.J., Noske, P.J., 2006. Changes in sediment loads and discharge from
small mountain catchments following wildfire in south eastern Australia. Journal of
Hydrology 331, 495-510.
Le Bissonnais, Y., Cerdan, O., Lecomte, V., Benkhadra, H., Souchère, V., Martin, P., 2005.
Variability of soil surface characteristics influencing runoff and interrill erosion. Catena 62,
111-124.
Le Maitre, D.C., Scott, D.F., Colvin, C., 1999. Review of information on interactions between
vegetation and groundwater. Water SA 25, 137-152.
Lenhart, T., Eckhardt, K., Fohrer, N., Frede, H.G., 2002. Comparison of two different
approaches of sensitivity analysis. Phys. Chem. Earth 27, 645-654.
Lenzi, M.A., Marchi, L., 2000. Suspended sediment load during floods in a small stream of
the Dolomites (northeastern Italy). CATENA 39, 267-282.
Leonard, J., Andrieux, P., 1998. Infiltration characteristics of soils in Mediterranean vineyards
in Southern France. Catena 32, 209-223.
Lesack, L.F.W., 1993. Water Balance and Hydrologic Characteristics of a Rain Forest
Catchment in the Central Amazon Basin. Water Resources Research 29, 759-773.
Leuning, R., Dunin, F.X., Wang, Y.P., 1998. A two-leaf model for canopy conductance,
photosynthesis and partitioning of available energy. II. Comparison with measurements.
Agricultural and Forest Meteorology 91, 113-125.
Lewis, J., 1996. Turbidity-Controlled Suspended Sediment Sampling for Runoff-Event Load
Estimation. Water Resour. Res. 32, 2299-2310.
Lewis, J., 2003. Turbidity-controlled sampling for suspended sediment load estimation. In:
IAHS (Ed.), Erosion and sediment transport measurement in rivers: technological and
methodological advances. IAHS, Oslo, Norway, pp. 13-20.
231
Lewis, J., Eads, R., 2001. Turbidity threshold sampling for suspended sediment load
estimation. 7th Federal Interagency Sedimentation Conference, Nevada, USA, pp. III110 -
III117.
Lindström, G., 1997. A Simple Automatic Calibration Routine for the HBV Model. Nordic
Hydrology 28, 153–168.
Lu, H., Moran, C.J., Prosser, I.P., 2006. Modelling Sediment Delivery Ratio Over The
Murray Darling Basin. Environmental Modelling & Software 21, 1297-1308.
Lyngbæk, A., Muschler, R.G., sinclair, F., 2001. Productivity and profitability of multistrata
organic versus conventional coffee farms in Costa Rica. Agroforestry Systems 53, 205-213.
Lloyd, J., Grace, J., Miranda, A.C., Meir, P., Wong, S.C., Miranda, H.S., Wright, I.R., Gash,
J.H.C., McIntyre, J., 1995. A simple calibrated model of Amazon rainforest productivity
based on leaf biochemical properties. Plant, Cell & Environment 18, 1129-1145.
Madsen, H., 2000. Automatic calibration of a conceptual rainfall-runoff model using multiple
objectives. Journal of Hydrology 235, 276-288.
Maeda, K., Tanaka, T., Park, H., Hattori, S., 2006. Spatial distribution of soil structure in a
suburban forest catchment and its effect on spatio-temporal soil moisture and runoff
fluctuations. Journal of Hydrology 321, 232-256.
Manik, T.K., 2008. Hydrological consequences of converting forestland to coffee plantations
and other agriculture crops on Sumber Jaya watershed, West Lampung, Indonesia. PhD
Thesis. National University of Singapore, Singapore, p. 281.
Maraux, F., Lafolie, F., 1998. Modeling Soil Water Balance of a Maize-Sorghum Sequence.
Soil Sci. Soc. Am. J. 62, 75-82.
Maraux, F., Lafolie, F., Bruckler, L., 1998. Comparison between mechanistic and functional
models for estimating soil water balance: deterministic and stochastic approaches.
Agricultural Water Management 38, 1-20.
Marsden, C., le Maire, G., Stape, J.-L., Seen, D.L., Roupsard, O., Cabral, O., Epron, D.,
Lima, A.M.N., Nouvellon, Y., 2010. Relating MODIS vegetation index time-series with
structure, light absorption and stem production of fast-growing Eucalyptus plantations. Forest
Ecology and Management 259, 1741-1753.
232
Mathevet, T., 2005. Quels modèles pluie-debit globaux au pas de temps horaire ?
Développements empiriques et comparaison de modèles sur un large échantillon de bassins
versants. PhD Thesis, ENGREF, Paris, p. 354.
McCuen, R.H., 2005. Hydrologic analysis and design. Pearson Prentice Hall, Upper Saddle
River, NJ, USA.
MEA, 2005. Ecosystems and Human Well-Being: Synthesis. Millennium Ecosystem
Assessment, Island Press, Washington, USA, p. 155.
MINAE-RECOPE, 1991. Mapa geológico de Costa Rica, Scale 1:200000. San José, Costa
Rica.
Minella, J.P.G., Merten, G.H., Reichert, J.M., Clarke, R.T., 2008. Estimating suspended
sediment concentrations from turbidity measurements and the calibration problem. pp. 1819-
1830.
Monteith, J.L., 1965. Evaporation and environment. In: Fogg, G.E. (Ed.), The State and
Movement of Water in Living Organisms, Symposium of the Society for Experimental
Biology. Cambridge University Press, New York, pp. 205-234.
Mora-Chinchilla, R., 2000. Geomorfología de la cuenca del Río Turrialba (unpublished).
Universidad de Costa Rica, San José, Costa Rica.
Moussa, R., Bocquillon, C., 1996. Algorithms for solving the diffusive wave flood routing
equation. Hydrol. Process. 10, 105-123.
Moussa, R., Chahinian, N., 2009. Comparison of different multi-objective calibration criteria
using a conceptual rainfall-runoff model of flood events. Hydrol. Earth Syst. Sci. 13, 519-535.
Moussa, R., Chahinian, N., Bocquillon, C., 2007a. Distributed hydrological modelling of a
Mediterranean mountainous catchment - Model construction and multi-site validation. Journal
of Hydrology 337, 35-51.
Moussa, R., Chahinian, N., Bocquillon, C., 2007b. Erratum to "Distributed hydrological
modelling of a Mediterranean mountainous catchment - Model construction and multi-site
validation" [J. Hydrol. 337 (2007) 35-51]. J. Hydrol. 345, 254-254.
233
Moussa, R., Voltz, M., Andrieux, P., 2002. Effects of the spatial organization of agricultural
management on the hydrological behaviour of a farmed catchment during flood events.
Hydrol. Process. 16, 393-412.
Nachtergaele, J., Poesen, J., Sidorchuk, A., Torri, D., 2002. Prediction of concentrated flow
width in ephemeral gully channels. Hydrol. Process. 16, 1935-1953.
Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models part I - A
discussion of principles. Journal of Hydrology 10, 282-290.
Nearing, M., Renard, K., Nichols, M., 2006. Erosion Prediction and Modeling. Encyclopedia
of Hydrological Sciences. John Wiley & Sons, Ltd, pp. 1221-1227.
Neilson, R.P., 1995. A Model for Predicting Continental-Scale Vegetation Distribution and
Water Balance. Ecological Applications 5, 362-385.
Nelder, J.A., Mead, R., 1965. A Simplex Method for Function Minimization. The Computer
Journal 7, 308-313.
Nilson, T., 1971. A theoritical analysis of the frequency of gaps in plant stands. Agr.
Meteorol. 8, 25-38.
Norman, J.M., 1980. Interfacing leaf and canopy light interception models. In: Hesketh, J.D.,
Jones, J.W. (Eds.), Predicting Photosynthesis for Ecosystem Models. CRC Press, Florida,
USA., pp. 49-67.
Notter, B., MacMillan, L., Viviroli, D., Weingartner, R., Liniger, H.-P., 2007. Impacts of
environmental change on water resources in the Mt. Kenya region. J. Hydrol. 343, 266-278.
Ogden, C.B., van Es, H.M., Schindelbeck, R.R., 1997. Miniature Rain Simulator for Field
Measurement of Soil Infiltration. Soil Sci. Soc. Am. J. 61, 1041-1043.
Pagiola, S., 2008. Payments for environmental services in Costa Rica. Ecological Economics
65, 712-724.
Park, A., Cameron, J.L., 2008. The influence of canopy traits on throughfall and stemflow in
five tropical trees growing in a Panamanian plantation. Forest Ecol. Manag. 255, 1915-1925.
Parsons, A.J., 2006. Erosion and Sediment Transport by Water on Hillslopes. Encyclopedia of
Hydrological Sciences. John Wiley & Sons, Ltd, pp. 1199-1208.
234
Parsons, A.J., Stromberg, S.G.L., 1998. Experimental analysis of size and distance of travel of
unconstrained particles in interrill flow. Water Resour. Res. 34, 2377-2381.
Parsons, A.J., Stromberg, S.G.L., Greener, M., 1998. Sediment-transport competence of rain-
impacted interrill overland flow. Earth Surface Processes and Landforms 23, 365-375.
Parton, W.J., Schimel, D.S., Cole, C.V., Ojima, D.S., 1987. Analysis of Factors Controlling
Soil Organic Matter Levels in Great Plains Grasslands. Soil Science Society of America
Journal 51, 1173-1179.
Payán, F., Jones, D., Beer, J., Harmand, J.-M., 2009. Soil characteristics below Erythrina
poeppigiana in organic and conventional Costa Rican coffee plantations. Agroforestry
Systems 76, 81-93.
Pearce, A.J., Rowe, L.K., 1979. Forest management effects on interception, evaporation and
water yield. Journal of Hydrology (New Zealand) 18, 73-87.
Peel, M.C., Finlayson, B.L., McMahon, T.A., 2007. Updated world map of the Köppen-
Geiger climate classification. Hydrol. Earth Syst. Sci. 11, 1633-1644.
Penman, H.L., 1948. Natural Evaporation from Open Water, Bare Soil and Grass. Proc. R.
Soc. Lond. A 193, 120-145.
Perrin, C., 2000. Vers une amélioration d’un modèle global pluie-débit au travers d’une
approche comparative. PhD Thesis, INPG-Cemagref, Paris.
Perrin, C., Michel, C., Andréassian, V., 2001. Does a large number of parameters enhance
model performance? Comparative assessment of common catchment model structures on 429
catchments. Journal of Hydrology 242, 275-301.
Pickup, G., 1977. Testing the efficiency of algorithms and strategies for automatic calibration
of rainfall-runoff models. Hydrological Sciences Bulletin 22, 257-274.
Poesen, J., 1985. An improved splash transport model. Zeitschrift für Geomorphologie 29,
193–211.
Poesen, J., Nachtergaele, J., Verstraeten, G., Valentin, C., 2003. Gully erosion and
environmental change: importance and research needs. CATENA 50, 91-133.
235
Poesen, J., Vandaele, K., van Wesemael, B., 1998. Gully erosion: importance and model
implications. In: Boardman, J., Favis-Mortlock, D. (Eds.), Modelling Soil Erosion by Water,
NATO ASI Series 155. Springer-Verlag, Berlin, Germany, pp. 285-311.
Postel, S.L., Thompson, B.H., 2005. Watershed protection: Capturing the benefits of nature's
water supply services. Natural Resources Forum 29, 98-108.
Poulenard, J., Podwojewski, P., Janeau, J.-L., Collinet, J., 2001. Runoff and soil erosion under
rainfall simulation of Andisols from the Ecuadorian Páramo: effect of tillage and burning.
Catena 45, 185-207.
Quansah, C., 1981. The effect of soil type, slope, rain intensity and their interactions on
splash detachment and transport. Journal of Soil Science 32, 215-224.
Querner, E.P., 1997. Description and application of the combined surface and groundwater
flow model MOGROW. Journal of Hydrology 192, 158-188.
Raclot, D., Le Bissonnais, Y., Louchart, X., Andrieux, P., Moussa, R., Voltz, M., 2009. Soil
tillage and scale effects on erosion from fields to catchment in a Mediterranean vineyard area.
Agriculture, Ecosystems & Environment 134, 201-210.
Raupach, M.R., Finnigan, J.J., 1988. 'Single-Layer Models of Evaporation From Plant
Canopies Are Incorrect but Useful, Whereas Multilayer Models Are Correct but Useless':
Discuss. Australian Journal of Plant Physiology 15, 705-716.
Rauws, G., Govers, G., 1988. Hydraulic and soil mechanical aspects of rill generation on
agricultural soils. Journal of Soil Science 39, 111-124.
Refsgaard, J.C., 1997. Parameterisation, calibration and validation of distributed hydrological
models. Journal of Hydrology 198, 69-97.
Renard, K.G., Foster, G.R., Weesies, G.A., McCool, D.K., Yoder, D.C., 1997. Predicting Soil
Erosion by Water: A Guide to Conservation Planning With the Revised Universal Soil Loss
Equation (RUSLE). U.S. Department of Agriculture, Agriculture Handbook No. 703,
Washington, USA, p. 404.
Renard, K.G., Foster, G.R., Weesies, G.A., Porter, J.P., 1991. RUSLE: Revised universal soil
loss equation. Journal of Soil and Water Conservation 46, 30-33.
236
Richards, L.A., 1931. Capillary conduction of liquids through porous mediums. Physics 1,
318-333.
Rijsdijk, A., 2005. Evaluating sediment sources and delivery in a tropical volcanic watershed.
IAHS publication 292, 16-23.
Rijsdijk, A., Bruijnzeel, L.A., Bremmer, C.N., Kaatee, E.G., S., C.K., 1990. Sediment sources
and their evaluation in the Upper Konto watershed, East Java, Indonesia. In: IAHS (Ed.),
Proceedings of the Fiji Symposium, June 1990. IAHS, Suva, Fiji, p. 391.
Rijsdijk, A., Bruijnzeel, L.A., Prins, T.M., 2007a. Sediment yield from gullies, riparian mass
wasting and bank erosion in the Upper Konto catchment, East Java, Indonesia.
Geomorphology 87, 38-52.
Rijsdijk, A., Sampurno Bruijnzeel, L.A., Sutoto, C.K., 2007b. Runoff and sediment yield
from rural roads, trails and settlements in the upper Konto catchment, East Java, Indonesia.
Geomorphology 87, 28-37.
Ritchie, J.T., 1985. A user-oriented model of the soil water balance in wheat. In: Day, W.,
Atkin, R.K. (Eds.), Wheat growth and modeling. Plenum Publishing Corporation, New York,
USA, pp. 293-305.
Roberts, G., Harding, R.J., 1996. The use of simple process-based models in the estimate of
water balances for mixed land use catchments in East Africa. Journal of Hydrology 180, 251-
266.
Robinson, M., Cognard-Plancq, A.L., Cosandey, C., David, J., Durand, P., Führer, H.W.,
Hall, R., Hendriques, M.O., Marc, V., McCarthy, R., McDonnell, M., Martin, C., Nisbet, T.,
O'Dea, P., Rodgers, M., Zollner, A., 2003. Studies of the impact of forests on peak flows and
baseflows: a European perspective. Forest Ecol. Manag. 186, 85-97.
Rojas, M., 2002. El Sistema de Pago por Servicios Ambientales en Costa Rica: Algunas
Implicaciones para Pequeños Proyectos Hidroeléctricos. Seminario Internacional Servicios
Hidrológicos de los Ecosistemas Forestales. FONAFIFO, San Jose, Costa Rica.
Rojas, M., Aylward, B., 2002. Cooperation between a small private hydropower producer and
a conservation NGO for forest protection: The case of La Esperanza, Costa Rica. In: FAO
237
(Ed.), Land-water Linkages in Rural Watersheds Case Study Series. Food and Agriculture
Organization, Rome.
Rojas, M., Aylward, B., 2003. What are we learning from experiences with markets for
environmental services in Costa Rica? A review and critique of the literature. International
Institute for Environment and Development, London, UK, p. 102.
Roupsard, O., Bonnefond, J.-M., Irvine, M., Berbigier, P., Nouvellon, Y., Dauzat, J., Taga, S.,
Hamel, O., Jourdan, C., Saint-André, L., Mialet-Serra, I., Labouisse, J.-P., Epron, D., Joffre,
R., Braconnier, S., Rouzière, A., Navarro, M., Bouillet, J.-P., 2006. Partitioning energy and
evapo-transpiration above and below a tropical palm canopy. Agricultural and Forest
Meteorology 139, 252-268.
Roupsard, O., Dauzat, J., Nouvellon, Y., Deveau, A., Feintrenie, L., Saint-André, L., Mialet-
Serra, I., Braconnier, S., Bonnefond, J.-M., Berbigier, P., Epron, D., Jourdan, C., Navarro, M.,
Bouillet, J.-P., 2008. Cross-validating Sun-shade and 3D models of light absorption by a tree-
crop canopy. Agricultural and Forest Meteorology 148, 549-564.
Rouse, J.W., Haas, R.H., Schell, J.A., Deering, D.W., 1974. Monitoring vegetation systems in
the Great Plains with ERTS. In: Freden, S.C., Mercanti, E.P., Becker, M.A. (Eds.), Proc.
Third Earth Resources Tech. Satellite-1 Symp. Vol. 1: Technical Presentations, Sec. A,
NASA SP-351. NASA Science and Technology Information Office, Washington, USA, pp.
309-317.
Ruiz, L., Varma, M.R.R., Kumar, M.S.M., Sekhar, M., Maréchal, J.-C., Descloitres, M.,
Riotte, J., Kumar, S., Kumar, C., Braun, J.-J., 2010. Water balance modelling in a tropical
watershed under deciduous forest (Mule Hole, India): Regolith matric storage buffers the
groundwater recharge process. J. Hydrol. 380, 460-472.
Sahin, V., Hall, M.J., 1996. The effects of afforestation and deforestation on water yields.
Journal of Hydrology 178, 293-309.
Saint-André, L., Roupsard, O., Marsden, C., Thongo M’Bou, A., D’Annunzio, R., De
Grandcourt, A., Jourdan C., Derrien, D., Picard, N., Zeller, B., Harmand, J.-M., Levillain, J.,
Henry, M., Nouvellon, Y., Deleporte, P., Bouillet, J.-P., J-P., L., 2007. Literature review on
current methodologies to assess C balance in CDM Afforestation/reforestation projects and a
few relevant alternatives for assessing water and nutrient balance, as a complement to carbon
238
sequestration assessments. CARBOAFRICA Project: Quantification, understanding and
prediction of carbon cycle, and other GHG gases, in Sub-Saharan Africa. CIRAD,
Montpellier, France, p. 105.
Saint-Venant, B., 1871. Théorie du mouvement non permanents des eaux avec application des
rivières et à l’introduction des marées dans leur lit. Comptes Rendus de l’Académie des
Sciences 73, 148-154.
Saison, C., Cattan, P., Louchart, X., Voltz, M., 2008. Effect of Spatial Heterogeneities of
Water Fluxes and Application Pattern on Cadusafos Fate on Banana-Cultivated Andosols. J.
Agr. Food Chem. 56, 11947-11955.
Samra, J.S., Sikka, A.K., Sharda, V.N., 2001. Hydrological implications of planting bluegum
in natural shola and grassland watersheds of southern India. In: Stott, D.E., Mohtar, R.H.,
Steinhardt, G.C. (Eds.), Sustaining the Global Farm. Selected papers from the 10th
International Soil Conservation Organization Meeting. Purdue University and USDA-ARS
National Soil Erosion Research Laboratory, USA, pp. 338-343.
Sansoulet, J., 2007. Transfert d'eau et des ions potassium et nitrate dans un sol à capacité
d'échange anionique sous un couvert redistributeur de la pluie. Etude expérimentale et
modélisation dans une bananeraie fertilisée sur un andosol. PhD thesis. INA-PG, Paris,
France, p. 135.
Scott, D.F., Bruijnzeel, L.A., Vertessey, R.A., Calder, I.R., 2004. Impacts of forest plantations
on streamflow. In: Burley, J., Evans, J., Youngquist, J. (Eds.), Encyclopedia of Forest
Science. Elsevier, Oxford, pp. 367–377.
Scott, D.F., Smith, R.E., 1997. Preliminary empirical models to predict reductions in total and
low flows resulting from afforestation. Water SA 23, 135-140.
SCS, 1972. Estimation of direct runoff from storm rainfall. National Engineering Handbook.
Section 4, Hydrology. Soil Conservation Service, USDA, Washington, USA, pp. 10.11-10.24.
Schoellhamer, D.H., Wright, S.A., 2003. Continuous measurement of suspended-sediment
discharge in rivers by use of optical backscatterance sensors. IAHS publication 283, 1 - 9.
239
Schwab, M., Rieke-Zapp, D., Schneider, H., Liniger, M., Schlunegger, F., 2008. Landsliding
and sediment flux in the Central Swiss Alps: A photogrammetric study of the Schimbrig
landslide, Entlebuch. Geomorphology 97, 392-406.
Seeger, M., Errea, M.P., Beguería, S., Arnáez, J., Martí, C., García-Ruiz, J.M., 2004.
Catchment soil moisture and rainfall characteristics as determinant factors for
discharge/suspended sediment hysteretic loops in a small headwater catchment in the Spanish
pyrenees. Journal of Hydrology 288, 299-311.
Sellers, P.J., Berry, J.A., Collatz, G.J., Field, C.B., Hall, F.G., 1992. Canopy reflectance,
photosynthesis, and transpiration. III. A reanalysis using improved leaf models and a new
canopy integration scheme. Remote Sensing of Environment 42, 187-216.
SEPSA, 2009. Boletín Estadístico Agropecuario 19. Secretaría Ejecutiva de Planificación
Sectorial Agropecuaria, San José, Costa Rica.
Sharda, V.N., Samraj, P., Samra, J.S., Lakshmanan, V., 1998. Hydrological behaviour of first
generation coppiced bluegum plantations in the Nilgiri sub-watersheds. Journal of Hydrology
211, 50-60.
Sharma, P.P., Gupta, S.C., Foster, G.R., 1993. Predicting Soil Detachment by Raindrops. Soil
Sci. Soc. Am. J. 57, 674-680.
Sharma, P.P., Gupta, S.C., Foster, G.R., 1995. Raindrop-Induced Soil Detachment and
Sediment Transport from Interrill Areas. Soil Sci. Soc. Am. J. 59, 727-734.
Sherman, L.K., 1932. Streamflow from rainfall by the unit-graph method. Engineering News
Record 108, 501-505.
Shuttleworth, W.J., Wallace, J.S., 1985. Evaporation from sparse crops-an energy
combination theory. Quarterly Journal of the Royal Meteorological Society 111, 839-855.
Sidle, R.C., Ziegler, A.D., Negishi, J.N., Nik, A.R., Siew, R., Turkelboom, F., 2006. Erosion
processes in steep terrain--Truths, myths, and uncertainties related to forest management in
Southeast Asia. Forest Ecology and Management 224, 199-225.
Siles, P., 2007. Hydrological processes (water use and balance) in a coffee (Coffea arabica L.)
monoculture and a coffee plantation shaded by Inga densiflora in Costa Rica. PhD Thesis.
Université Henri Poincaré, Nancy-I, Nancy, France, p. 153.
240
Siles, P., Harmand, J.-M., Vaast, P., 2010a. Effects of Inga densiflora on the microclimate of
coffee Coffea arabica L.) and overall biomass under optimal growing conditions in Costa
Rica. Agroforest. Syst. 78, 269-286.
Siles, P., Vaast, P., Dreyer, E., Harmand, J.-M., 2010b. Rainfall partitioning into throughfall,
stemflow and interception loss in a coffee (Coffea arabica L.) monoculture compared to an
agroforestry system with Inga densiflora. J. Hydrol. 395, 39-48.
Sinclair, T.R., Murphy, C.E., Knoerr, K.R., 1976. Development and Evaluation of Simplified
Models for Simulating Canopy Photosynthesis and Transpiration. Journal of Applied Ecology
13, 813-829.
Singh, V.P., 1995. Computer Models of Watershed Hydrology. Water Resources
Publications, Colorado, USA.
Sinoquet, H., Le Roux, X., Adam, B., Ameglio, T., Daudet, F.A., 2001. RATP: a model for
simulating the spatial distribution of radiation absorption, transpiration and photosynthesis
within canopies: application to an isolated tree crown. Plant, Cell & Environment 24, 395-
406.
Smerdon, B.D., Allen, D.M., Grasby, S.E., Berg, M.A., 2009. An approach for predicting
groundwater recharge in mountainous watersheds. Journal of Hydrology 365, 156-172.
Smith, R.E., Scott, D.F., 1992. The effects of afforestation on low flows in various regions of
South Africa. Water SA 18, 185-194.
Sorooshian, S., Gupta, V.K., 1983. Automatic calibration of conceptual rainfall-runoff
models: The question of parameter observability and uniqueness. Water Resour. Res. 19, 260-
268.
Southgate, D., Wunder, S., 2007. Paying for watershed services in Latin America: a review of
current initiatives. Office of International Research, Education, and Development (OIRED),
Virginia, USA, p. 27.
Stephenson, N.L., 1990. Climatic control of vegetation distribution: The role of the water
balance. The American Naturalist 135, 649-670.
Sun, H., Cornish, P.S., Daniell, T.M., 2001. Turbidity-based erosion estimation in a
catchment in South Australia. Journal of Hydrology 253, 227-238.
241
Taugourdeau, S., 2010. Indice foliaire d’un système agroforestier à base café, mesure,
dynamique et relation avec la production. Master FENEC: Université des Sciences et
Technologies du Languedoc, Montpellier, p. 12 + annexes.
Torri, D., Poesen, J., 1992. The effect of soil surface slope on raindrop detachment. CATENA
19, 561-578.
Torri, D., Sfalanga, M., Del Sette, M., 1987. Splash detachment: Runoff depth and soil
cohesion. CATENA 14, 149-155.
Tripathi, M.P., Raghuwanshi, N.S., Rao, G.P., 2006. Effect of watershed subdivision on
simulation of water balance components. Hydrol. Process. 20, 1137-1156.
Truman, C.C., Bradford, J.M., 1990. Effect of Antecedent Soil Moisture on Splash
Detachment Under Simulated Rainfall. Soil Science 150, 787-798.
UNESCO, 2007. Balance hídrico superficial de Costa Rica: período 1970-2002. Documento
técnico del PHI-LAC 10, Montevideo, Uruguay, p. 55.
USDA-ARS-NSL, 2003. RUSLE1.06c and RUSLE2. Internet
site:www.sedlab.olemiss.edu/rusle. U.S. Department of Agriculture, Agricultural Research
Service, National Sediment Laboratory.
USDA, 1999. Soil Taxonomy, A Basic System of Soil Classification for Making and
Interpreting Soil Surveys. Washington, USA.
Vaast, P., Beer, J., Harvey, C., Harmand, J.M., 2005. Environmental services of coffee
agroforestry systems in Central America: a promising potential to improve the livelihoods of
coffee farmers’ communities. Intregrated Management of Environmental Services in Human-
Dominated Tropical Landscapes. CATIE, Turrialba, Costa Rica.
van Kanten, R., Vaast, P., 2006. Transpiration of Arabica Coffee and Associated Shade Tree
Species in Sub-optimal, Low-altitude Conditions of Costa Rica. Agroforestry Systems 67,
187-202.
Van Lill, W.S., Kruger, F.J., Van Wyk, D.B., 1980. The effect of afforestation with
Eucalyptus grandis Hill ex Maiden and Pinus patula Schlecht. et Cham. on streamflow from
experimental catchments at Mokobulaan, Transvaal. Journal of Hydrology 48, 107-118.
242
van Oijen, M., Dauzat, J., Harmand, J.-M., Lawson, G., Vaast, P., 2010a. Coffee agroforestry
systems in Central America: I. A review of quantitative information on physiological and
ecological processes. Agroforestry Systems 80, 341-359.
van Oijen, M., Dauzat, J., Harmand, J.-M., Lawson, G., Vaast, P., 2010b. Coffee agroforestry
systems in Central America: II. Development of a simple process-based model and
preliminary results. Agroforestry Systems 80, 361-378.
Vanacker, V., Molina, A., Govers, G., Poesen, J., Deckers, J., 2007. Spatial variation of
suspended sediment concentrations in a tropical Andean river system: The Paute River,
southern Ecuador. Geomorphology 87, 53-67.
Vega, F.E., Rosenquist, E., 2001. The coffee berry borer and coffee research at the United
States Department of Agriculture. Proc. First ICO World Coffee Conference. International
Coffee Organization, London, UK.
Verbist, B., Poesen, J., van Noordwijk, M., Widianto, Suprayogo, D., Agus, F., Deckers, J.,
2010. Factors affecting soil loss at plot scale and sediment yield at catchment scale in a
tropical volcanic agroforestry landscape. CATENA 80, 34-46.
Verbist, B.J.P., Widayati, A., van Noordwijk, M., 2003. The link between land and water
prediction of sediment sources in a previously forested watershed in Lampung, Sumatra,
Indonesia. Modsim 2003: International Congress on Modelling and Simulation, Vols 1-4 -
Vol 1: Natural Systems, Pt 1; Vol 2: Natural Systems, Pt 2; Vol 3: Socio-Economic Systems;
Vol 4: General Systems. Univ Western Australia, Nedlands, pp. 560-565.
Vertessy, R.A., 1999. The impacts of forestry on streamflows: a review. In: J. Croke, P.L.
(Ed.), Forest management for water quality and quantity: proceedings of the Second Forest
Erosion Workshop. Cooperative Research Centre for Catchment Hydrology, Warburton,
Australia, pp. 93-109.
Wahl, T.L., Clemmens, A.J., Replogle, J.A., Bos, M.G., 2000. WinFlume - Windows-based
software for the design of long-throated measuring flumes. Fourth Decennial National
Irrigation Symposium. American Society of Agricultural Engineers, Phoenix, Arizona.
Walling, D.E., 1977. Assessing Accuracy of Suspended Sediment Rating Curves for a Small
Basin. Water Resources Research 13, 530-538.
243
Walling, D.E., 1983. The sediment delivery problem. Journal of Hydrology 65, 209-237.
Walling, D.E., Webb, B.W., 1982. Sediment availability and the prediction of storm-period
sediment yields. IAHS publication 137, 327-337.
Walling, D.E., Webb, B.W., Woodward, J.C., 1992. Some sampling considerations in the
design of effective strategies for monitoring sediment-associated transport. In: IAHS (Ed.),
Erosion and Sediment Transport Monitoring Programmes in River Basins. IAHS, Oslo,
Norway, pp. 279-288.
Wang, Y.P., Leuning, R., 1998. A two-leaf model for canopy conductance, photosynthesis
and partitioning of available energy I: Model description and comparison with a multi-layered
model. Agricultural and Forest Meteorology 91, 89-111.
Wass, P.D., Leeks, G.J.L., 1999. Suspended sediment fluxes in the Humber catchment, UK.
pp. 935-953.
Wass, P.D., Marks, S.D., Finch, J.W., Leeks, G.J.L., Ingram, J.K., 1997. Monitoring and
preliminary interpretation of in-river turbidity and remote sensed imagery for suspended
sediment transport studies in the Humber catchment. Science of The Total Environment 194-
195, 263-283.
Watson, D.A., Laflen, J.M., Franti, T.G., 1986. Estimating Ephemeral Gully Erosion. Paper
No. 86–2020. American Society of Agricultural Engineers, St. Joseph, USA.
Webb, B.W., Walling, D.E., 1982. The magnitude and frequency characteristics of fluvial
transport in a devon drainage basin and some geomorphological implications. CATENA 9, 9-
23.
Weiss, M., Baret, F., Smith, G.J., Jonckheere, I., Coppin, P., 2004. Review of methods for in
situ leaf area index (LAI) determination: Part II. Estimation of LAI, errors and sampling.
Agricultural and Forest Meteorology 121, 37-53.
White, M.A., Thornton, P.E., Running, S.W., Nemani, R.R., 2000. Parameterization and
Sensitivity Analysis of the BIOME-BGC Terrestrial Ecosystem Model: Net Primary
Production Controls. Earth Interact. 4, 1-85.
White, S.M., 1990. The Influence of tropical cyclones as soil eroding and sediment
transporting events. An example from the Philippines. IAHS publication 192, 259-269.
244
Whitehead, D., Beadle, C.L., 2004. Physiological regulation of productivity and water use in
Eucalyptus: a review. Forest Ecology and Management 193, 113-140.
Wigmosta, M.S., Vail, L.W., Lettenmaier, D.P., 1994. A distributed hydrology-vegetation
model for complex terrain. Water Resour. Res. 30, 1665-1679.
Wilk, J., Andersson, L., Plermkamon, V., 2001. Hydrological impacts of forest conversion to
agriculture in a large river basin in northeast Thailand. Hydrol. Process. 15, 2729-2748.
Wilson, K.B., Hanson, P.J., Mulholland, P.J., Baldocchi, D.D., Wullschleger, S.D., 2001. A
comparison of methods for determining forest evapotranspiration and its components: sap-
flow, soil water budget, eddy covariance and catchment water balance. Agricultural and
Forest Meteorology 106, 153-168.
Willmott, C.J., Ackleson, S.G., Davis, R.E., Feddema, J.J., Klink, K.M., Legates, D.R.,
O'Donnell, J., Rowe, C.M., 1985. Statistics for the Evaluation and Comparison of Models. J.
Geophys. Res. 90, 8995-9005.
Wischmeier, W.H., Johnson, C.B., Cross, V., 1971. A soil erodibility nomograph for farmland
and construction sites. Journal of Soil and Water Conservation 26, 189-193.
Wischmeier, W.H., Smith, D.D., 1960. A universal soil-loss equation to guide conservation
farm planning. Transactions of the 7th International Congress Soil Science, Brussels,
Belgium, pp. 418-425.
Wischmeier, W.H., Smith, D.D., 1965. Predicting rainfall-erosion losses from cropland east
of the Rocky Mountains. Guide for selection of practices for soil and water conservation,
Agricultural Handbook No. 282. Agricultural Research Service, USDA, Purdue Agricultural
Experiment Station, Washington, D.C., p. 44.
Wischmeier, W.H., Smith, D.D., 1978. Predicting Rainfall Erosion Losses: A Guide to
Conservation Planning. Agriculture Handbook No. 537. USDA / Science and Education
Administration, Washington, DC, p. 58.
Wunder, S., Engel, S., Pagiola, S., 2008. Taking stock: A comparative analysis of payments
for environmental services programs in developed and developing countries. Ecological
Economics 65, 834-852.
245
Xu, C.Y., Singh, V.P., 1998. A Review on Monthly Water Balance Models for Water
Resources Investigations. Water Resour. Manag. 12, 31-50.
Yapo, P.O., Gupta, H.V., Sorooshian, S., 1998. Multi-objective global optimization for
hydrologic models. Journal of Hydrology 204, 83-97.
Zaehle, S., Sitch, S., Smith, B., Hatterman, F., 2005. Effects of parameter uncertainties on the
modeling of terrestrial biosphere dynamics. Global Biogeochem. Cy. 19, GB3020.
Zbinden, S., Lee, D.R., 2005. Paying for Environmental Services: An Analysis of
Participation in Costa Rica's PSA Program. World Development 33, 255-272.
Zhang, L., Dawes, W.R., Walker, G.R., 1999. Predicting the effect of vegetation changes on
catchment average water balance. Technical Report 99/12. CRC for Catchment Hydrology,
Victoria, Australia.
Ziegler, A.D., Giambelluca, T.W., Sutherland, R.A., Nullet, M.A., Yarnasarn, S., Pinthong, J.,
Preechapanya, P., Jaiaree, S., 2004. Toward understanding the cumulative impacts of roads in
upland agricultural watersheds of northern Thailand. Agriculture, Ecosystems & Environment
104, 145-158.
Ziegler, A.D., Sutherland, R.A., Giambelluca, T.W., 2000. Runoff generation and sediment
production on unpaved roads, footpaths and agricultural land surfaces in northern Thailand.
Earth Surface Processes and Landforms 25, 519-534.
246
247
List of tables
!"#$%&'%#() %*+%),-"*#)%!./%0!1(!"#$)%(.%"(#2*3%451!.($1 ET AL.6%&7778' ......................................................... 25
!"#$%9'%-$!)31$/%:,/1*#*5(;%0!1(!"#$)6%1$;*1/%!./%5!<)%<$1(*/)'.................................................................. 75
!"#$%='%<!1!-$ $1%/$);1(< (*.6%1!.5$%+*1%*< (-(>! (*.6%*< (-(>$/%0!#3$%!./%1!.5$%+*1%)$.)( (0( ,%!.!#,)()'%
-&?%*1(5(.!#%&@A<!1!-$ $1%-*/$#6%-9?%)(-<#(+($/%BA<!1!-$ $1%-*/$#' ................................................... 97
!"#$%C'%)$.)( (0( ,%(./$D$)%+*1%$!;:%*+% :$%&@%;!#("1! (*.%<!1!-$ $1)%0)'% :$%.)%;*$++(;($. '%<$!1?%<$!1)*.%
<1*/3; %-*-$. %;*11$#! (*.%;*$++(;($. 6%)<$!?%)<$!1-!.%;*$++(;($. 6%)1;?%) !./!1/(>$/%1$51$))(*.%
;*$++(;($. %!./%)11;?%) !./!1/(>$/%1!.E%1$51$))(*.%;*$++(;($. )'%(./$D%0!#3$)%(.%"*#/%!1$%
) ! () (;!##,%)(5.(+(;!. %! %7FG%;*.+(/$.;$%#$0$#' ............................................................................... 108
!"#$%F'%;*-<!1()*.%*+%!..3!#%H! $1%"!#!.;$%;*-<*.$. )%4!)%G%*+%1!(.+!##8%+*1%/(++$1$. %) 3/($)%! %<#* %
!./%"!)(.%);!#$)6%(.% 1*<(;!#%1$5(*.)' ...................................................................................................... 112
!"#$%I?%<$1+*1-!.;$%*+% :$%)(D%5!<A+(##(.5%-$ :*/)%!<<#($/% *% :$%-())(.5%0!#3$)%*+% :$%*1(5(.!#% 31"(/( ,%
1$;*1/6%H( :%1$)<$; % *% H*%$0!#3! (*.%;1( $1(!?%$/%!./%0('................................................................ 138
!"#$%B'%;*-<*.$. )%*+%3.;$1 !(. ,%(.% :$%$) (-! (*.%*+%!..3!#%;3-3#! (0$%)$/(-$. %,($#/%!./%$J3! (*.)'
................................................................................................................................................................... 141
!"#$%K?%:,/1*#*5(;!#%!./%)$/(-$. %<1*<$1 ($)%*+% :$%9C%) *1-%$0$. )%1$;*1/$/%/31(.5%*.$%,$!1%*+%
*")$10! (*.)%! % :$%$D<$1(-$. !#%"!)(.'%R?%1!(.+!##6%I&@?%-!D(-3-%(. $.)( ,%(.%&@%-(.6%QPR?%) 1$!-+#*H%
<$!E%1! $6%B?%"!)$+#*H6%SRB?%)31+!;$%13.*++%! %"!)(.%);!#$6%SRR?%)31+!;$%13.*++%+1*-%!##% :$%"!)(.%
1*!/)6%S?%)3)<$./$/%)$/(-$. %;*.;$. 1! (*.6%Q?%) 1$!-+#*H6%Y?%$0$. %)$/(-$. %,($#/6%SR?%)31+!;$%
13.*++6%!+?%!51*+*1$) 1,%<#* %!./%-;?%-*.*;3# 31$%<#* '%!##%0!#3$)%+1*-%/(1$; %-$!)31$-$. )%!1$%
5(0$.%(.%1*-!.% ,<$6%H:(#$%;!#;3#! $/%*1%$) (-! $/%0!#3$)%!1$%<1$)$. $/%(.%( !#(;% ,<$' ................... 145
!"#$%7'%;*-<!1(.5%H! $1%,($#/%+1*-%) *1-%$0$. )% *%!..3!#%);!#$6%!./%+1*-%<#* % *%"!)(.%);!#$'% :$%) *1-%
)3--!1,%H!)%;*-<3 $/%",%!//(.5% :$%<1*<$1 ($)%*+% :$%9C%-!(.%) *1-%$0$. )%/31(.5% :$%*.$A,$!1%
-$!)31$-$. %<$1(*/'%R?%1!(.+!##6%QD?%) 1$!-+#*H%/$< :6%SRB?%"!)(.%)31+!;$%13.*++6%SR!+?%)31+!;$%
13.*++%! % :$%!51*+*1$) 1,%<#* 6%SR-;?%)31+!;$%13.*++%! % :$%-*.*;3# 31$%<#* 6%SR?%)31+!;$%13.*++'
................................................................................................................................................................... 153
!"#$%&@'%(.( (!#%<!1!-$ $1%0!#3$)%+*1% :$%;!#("1! (*.%*+% :$%+*31%-*/$#)% $) $/'..................................... 185
!"#$%&&%<$1+*1-!.;$%*+% :$%+*31%-*/$#)% $) $/%*.%;*. (.3*3)% (-$'%.)?%.!):A)3 ;#(++$%$++(;($.;,%
;*$++(;($. 6%;9-?%"*3./$/%.!):A)3 ;#(++$6%05?%5#*"!#%0*#3-$%$11*16%01?%1$#! (0$%0*#3-$%$11*1' ........ 190
!"#$%&9'%-$!.%0!#3$)%*+% :$%-*/$#%<$1+*1-!.;$)%$0!#3! $/%*.%9C%) *1-%$0$. )6%!;;*1/(.5% *%)(D%
$0!#3! (*.%;1( $1(!'% NS ?%$0$. A-$!.%.!):A)3 ;#(++$%$++(;($. ,%;*$++(;($. 6% 2MC ?$0$. A-$!.%
"*3./$/%.!):A)3 ;#(++$6% gV ?%$0$. A-$!.%5#*"!#%0*#3-$%$11*16% rV ?%$0$. A-$!.%1$#! (0$%0*#3-$%
$11*16%gP ?%$0$. A-$!.%5#*"!#%<$!E%$11*1%!./% rP ?%$0$. A-$!.%1$#! (0$%<$!E%$11*1'.......................... 195
248
249
List of figures
+(531$%&'%1$/3; (*.%/$)%/$"( )%-$)31$$%/!.)%;(.J%$D<$1($.;$)%/$%1$"*()$-$. %/$%"!))(.)%$.%!+1(J3$%/3%)3/'%
#$)%;*31"$)%)*. %!23) $$)%<*31%&@@G%/$%<#!. ! (*.%/!.)%#$%"!))(.%$ %#())$$)%<*31%#$%/$"( %!..3$#%
-*,$.%!0!. %#!%<#!. ! (*.%%4/L!<1$)%);* ET AL.6%9@@C8 ........................................................................... X
+(531$%9'%5!--$)%/$%<1*/3; (*.%/$%)$/(-$. )%<!1%/$)%"!))(.)%$.%!)($%/3%)3/A$) %$.%+*.; (*.%/3%)3") 1! %
5$*#*5(J3$%$ %/$%#L3 (#()! (*.%/$)% $11$)'%;! $5*1($)?%(6%+*1$ )6%51!.( 6%((6%+*1$ )6%%51$)%M%);:() $)N%(((6%
+*1$ )6%1*;:$)%0*#;!.(J3$)N%(06%+*1$ )6%-!1.$)N%06%+*1$ )%$D<#*( $$)%41(#?%$D<#*( ! (*.%!%+!("#$%(-<!; 8N%
0(6%+*1$ )%;*3<$$)6%1*;:$)%)$/(-$. !(1$)%4"!11$%(.+$1($31$?%-(;1*A"!))(.)8N%0((6%+*1$ )%;*3<$$)6%1*;:$)%
0*#;!.(J3$)N%0(((6%+*1$ )%;*3<$$)6%-!1.$)6%(D6%"!))(.)%-*,$.)%!%51!./)6%3 (#()! (*.%-(D $%/$)% $11$)6%
51!.( N%D6%(/$-6%1*;:$)%0*#;!.(J3$)N%D(6%(/$-6%1*;:$)%0*#;!.(J3$)%$ %-!1.$)N%D((6%31"!.()$%4"!11$%
(.+$1($31$8%6%$D<#*( ! (*.%-(.($1$%$ %#!%;*.) 13; (*.%/$%1*3 $)%4"!11$%)3<$1($31$8'4/L!<1$)%"13(2.>$$#6%
9@@C8............................................................................................................................................................ XI
+(531$%='%$D$-<#$)%/$%<1*/3; (*.%/$%)$/(-$. )%$.%)3)<$.)(*.%4)),8%$.%+*.; (*.%/$%#!%)3<$1+(;($%/3%"!))(.%4!8'%
4!8%3.$%1$#! (*.%.$5! (0$%$) %*")$10$$%$. 1$%!%$ %)),%<*31%<#3)($31)%"!))(.)%!% 1!0$1)%#$%-*./$6%
51*3<$)%$.%&K%1$5(*.)%4;:!J3$%;*31"$%1$<1$)$. $%3.$%1$5(*.8'%4"8%$D$-<#$%/$%1$#! (*.)%<*)( (0$)%
$. 1$%!%$ %)),%<*31%<#3)($31)%"!))(.)%51*3<$)%$.%&&%1$5(*.)%4/L!<1$)%/$%0$. $ ET AL.6%9@@B8' ............. XIII
+(531$%C'%;*-<!1!()*.%/$%#!%5!--$%-$)31$$%/$%<1*/3; (*.%/$%)$/(-$. )%4 %:!A&%!.A&8%!3%.(0$!3%/$%#!%<!1;$##$%
4@6@@C%!%@6@&9%:!8%$ %#$%.(0$!3%/3%"!))(.%4"!)86%<*31%#$)%"!))(.)%H!,%1(.5E(:%4H16%B=K%:!86%H!,% $"3%
4H 6%BB&%:!8%$ %<$ !(%H!,%4H<6%&=7&%:!8%/!.)%)3-"$12!,!6%$.%(./*.$)($%4/L!<1$)%0$1"() ET AL.6%9@&@8
....................................................................................................................................................................XIV
+(531$%F'%!8%$-<#!;$-$. %/3%"!))(.%/3%+#$30$%1$0$. !>*.%!3%;*) !%1(;!6%$.%!-$1(J3$%;$. 1!#$'%"8%<*)( (*.%
/3%"!))(.%$D<$1(-$. !#%!%#L(. $1($31%/3%"!))(.%/3%1$0$. !>*.'%;8%#$%"!))(.%$D<$1(-$. !#%/3%<1*2$ %
O%;*++$$A+#3D%P%/!.)%#!%+$1-$%/Q!J3(!1$)%/$) (.$%!%-$)31$1%#$)%;*-<*)!. $)%/3%"(#!.%:,/1(J3$'.......XVI
+(531$%I'%R:,/1*A)0! S6%#$%-*/$#$%$;*A:,/1*#*5(J3$%;*.;$< 3$#%<1*<*)$%<*31%#$%"!))(.%$D<$1(-$. !#6%
1$<$)$. $%;*--$%3.$%$. ( $%3.(J3$' ........................................................................................................XVII
+(531$%B'%<!1 ( (*.%/3%"(#!.%:,/1(J3$%)(-3#$$%<!1%O%:,/1*A)0! %P%/!.)%#$%"!))(.%$D<$1(-$. !#%/Q!J3(!1$)
..................................................................................................................................................................XVIII
+(531$%K%/$"( %/$%1(0($1$%Q%$ %;*.;$. 1! (*.%/$%)$/(-$. )%$.%)3)<$.)(*.%S%4#(5.$)%.*(1$)8%/!.)%#$%"!))(.%
$D<$1(-$. !#%<*31%3.$%<$1(*/$%/$%-$)31$%/L3.%!.'%3.%(. $10!##$%/$%;*.+(!.;$%7FG%4>*.$%51()$8%$) %
!++(;:$%!3 *31%/$)%0!#$31)%$) (-$$)'%TT%U%=@%-(.' ....................................................................................XX
+(531$%7'%)(-3#! (*.)%/$%#!%<!1 ( (*.%/$)%/$"( )%/$%1(0($1$%!3%;*31)%/$%/$3D%$0$.$-$. )%/$%+*1 $)%<#3($)'%
$0$.$-$. %/3%9CM@IM9@@7?%4!8%-&?%-*/$#$%)(-<#$6%4"8%-9?%-*/$#$%:,/1*A)0! %&@%<!1!-$ 1$)6%4;8%-=?%
-*/$#$%:,/1*A)0! %B%<!1!-$ 1$)6%$ %4/8%-C?%:,/1*A)0! %!0$;%13())$##$-$. %)31%1*3 $)'%$0$.$-$. %
/3%@9M&&M9@@7?%4$8%-&6%4+8%-96%458%-=6%4:8%-C'%QMEA?%/$"( %-$)31$%N%/$"( %/$%"!)$?%)$<!1! (*.%-!.3$##$%/3%
250
/$"( %/$%"!)$%N%QBF?%/$"( %/$%"!)$%-*/$#()$6%QSR?%13())$##$-$. %/$%)31+!;$%-*/$#()$6%QBF%V%QSR?% * !#%/$)%
/$"( )%-*/$#()$)'.......................................................................................................................................XXII
+(531$%&@'%!.5*) 31!%1$)$10*(1%)3++$1)%!.%$D $./$/%)$/(-$. ! (*.%<1*"#$-6%$. !(#(.5%#!15$%;#$!.(.5%;*) )'%
4;*31 $),%*+%(;$8 ......................................................................................................................................... 2
+(531$%&&'%1$#! (*.):(<%"$ H$$.%1$/3; (*.%(.%+*1$) %;*0$1%!./%(.;1$!)$%(.%"!)(.%H! $1%,($#/'%4+(531$%
$D 1!; $/%+1*-%"13(2.>$$#6%9@@C8............................................................................................................... 4
+(531$%&9'%1$/3; (*.)%(.%) 1$!-+#*H%!)%-$!)31$/%(.%+(0$%"!)(.%!++*1$) ! (*.%$D<$1(-$. )%(.%)*3 :%!+1(;!'%
:$%;310$)%!1$%);!#$/%+*1%&@@G%<#!. (.5%*+% :$%"!)(.%!./%)-** :$/% *% :$%-$!.%!..3!#%) 1$!-+#*H%
<1(*1% *%<#!. (.5'%4+(531$%$D 1!; $/%+1*-%);* ET AL.6%9@@C8.................................................................. 5
+(531$%&='%1!.5$)%(.%1$<*1 $/%"!)(.%)$/(-$. %,($#/)%(.%)*3 :$!) %!)(!%!)%!%+3.; (*.%*+%5$*#*5(;!#%)3") 1! $%
!./%#!./%3)$'%;! $5*1($)?%(6%+*1$) 6%51!.( $N%((6%+*1$) 6%)!./) *.$)M):!#$)N%(((6%+*1$) 6%0*#;!.(;)N%(06%
+*1$) 6%-!1#)N%06%#*55$/%41(#?%1$/3;$/%(-<!; %#*55(.58N%0(6%;#$!1$/6%)$/(-$. !1,%1*;E)%4#*H$1%"!1?%
-(;1*A"!)(.)8N%0((6%;#$!1$/6%0*#;!.(;)N%0(((6%;#$!1$/6%-!1#)N%(D6%-$/(3-A#!15$%"!)(.)6%-(D$/%#!./%3)$6%
51!.( $N%D6%(/$-6%0*#;!.(;)N%D(6%(/$-6%0*#;!.(;)%<#3)%-!1#)N%D((6%31"!.()$/%4#*H$1%"!186%-(.(.5%!./%1*!/%
"3(#/(.5%43<<$1%"!18'%4+(531$%$D 1!; $/%+1*-%"13(2.>$$#6%9@@C8 .............................................................. 7
+(531$%&C'%-!<%*+%;*) !%1(;!'%;*#*31$/%!1$!)%(./(;! $% :$%)$0$.%;*++$$%<1*/3;(.5%1$5(*.)%4+1*-%.*1H$) % *%
)*3 :$!) ?%.*1 :%>*.$6%*;;(/$. !#6%;$. 1!#%0!##$,6%R#*)%)!. *)S6%R 311(!#"!S6%R<$1$>%>$#$/W.S%!./%
R;* *%"13)S8'%/!1E%"1*H.%!1$!)%/$.* $%;*++$$%<#!. ! (*.)'%"#!;E% :(;E%#(.$)%/$#(-( % :$% :(1 ,%+*31%
/1!(.!5$%"!)(.)%*+%;*) !%1(;!'%51$$.%<*(. )%!./%1$/%#(.$)%(./(;! $% :$%<1$)$.;$%*+%:,/1*$#$; 1(;%
<#!. )%*1%<1*2$; )'%)*31;$)?%)(5(;!+$%A%(;!+X%!./%(;$%49@@78'................................................................ 12
+(531$%&F'%H! $1%"!#!.;$%<!1 ( (*.6%<1$)$. (.5% :$%-!(.%$;*A:,/1*#*5(;!#%<1*;$))$)%(.%!%/1!(.!5$%"!)(.'%R?%
1!(.+!##6%IN?%(. $1;$< (*.6%T?% 1$$% 1!.)<(1! (*.6%TH?% :1*35:+!##6%ST?%) $-+#*H6%EU?%)*(#%$0!<*1! (*.6%
RE?%1** %$D 1!; (*.6%SR?%)31+!;$%13.*++6%I?%(.+(# 1! (*.6%D?%/1!(.!5$6%IF?%(. $1+#*H%4.*.A)! 31! $/%
+#*H86%BF?%"!)$+#*H6%DP?%/$$<%<$1;*#! (*.%!./%Q?%) 1$!-+#*H' .............................................................. 17
+(531$%&I'%1! (*%T/PET%;!#;3#! $/%+1*-%)!<%+#*H%-$!)31$-$. )%(.%!.%*!E%) !./%!)%!%+3.; (*.%*+%1$#! (0$%
$D 1!; !"#$%H! $1%4REW8%;!#;3#! $/%+1*-%.$3 1*.%<1*"$%-$!)31$-$. )%4+1*-%"1X/!%!./%51!.($16%
&77I8'% H*%/! !%)$ )%!1$%1$<*1 $/?%LAIUI%4"#!;E%;(1;#$)8%!./%LAIUC'F%4*<$.% 1(!.5#$)8'% :$%/* $/%
#(.$%):*H)% :$%;1( (;!#%REW%4REWC8'%4+(531$%$D 1!; $/%+1*-%51!.($1 ET AL.6%&7778 ............................ 20
+(531$%&B'%$D!-<#$)%*+%)3)<$./$/%)$/(-$. %,($#/%4)),8%!)%!%+3.; (*.%*+%"!)(.%!1$!%4!8'%4!8%!%.$5! (0$%
1$#! (*.):(<%()%*")$10$/%"$ H$$.%!%!./%)),%+*1%)$0$1!#%"!)(.)%H*1#/H(/$6%51*3<$/%(.%)K%1$5(*.)%4$!;:%
;310$%1$<1$)$. )%!%1$5(*.8'%4"8%$D!-<#$%*+%<*)( (0$%1$#! (*.)%"$ H$$.%!%!./%)),%+*1%)$0$1!#%"!)(.)%
513<$/%(.%&&%1$5(*.)'%4+(531$%$D 1!; $/%+1*-%/$%0$. $ ET AL.6%9@@B8'..................................................... 38
+(531$%&K% 311(!#"!%1(0$1%"!)(.6%-$2Y!)%;1$$E%-(;1*A"!)(.%!./%;*++$$%<#!. ! (*.)'%4)*31;$%*+%;*++$$%
<#!. ! (*.)%-!<?%(;!+X8' ............................................................................................................................. 65
+(531$%&7'%!8%#*;! (*.%*+%1$0$. !>W.%1(0$1%"!)(.%(.%;*) !%1(;!6%;$. 1!#%!-$1(;!'%"8%<*)( (*.%*+%
$D<$1(-$. !#%"!)(.%(.)(/$%*+%1$0$. !>W.%"!)(.'%;8% :$%R;*++$$A+#3DS%$D<$1(-$. !#%"!)(.%(.%!J3(!1$)%
+!1-%!./%( )%$D<$1(-$. !#%)$ 3<% *%-$!)31$% :$%H! $1%"!#!.;$%;*-<*.$. )' ....................................... 66
251
+(531$%9@'%-$!.%-*. :#,%1!(.+!##%! %!J3(!1$)%+!1-%) ! (*.%4<$1(*/%&7B=A9@@78%;*-<!1$/% *% :$%1!(.+!##%
(.% :$%$D<$1(-$. !#%"!)(.%(.%9@@7' ............................................................................................................ 67
+(531$%9&%1!(.+!##%) ! (*.)%(.% :$%4!8%-(//#$%!./% :$%4"8%3<<$) %<!1 )%*+% :$%$D<$1(-$. !#%"!)(.'............... 68
+(531$%99?%4!8%#*.5% :1*! %+#3-$%! % :$%*3 #$ %*+% :$%$D<$1(-$. !#%"!)(.'%4"8%;!#("1! (*.%*+% :$%+#3-$%",%
<,5-,%;311$. %-$ $1%4;*31 $),%*+%(;$8' ................................................................................................... 69
+(531$%9=?%-$!)31(.5%Z%*.%!%;*++$$%)#*<$%H( :%/(0(.$19@@@%+/1'.................................................................... 70
+(531$%9C?%$//,%;*0!1(!.;$% *H$1%49I%-%:(5:86%1$;*1/(.5% :$%!; 3!#%$0!<* 1!.)<(1! (*.6%$.$15,%!./%;*9%
+#3D$)%*+% :$%;*++$$%!+%),) $-'%(.% :$%1$!16%*.$%;!.%)$$% :$% 311(!#"!%0*#;!.* ................................... 71
+(531$%9F?%#!(A9@@@%/$0(;$% *%-$!)31$%#(5: % 1!.)-( !.;$'........................................................................... 71
+(531$%9I?%!%<($>*-$ 1(;%H$##%!./%!%-(.(A/(0$1%)$.)*1%(.% :$%$D<$1(-$. !#%"!)(.' ......................................... 73
+(531$%9B?%#*;! (*.%*+% :$%"!;E);! $1%)$.)*1%(.% :$%+#3-$6%H( :%1$)<$; % *% :$%H! $1%(.+#*H'.................. 73
+(531$%9K?%$D<$1(-$. !#%<#* )%4!8%(.% :$(1%H:*#$%$D $.)(*.?% :$%):!/$/%<#* %4!+8%! %#$+ 6% :$%.*.A):!/$/%
<#* %4-;8%! %1(5: N%4"8%! % :$%#*H$1%$./%H:$1$% :$%53 $1)%;*##$; %H! $1%!./%<#* %$1*)(*.%4 :$%
<#!) (;%;!.*<,%*0$1% :$%53 $1)%:!)%"$$.%1$ 1!; $/%+*1% :$%<:* *8' ..................................................... 74
+(531$%97'% :$%#3-<$/%;*.;$< 3!#%:,/1*#*5(;!#%-*/$#%<1*<*)$/%+*1% :$%$D<$1(-$. !#%"!)(.'...................... 87
+(531$%=@'%A8%(.+(# 1! (*.%+1*-%)31+!;$%1$)$10*(1%!)%!%+3.; (*.%*+%)*(#%H! $1%;*. $. %(.% :$%.*.A)! 31! $/%
.*.A1** %1$)$10*(16%B8% :$%1! (*%R%U T%ET@A&%!)%!%+3.; (*.%*+%1$#! (0$%$D 1!; !"#$%H! $1%REW%!./%+*1%
/(++$1$. %0!#3$)%*+%LAI%4:$1$6%LAI&%[%LAI9%[%\%[%LAII86%C8%)31+!;$%13.*++%+1*-% :$%)31+!;$%1$)$10*(1%!)%!%
+3.; (*.%*+%( )%H! $1%;*. $. 6%D8%"!)$+#*H%+1*-% :$%!J3(+$1%1$)$10*(1%!)%!%+3.; (*.%*+%( )%H! $1%
;*. $. ' ...................................................................................................................................................... 89
+(531$%=&'%-$!)31$-$. )%(.% :$%$D<$1(-$. !#%"!)(.%+*1%9@@7?%A8%1!(.+!##%R%!./%) 1$!-+#*H%Q6%TTU=@%-(.6%B8%
#$!+%!1$!%(./$D%4LAI8%H( :(.%&%) /'%/$0'%;*.+(/$.;$%"!./)%451$,86%C8%1$+$1$.;$%4ET@8%!./%-$!)31$/%
4ETR8%/!(#,%$0!<* 1!.)<(1! (*.6%D8%)*(#%H! $1%;*. $. %!%H( :%*.$%) !./!1/%/$0(! (*.%"!1)%!./%E8%
H! $1% !"#$%#$0$#%Z%(.% :$%+*31%<($>*-$ $1)% :1*35:*3 % :$%"!)(.'%51!<:)%*.% :$%1(5: %)(/$%
;*11$)<*./% *%!% H*A-*. :%<$1(*/%423.$A23#,%9@@78%+*1%"$ $1%(##3) 1! (*.' ................................... 100
+(531$%=9'%H! $1%"!#!.;$%(.% :$%$D<$1(-$. !#%"!)(.%+*1%9@@7'%*.#,%-$!)31$/%0!#3$)%!1$%<1$)$. $/%:$1$'102
+(531$%=='%-*/$#%1$)3# )?%A8%-$!)31$/%0)'%-*/$##$/%) 1$!-+#*H%!./%B8%;*-<*.$. )%*+%-*/$##$/%
) 1$!-+#*H'%H! $1%#$0$#%(.% :$%-*/$#%1$)$10*(1)?%C8%)31+!;$%1$)$10*(16%D8%.*.A)! 31! $/%1** %
1$)$10*(1%4*++)$ %",%!R%CH% *%):*H% :$%!")*#3 $%H! $1%;*. $. %(.% :$%)*(#%#!,$186%E8%.*.A)! 31! $/%
.*.A1** %1$)$10*(1%4*++)$ %",%!R%DH8%!./%F8%!J3(+$1%1$)$10*(1'%51!<:)%*.% :$%1(5: %)(/$%;*11$)<*./% *%
!% H*A-*. :%<$1(*/%423.$A23#,%9@@78%+*1%"$ $1%(##3) 1! (*.' ............................................................ 103
+(531$%=C'%0!#(/! (*.%*+?%A8%/!(#,%$0!<* 1!.)<(1! (*.%ETRM6%B8%)*(#%H! $1%;*. $. %!%! % :$%&'I%-%1** A
1$)$10*(16%C8%H! $1% !"#$%#$0$#%Z%!./%D8%) 1$!-+#*H%Q%+*1%$(5: %-*. :)%*+%9@&@'%419?%/$ $1-(.! (*.%
;*$++(;($. 6%11-)$?%1$#! (0$%1** %-$!.%)J3!1$/%$11*18' ..................................................................... 105
+(531$%=F'%-$!)31$/%) 1$!-+#*H%!5!(.) %$-<(1(;!#%7FG%;*.+(/$.;$%(. $10!#' ............................................ 107
+(531$%=I'%)<$!1-!.%(./$D$)%"$ H$$.%Q%!./?%A8%!##% :$%<!1!-$ $1)?%"#!;E%:*1(>*. !#%#(.$)%(./(;! $% :$%7FG%
;*.+(/$.;$%(. $10!#%*3 %*+%H:(;:% :$%)<$!1-!.%()%)(5.(+(;!. 6%B8%<!1 ( (*.%<!1!-$ $1% 6%C8%)31+!;$%
252
/();:!15$%KB6%D8%(.+(# 1! (*.%1! $%! %+($#/%;!<!;( ,%FC6%E8%-!D(-3-%(.+(# 1! (*.%"6%F8%1** A1$)$10*(1%
/();:!15$%KC6%G8%.*.A1** %/();:!15$%KD6%H8% :1$):*#/%+*1%):!##*H%!J3(+$1%EX6%I8%):!##*H%!J3(+$1%
/();:!15$ KE96%J8%/$$<%!J3(+$1%/();:!15$%KE&%!./ K8%/$$<%<$1;*#! (*.%/();:!15$%KE='% :$%"#!;E%/* )%
(./(;! $% :$% (-$%) $<)%+*1%H:(;:% :$%)<$!1-!.%()%)(5.(+(;!. %! % :$%7FG%;*.+(/$.;$%#$0$#6%H:(#$%51$,%
/* )%(./(;! $%.*.A)(5.(+(;!. %;*11$#! (*.)'........................................................................................... 110
+(531$%=B'%1$#! (*.):(<)%"$ H$$.% 31"(/( ,%4#8%!./%)3)<$./$/%)$/(-$. %;*.;$. 1! (*.%4S86%+( $/%",%!8%
<*H$1%!./%"8%#(.$!1%;310$)%!./%/$.* $/%",% :$% :(;E%#(.$%(.%$!;:%51!<:'% :(.%#(.$)%1$<1$)$. % :$%7FG%
;*.+(/$.;$%"!./)%!"*3 % :$%1$51$))(*.%#(.$%4/* $/8%!./%+*1%(./(0(/3!#%<1$/(; (*.)%4/();*. (.3*3)8'%
19?%/$ $1-(.! (*.%;*$++(;($. ' ................................................................................................................ 139
+(531$%=K%H! $1%"!#!.;$%!./%)$/(-$. %,($#/%! %"!)(.%!./%<#* %);!#$)6%+*1% :$%) *1-%$0$. %*+%@FM@IM9@@7?%
!8%1!(.+!##%R6%"8%) 1$!-+#*H%Q6%( )%"!)$+#*H%"%!./%)3)<$./$/%)$/(-$. %;*.;$. 1! (*.%S%!./%;8%)31+!;$%
13.*++%SR%+*1%!51*+*1$) 1,%!+%!./%-*.*;3# 31$%-;%<#* )'%TT%U%&@%-(.' ............................................ 144
+(531$%=7?%) 1$!-+#*H%<$!E%1! $%4QPR8%+*1%9C%) *1-%$0$. )6%$D<#!(.$/%",%<#* %)31+!;$%13.*++%3./$1%;*++$$%
!51*+*1$) 1,%4!+8%!./%;*++$$%-*.*;3# 31$%4-;8%),) $-)'................................................................... 146
+(531$%C@?%"!)(.%)31+!;$%13.*++%4SRB8%+*1%9C%) *1-%$0$. )6%!)%+3.; (*.%*+%$0$. % * !#%1!(.+!##%4R86%!./%
:$(1%#(.E% *%)$/(-$. %,($#/%4Y8%!)%"3""#$%H(/ :' ................................................................................... 147
+(531$%C&'%"!)(.%)$/(-$. %,($#/%4Y8%+*1%9C%) *1-%$0$. )6%!;;*1/(.5% *% :$%"!)(.%)31+!;$%13.*++%4SRB8'...... 147
+(531$%C9%) 1$!-+#*H%Q%!./%)3)<$./$/%)$/(-$. %;*.;$. 1! (*.%S%4"#!;E%#(.$)8%(.% :$%$D<$1(-$. !#%"!)(.%
+*1% :$%*.$A,$!1%-$!)31$-$. %<$1(*/'%!%7FG%;*.+(/$.;$%(. $10!#%451$,%1$5(*.8%()%/()<#!,$/%!1*3./%
:$%$) (-! $/%0!#3$)'%TT%U%=@%-(.' ........................................................................................................... 149
+(531$%C=%!..3!#%;3-3#! (0$%H! $1%"!#!.;$%*+% :$%$D<$1(-$. !#%"!)(.6%+1*-%@BM@=M9@@7% *%@IM@=M9@&@'%
R?%1!(.+!##6%Q?%) 1$!-+#*H6%!$ ?%-$!)31$/%!; 3!#%$0!<* 1!.)<(1! (*.6%SR!+?%)31+!;$%13.*++%! %;*++$$%
!51*+*1$) 1,%<#* %!./%SR-;?%)31+!;$%13.*++%! %;*++$$%-*.*;3# 31$%<#* ' ......................................... 150
+(531$%CC'%!8%!..3!#%;3-3#! (0$%)$/(-$. %,($#/%(.% :$%$D<$1(-$. !#%"!)(.%4"#!;E%#(.$8%H( :%7FG%;*.+(/$.;$%
(. $10!#%451!,%1$5(*.8%+*1% :$%*.$A,$!1%-$!)31$-$. %<$1(*/'%"8%!..3!#%;3-3#! (0$%$1*)(*.%(.% :$%
!51*+*1$) 1,%4!+8%!./%-*.*;3# 31$%4-;8%$D<$1(-$. !#%<#* )' ........................................................... 151
+(531$%CF?%)31+!;$%13.*++%(.% :$%"!)(.%SRB%!)%!%+3.; (*.%*+%!8%)31+!;$%13.*++%*.% :$%1*!/)%SRR%!./%"8%)31+!;$%
13.*++%*.% :$%<#* )%SRPLOT'........................................................................................................................ 155
+(531$%CI?%!8%/() 1("3 (*.%!./%"8%)3-%*+% :$%)$/(-$. %,($#/%$) (-! $)%4! %=@%-(.% (-$%) $<86%",%) 1$!-+#*H%
/$;(#$%51*3<'%4+*1% :$%#$+ %51!<:?% :$%0$1 (;!#%);!#$% 13.;)% :$%#!15$) %0!#3$)%*+%)$/(-$. %,($#/%(.%
:$%&@ :%/$;(#$%51*3<6% :$%"*D%)(>$%1$<1$)$. )% :$%(. $1J3!1 (#$%1!.5$6% :$%H:()E$1)%$D $./% *% :$%
-*) %$D 1$-$%/! !%<*(. %H:(;:%()%.*%-*1$% :!.% :$%(. $1J3!1 (#$%1!.5$6%;(1;#$)%!1$%0!#3$)%"$,*./%
H:()E$1)%#(-( )8' ....................................................................................................................................... 158
+(531$%CB'%+*31%/(++$1$. %-*/$#%) 13; 31$)% *% $) ?%4!8%)(-<#$%-*/$#?%-&6%4"8%:,/1*A)0! %-*/$#%(.% H*%
/(++$1$. %0$1)(*.)?%-9%4&@%<!1!-$ $1)8%!./%-=%4B%<!1!-$ $1)86%!./%4;8%:,/1*A)0! %H( :%1*!/A
(. $1;$< (*.%-*/$#?%-C' ........................................................................................................................... 183
+(531$%CK'%1*!/A13.*++%+!; *1%!)%!%+3.; (*.%*+% :$%1!(.+!#%(. $.)( ,' ......................................................... 184
253
+(531$%C7'%<$1+*1-!.;$%*+% :$%+*31%-*/$#)% *%<1$/(; %) 1$!-+#*H%4/()<#!,$/%(.%#*5!1( :-(;%);!#$8%*.%
;*. (.3*3)% (-$6%"* :%!)% (-$%)$1($)%4#$+ %)(/$8%!./%!)% :$%-$!)31$/A-*/$##$/%<#* %41(5: %)(/$86%+*1?%
4!8%-&?%)(-<#$%-*/$#6%4"8%-9?%:,/1*A)0! %-*/$#%&@%<!1!-$ $1)6%4;8%-=?%:,/1*A)0! %-*/$#%B%
<!1!-$ $1)%!./%4/8%-C?%:,/1*A)0! %H( :%1*!/A(. $1;$< (*.%1*3 (.$' ............................................... 191
+(531$%F@'%<$1+*1-!.;$%*+% :$%+*31%-*/$#)% $) $/%*.%;*. (.3*3)% (-$6%(.%;*-<!1()*.% *% :$%-$!)31$/%
) 1$!-+#*H6%!)%!%;3-3#! (0$%/() 1("3 (*.%+3.; (*.'.............................................................................. 192
+(531$%F&'%-*/$#%<$1+*1-!.;$%*.%) *1-%$0$. )6%!;;*1/(.5% *%)(D%;1( $1(!?%4!8%.!):A)3 ;#(++$%.)6%4"8%
"*3./$/%.!):A)3 ;#(++$%;9-6%4;8%5#*"!#%0*#3-$%$11*1%056%4/8%1$#! (0$%0*#3-$%$11*1%016%4$8%5#*"!#%<$!E%
$11*1%<56%4+8%1$#! (0$%<$!E%$11*1%<1' ........................................................................................................ 194
+(531$%F9'%<$1+*1-!.;$%"*D<#* )%!;;*1/(.5% *% :$%+*31%-*/$#)% $) $/%*.%) *1-%$0$. )%+*1?%4!8%.!):A
)3 ;#(++$?%.)4I86%4"8%"*3./$/%.!):A)3 ;#(++$?%;9-4I86%4;8%5#*"!#%0*#3-$%$11*1?%054I86%4/8%1$#! (0$%
0*#3-$%$11*1?%014I86%4$8%5#*"!#%<$!E+#*H%1! $%$11*1?%<54I86%4+8%1$#! (0$%<$!E+#*H%1! $%$11*1?%<14I8'%
)*-$%*3 #($1)%!1$%.* %):*H.%5(0$.% :$%;:*)$.%1!.5$%+*1% :$%0$1 (;!#%!D()'% :$%"*D%)(>$%1$<1$)$. )% :$%
(. $1J3!1 (#$%1!.5$6% :$%H:()E$1)%$D $./% *% :$%-*) %$D 1$-$%/! !%<*(. 6%H:(;:%()%.*%-*1$% :!.%*.$%
(. $1J3!1 (#$%1!.5$6%;(1;#$)%!1$%0!#3$)%"$,*./%H:()E$1)%#(-( )'% :$%]%-!1E%(./(;! $)% :$%-$!.%0!#3$'
................................................................................................................................................................... 196
+(531$%F='%<$1+*1-!.;$%;3-3#! (0$%/() 1("3 (*.)%+*1% :$%+*31%-*/$#)% $) $/%*.%) *1-%$0$. )?%4!8%.!):A
)3 ;#(++$%;*$++(;($. %.)4I86%4"8%"*3./$/%.!):A)3 ;#(++$%;9-4I86%4;8%5#*"!#%0*#3-$%$11*1%054I86%4/8%
1$#! (0$%0*#3-$%$11*1%014I8%4$8%5#*"!#%<$!E%$11*1%<54I8!./%4+8%1$#! (0$%<$!E%$11*1%<14I8'% :$%
:*1(>*. !#%);!#$%1$+$1)% *% :$%;3-3#! (0$%+1$J3$.;,%*+%) *1-%$0$. )?% :$%+1$J3$.;,%@'F%(./(;! $)%!%
:!#+%*+% :$%$0$. )6%H:(#$%&%1$<1$)$. )%!##% :$-' ................................................................................... 197
+(531$%FC'%-*/$##$/%) 1$!-+#*H%!./%( )%<!1 ( (*.%*0$1% H*%1$<1$)$. ! (0$%) *1-%$0$. )'%$0$. %*+%
9CM@IM9@@7?%4!8%-&?%)(-<#$%-*/$#6%4"8%-9?%:,/1*A)0! %-*/$#%&@%<!1!-$ $1)6%4;8%-=?%:,/1*A)0! %
-*/$#%B%<!1!-$ $1)6%!./%4/8%-C?%:,/1*A)0! %H( :%1*!/%(. $1;$< (*.'%$0$. %*+%@9M&&M9@@7?%4$8%-&6%
4+8%-96%458%-=6%4:8%-C'%Q-$!?%-$!)31$/%) 1$!-+#*HN%"!)$+#*H?%-!.3!##,%)$<!1! $/%"!)$+#*HN%QBF?%
-*/$##$/%"!)$+#*HN%QSR?%-*/$##$/%)31+!;$%13.*++N%QBF%V%QSR?% * !#%-*/$##$/%) 1$!-+#*H' .............. 199
254
255
APPENDIX:
Paper accepted:
Modelling the hydrological behaviour of a coffee agroforestry
basin in Costa Rica
5^_`aA/`bcdef6%+'6%1fghidje6%*'6% b`%-dkj`6%5'6% dgcfgje`dg6%)'6%<lj`a6%!'6% mdn%*ko`n6%-'6%
0ddip6%<'6%1dhke`b6%"'6%:dj_dne6%2'%-'6%0fbpa6%-'6%"fnn`qfne6%2'%-'6%(_rdst6%<'6%u%-fgiid6%1'%
%
:vejfbfcv%dne%$djpt%)vip`_%)sk`ns`i6%&F6%=I7A=796%9@&&%
%
efk?&@'F&7CMt`iiA&FA=I7A9@&&%
%
tpph?MMwww'tvejfbA`djptAivipAisk'n`pM&FM=I7M9@&&%
Hydrol. Earth Syst. Sci., 15, 369–392, 2011www.hydrol-earth-syst-sci.net/15/369/2011/doi:10.5194/hess-15-369-2011© Author(s) 2011. CC Attribution 3.0 License.
Hydrology andEarth System
Sciences
Modelling the hydrological behaviour of a coffee agroforestry basinin Costa RicaF. Gomez-Delgado1,2, O. Roupsard3,4, G. le Maire1, S. Taugourdeau1, A. Perez4, M. van Oijen5, P. Vaast1,B. Rapidel4,6, J. M. Harmand1, M. Voltz7, J. M. Bonnefond8, P. Imbach4, and R. Moussa71CIRAD, UPR80, Avenue d’Agropolis, 34398 Montpellier, France2ICE, CENPE, 10032 San Jose, Costa Rica3CIRAD, UMR Eco&Sols – Ecologie Fonctionnelle & Biogeochimie des Sols & Agroecosystemes(SupAgro-CIRAD-INRA-IRD), 2 Place Viala, 34060 Montpellier, France4CATIE, 7170, 30501 Turrialba, Costa Rica5CEH Edinburgh, Bush Estate, Penicuik, EH26 0QB, UK6CIRAD, UMR SYSTEM (CIRAD-INRA-SupAgro), 2 Place Viala, 34060 Montpellier Cedex 1, France7INRA, UMR LISAH, 2 Place Viala, 34060 Montpellier, France8INRA, EPHYSE, BP 81, 33883 Villenave d’Ornon, FranceReceived: 30 April 2010 – Published in Hydrol. Earth Syst. Sci. Discuss.: 17 May 2010Revised: 1 December 2010 – Accepted: 20 December 2010 – Published: 28 January 2011
Abstract. The profitability of hydropower in Costa Ricais affected by soil erosion and sedimentation in dam reser-voirs, which are in turn influenced by land use, infiltrationand aquifer interactions with surface water. In order to fosterthe provision and payment for Hydrological EnvironmentalServices (HES), a quantitative assessment of the impact ofspecific land uses on the functioning of drainage-basins isrequired. The present paper aims to study the water bal-ance partitioning in a volcanic coffee agroforestry micro-basin (1 km2, steep slopes) in Costa Rica, as a first step to-wards evaluating sediment or contaminant loads. The mainhydrological processes were monitored during one year, us-ing flume, eddy-covariance flux tower, soil water profilesand piezometers. A new Hydro-SVAT lumped model isproposed, that balances SVAT (Soil Vegetation AtmosphereTransfer) and basin-reservoir routines. The purpose of sucha coupling was to achieve a trade-off between the expectedperformance of ecophysiological and hydrological models,which are often employed separately and at different spa-tial scales, either the plot or the basin. The calibration ofthe model to perform streamflow yielded a Nash-Sutcliffe(NS) coefficient equal to 0.89 for the year 2009, while the
Correspondence to: F. Gomez-Delgado([email protected])
validation of the water balance partitioning was consistentwith the independent measurements of actual evapotranspi-ration (R2 = 0.79, energy balance closed independently), soilwater content (R2 = 0.35) and water table level (R2 = 0.84).Eight months of data from 2010 were used to validate mod-elled streamflow, resulting in a NS= 0.75. An uncertaintyanalysis showed that the streamflow modelling was precisefor nearly every time step, while a sensitivity analysis re-vealed which parameters mostly affected model precision,depending on the season. It was observed that 64% of theincident rainfall R flowed out of the basin as streamflowand 25% as evapotranspiration, while the remaining 11% isprobably explained by deep percolation, measurement errorsand/or inter-annual changes in soil and aquifer water stocks.The model indicated an interception loss equal to 4% of R, asurface runoff of 4% and an infiltration component of 92%.The modelled streamflow was constituted by 87% of base-flow originating from the aquifer, 7% of subsurface non-saturated runoff and 6% of surface runoff. Given the lowsurface runoff observed under the current physical conditions(andisol) and management practices (no tillage, planted trees,bare soil kept by weeding), this agroforestry system on a vol-canic soil demonstrated potential to provide valuable HES,such as a reduced superficial displacement-capacity for fertil-izers, pesticides and sediments, as well as a streamflow regu-lation function provided by the highly efficient mechanisms
Published by Copernicus Publications on behalf of the European Geosciences Union.
370 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
of aquifer recharge and discharge. The proposed combina-tion of experimentation and modelling across ecophysiolog-ical and hydrological approaches proved to be useful to ac-count for the behaviour of a given basin, so that it can beapplied to compare HES provision for different regions ormanagement alternatives.
1 IntroductionThe ability of ecosystems to infiltrate rainfall, sustainaquifers, and avoid erosion is a key determinant for the pro-vision of hydrological environmental services (HES), espe-cially in the humid tropics where surface fluxes can be veryhigh (MEA, 2005). Woody plants and, in particular, agro-forestry (AF) systems associating shade trees and perennialcrops with deep root systems, are assumed to enhance theseHES in comparison to traditional intensive cropping systems(Siles et al., 2010a; Ataroff and Monasterio, 1997; Vaast etal., 2005). Costa Rica is renowned as a promoter of HESby charging water users for the HES they receive from landowners (e.g. forest conservation), focusing on water quality(Pagiola, 2008). Hydropower producers, generating 78% ofthe total electricity consumption in Costa Rica during 2008(ICE, 2009), are major HES payers. Coffee is one of themost traded agricultural commodities in the world employing100 million people (Vega and Rosenquist, 2001). In CostaRica, coffee accounted for 15% of the agricultural exportsin 2008 and covered 2% of the territory (SEPSA, 2009). Ascoffee plantations are present in the main basins used for hy-droelectric generation in Costa Rica, the possible trade-offsof the payment for HES from hydropower producers to cof-fee farmers become evident. Negotiation for these paymentsis facilitated between providers and purchasers when the ser-vice, and/or the impact of a given practice on the provisionof the service, are clearly evaluated. However, links betweenland use, management practices (like tree planting) and hy-drology in Costa Rica have not been thoroughly investigatedby quantitative research (Anderson et al., 2006). There is aneed of both, experimentation at the basin scale in order toevaluate the main hydrological processes, and of integratedmodelling to understand the behaviour of all water compart-ments, including hidden ones (e.g. the aquifer).The partitioning of the water balance (WB) is a pre-
requisite to evaluate HES such as infiltration, aquifer reg-ulation capacity, erosion control and contaminants reten-tion in coffee AF systems. Comprehensive WB stud-ies at basin scale, including closure verification by inde-pendent methods, have been carried out in the developedworld and for other land covers, like those reported byRoberts and Harding (1996), Dawes et al. (1997), Cebal-los and Schnabel (1998), Wilson et al. (2001) and Maeda etal. (2006). Some experimental basins are located in the trop-ics, like those in Brazil, Costa Rica, Guadeloupe and Panama
(Charlier et al., 2008; Fujieda et al., 1997; Genereux et al.,2005; Kinner and Stallard, 2004), but no coffee AF basinshave been equipped so far. Some reports are available forcoffee AF systems but at the plot level and for some par-ticular fluxes such as throughfall and stemflow (Siles et al.,2010b), tree and coffee transpiration (Dauzat et al., 2001;van Kanten and Vaast, 2006), surface runoff (Harmand et al.,2007), energy balance and latent heat flux (Gutierrez et al.,1994). To our knowledge, there is no comprehensive studyof the water balance partitioning of coffee AF systems at thebasin level, including the behaviour of the aquifer.Truly balanced combinations of hydrological and eco-
physiological experiments and models remain scarce, al-though they intrinsically carry a more realistic and compre-hensive representation of plant, soil and aquifer componentsat plot and basin scales. Most hydrological studies at basinscale use flumes for monitoring the streamflow and simplyestimate evapotranspiration (ET), which prevents a true veri-fication of the water balance closure.As in the tropics we assumed that ET, including the re-
evaporation of intercepted water (RIn), is an important com-ponent of the water balance, even for precipitations around3000mmyr−1, we decided to measure it directly by eddy-covariance (Roupsard et al., 2006; Baldocchi and Meyers,1998; Wilson et al., 2001), choosing a 0.9 km2 micro-basinembedded in a very homogeneous coffee AF plantation. Asan additional advantage, the eddy-covariance method can bevalidated itself by closing the energy balance (Falge et al.,2001).Lumped, conceptual rainfall-streamflowmodels have been
used in hydrology since the 1960s (e.g. Crawford and Lins-ley, 1966; Cormary and Guilbot, 1969; Duan et al., 1992;Bergstrom, 1995; Donigan et al., 1995; Havnø et al., 1995;Chahinian et al., 2005). These models consider the basin asan undivided entity, and use lumped values of input variablesand parameters. For the most part (for a review, see Fleming,1975; Singh, 1995), they have a conceptual structure basedon the interaction between storage compartments, represent-ing the different processes with mathematical functions todescribe the fluxes between the compartments. Most hydro-logical models simplify the ET component based on refer-ence ET routines (Allen et al., 1998) or using very empiri-cal, non-validated models for actual ET. For instance, manymodels assume that actual evapotranspiration equals the ref-erence ET, or a constant fraction of it, or a variable frac-tion that depends only on the soil water content in the rootzone. However, improper parameterization of the crop co-efficient may severely affect the parameterization of hydro-logical resistances and fluxes. In constrast, ecophysiologicalmodels may operate efficiently at plot level but miss the parti-tioning between lateral non-saturated (subsurface) runoff andvertical drainage, and the dynamics of water in aquifers andrivers. This is a major limitation for the assessment of HES,which is mainly desired at the basin scale.
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 371
Fig. 1. (a) Location of Reventazon river basin in Costa Rica, Central America. (b) Position of experimental basin inside of Reventazonbasin. (c) The “Coffee-Flux” experimental basin in Aquiares farm and its experimental setup to measure the water balance components.
In the present study we attempted to couple two lumpedmodels into a new integrative one, chosen to be scalableand parsimonious: a basin reservoir model similar to theCREC model (Cormary and Guilbot, 1969) and employingthe Diskin and Nazimov (1995) production function as pro-posed by Moussa et al. (2007a,b), and the SVAT model pro-posed by Granier et al. (1999). While the basin model wasconsidered appropriate for its simplicity and capacity to sup-port new routines, the SVAT model was chosen for its par-simony (three parameters in its basic formulation), its ro-bustness (uses simple soil and stand data in order to pro-duce model runs for many years, avoiding hydraulic param-eters that are difficult to measure and scale up), its abilityto quantify drought intensity and duration in forest stands,and for its successful past validation in various forest standsand climatic conditions, including tropical basins (Ruiz et al.,2010).This paper aims to explain and model the hydrological
behaviour of a coffee AF micro-basin in Costa Rica, as-sessing its infiltration capacity on andisols. The method-ology consists of experimentation to assess the main waterfluxes and modelling to reproduce the behaviour of the basin.First, we present the study site and the experimental design.Second, we develop a new lumped eco-hydrological modelwith balanced ecophysiological/hydrological modules (thatwe called Hydro-SVAT model). This model was tailored tothe main hydrological processes that we recorded (stream-flow, evapotranspiration, water content in the non-saturated
zone and water table level) and that are described in the sub-sequent sections. Third, we propose a multi-variable calibra-tion/validation strategy for the Hydro-SVATmodel so we cal-ibrate using the streamflow in 2009 and validate using the re-maining three variables in 2009 and the streamflow in 2010.Fourth, we make an uncertainty analysis to produce a confi-dence interval around our modelled streamflow values, anda sensitivity analysis to assess from which parameters thisuncertainty might come. Fifth, we attempt to simplify theinitially proposed model based on the results of the sensitiv-ity analysis. Finally, we discuss the main findings concerningthe water balance in our experimental basin.
2 The study site2.1 Location, soil and climateThe area of interest is located in Reventazon river basin, inthe Central-Caribbean region of Costa Rica (Fig. 1a and b).It lies on the slope of the Turrialba volcano (central vol-canic mountain range of the country) and drains to theCaribbean Sea. The Aquiares coffee farm is one of thelargest in Costa Rica (6.6 km2), “Rainforest AllianceTM”certified, 15 km from CATIE (Centro Agronomico Tropicalde Investigacion y Ensenanza). Within the Aquiares farm,we selected the Mejıas creek micro-basin (Fig. 1c) for the“Coffee-Flux” experiment. The basin is placed between the
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
372 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
coordinates −83◦44′39′′ and −83◦43′35′′ (West longitude),and between 9◦56′8′′ and 9◦56′35′′ (North latitude) and ishomogeneously planted with coffee (Coffea arabica L., varCaturra) on bare soil, shaded by free-growing tall Erythrinapoeppigiana trees. The initial planting density for coffee was6300 plants ha−1, with a current age >30 years, 20% canopyopenness and 2.5m canopy height. It is intensively man-aged and selectively pruned (20% per year, around March).Shade trees have a density of 12.8 trees ha−1, with 12.3%canopy cover and 20m canopy height. The experimentalbasin has an area of 0.9 km2, an elevation range from 1020up to 1280m a.s.l. and a mean slope of 20%. Permanentstreams extend along 5.6 km, implying a drainage density of6.2 kmkm−2. The average slope of the main stream is 11%.According to the classification byMora-Chinchilla (2000),
the experimental basin is located along a 1.3 km wide stripof volcanic avalanche deposits, characterized by chaotic de-posits of blocks immersed in a matrix of medium-to-coarsesand, which is the product of the collapse of the south-eastern slope of Turrialba volcano’s ancient crater. The gen-eral classification given by the geological map of Costa Rica(MINAE-RECOPE, 1991) describes the general stratigraphyas shallow intrusive volcanic rocks, and the particular regionas proximal facies of modern volcanic rocks (Quaternary),with presence of lava flows, agglomerates, lahars and ashes.Soils belong to the order of andisols according to the USDAsoil taxonomy (USDA, 1999), which are soils developingfrom volcanic ejecta, under weathering and mineral trans-formation processes, very stable, with high organic mattercontent and biological activity and very large infiltration ca-pacities.According to Koppen-Geiger classification (Peel et al.,
2007), the climate is tropical humid with no dry seasonand strongly influenced by the climatic conditions in theCaribbean hillside. The mean annual rainfall in the studyregion for the period 1973–2009 was estimated as 3014mmat the Aquiares farm station (Fig. 2). At the experimentalbasin the rainfall in 2009 (3208mm) was close to the annualmean, but showed a monthly mean deviation of ±100mmaround the historical regime. Mean monthly net radiationranged in 2009 from 5.7 to 13.0MJm−2 d−1, air tempera-ture from 17.0 to 20.8 ◦C, relative humidity from 83 to 91%,windspeed at 2m high from 0.4 to 1.6m s−1 and Penman-Monteith reference evapotranspiration (Allen et al., 1998)from 1.7 to 3.8mmd−1.2.2 Experimental setupThe “Coffee-Flux” experimental basin and instrument lay-out was designed to trace the main water balance compo-nents employing spatially representative methods (Fig. 1c).It is part of the FLUXNET network for the monitoring ofgreenhouse gases of terrestrial ecosystems. The hydrologi-cal measurements were recorded from December 2008 up toFebruary 2010.
Fig. 2. Mean monthly rainfall at Aquiares farm station (period1973–2009) compared to the rainfall in the experimental basin in2009.
Rainfall and climate: rainfall was monitored at 3m aboveground in the middle of 3 transects of the basin, using threelab-intercalibrated ARG100 tipping-bucket (R. M. Young,MI, USA) connected to CR800 dataloggers (Campbell Sci-entific, Shepshed, UK), and integrated every 10min. Otherclimate variables were logged on top of the eddy-flux towerwith a CR1000, every 30 s, integrated half-hourly and using:Net radiation: NR-Lite (Kipp and Zonen, Delft, The Nether-lands); PPFD: Sunshine sensor BF3 (Delta-T devices Ltd,UK); temperature and humidity: HMP45C in URS1 shel-ter (Campbell Scientific); wind-speed and direction: 03001Wind Sentry (R. M. Young, MI, USA). The theoretical evap-otranspiration from a wet grass placed under local climateconditions, ET0, was computed in accordance with Allen etal. (1998).Streamflow: a long-throated steel flume (length: 3.9m;
width: 2.8m; height: 1.2m) was home-built to measure thestreamflow at the outlet of the experimental basin, to recordup to 3m3 s−1, the maximum estimated discharge for thestudy period from an intensity-duration-frequency analysis.The flume was equipped with a PDCR-1830 pressure trans-ducer (Campbell Scientific) to record water head at gaugepoint (30 s, 10min integration), while the rating curve wascalculated considering the geometric and hydraulic proper-ties of the flume usingWinflume software (Wahl et al., 2000).A validation of the rating curve was made successfully usingthe salt dilution method as well as a pygmy current meter.Soil water content: a frequency-domain-reflectometry
portable probe (FDR Diviner2000, Sentek Pty Ltd) was usedto survey 20 access tubes distributed in the three studytransects to provide the mean volumetric soil water in thebasin. The sensor measures at 10 cm intervals, reaching atotal depth of 1.6m. A measurement campaign through the20 sites was carried out every week. The sensors were cal-ibrated by digging sampling pits in the vicinity of six testtubes, to obtain the actual volumetric soil water content fromgravimetric content and dry bulk density.Evapotranspiration: the actual evapotranspiration from the
soil, coffee plants and shade trees was measured at refer-ence height (26m) on the eddy-covariance tower, similarly
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 373
Table 1. Measured hydrologic variables, record and gaps periods.
Variable Frequency of Measurement Data gaps Gap filling methodmeasurement period
(2009)Rainfall R 10min 1 Jan–31 Dec* No –StreamflowQ 10min 1 Jan–18 Jul 18 Jul–23 Jul Using the current model. Peak estimation by peakflow/
23 Jul–31 Dec* peak rainfall analysisMeasured 20Hertz 4 Mar–17 Jul 1 Jan–4 Mar Penman-Monteith model adjusting the canopyevapotranspiration ETR 30 Aug –31 Dec 17 Jul–30 Aug conductanceSoil water content θ 7 to 15 days 2 Apr–7 Dec 1 Jan–2 Apr –Water table level z 30min 2 Jun–31 Dec 1 Jan–2 Jun –
* Additional information corresponding to the period January–August 2010 was recorded and used for model validation purposes.
to Roupsard et al. (2006). 3-D wind components and tem-perature were measured with a WindMaster sonic anemome-ter (Gill Instruments, Lymington, UK) at 20Hz. H2O fluc-tuations were measured with a Li-7500 open path (LiCor,Lincoln, NE, USA). Raw data were collected and pre-processed by “Tourbillon” software (INRA-EPHYSE, Bor-deaux, France) for a time-integration period of 300 s, thenpost-processed using EdiRe software (University of Edin-burgh, UK) into half-hourly values and quality checked. Avalidation was made by direct comparison of the measurednet radiation Rn with the sum of sensible heat flux (H )and latent heat flux (λE): at daily time step, this yieldedH + λE = 0.92 Rn (R2 = 0.93) which was considered suffi-ciently accurate to assume that advection effects on λE couldbe neglected here. Due to lighting and sensor breakdown,45 days of data were lost between July and August 2009.To gap-fill the missing period we used the Penman-Monteithmodel, whose canopy conductance was adjusted using mea-sured values.Leaf Area Index (LAI): the coffee light transmittance was
measured monthly in diffuse light conditions, for five rings atdifferent zenital angles (LAI2000, Li-COR Corvallis, USA),along three 50m-long transects through the flux tower plot,similarly to Roupsard et al. (2008). Effective coffee LAI,obtained from this light transmittance, was converted intoactual LAI according to Nilson (1971), using a ratio of ef-fective to actual LAI that was estimated from a dedicatedcalibration. The actual coffee LAI was measured directlyon a small plot by counting total leaf number of 25 coffeeplants, measuring leaf length and width every 20 leaves andusing empirical relationships between leaf length and widthand leaf area (LI-3100C, Li-COR) (R2> 0.95). On the samesmall plot, the effective LAI was measured with LAI2000.The ratio of effective to actual LAI was then calculated onthis small plot (1.75) and was considered to be constant withtime and space in the micro-basin, allowing the estimation
of the actual LAI on the three LAI2000 transects. The LAIfor shade trees was estimated using their crown cover pro-jection (on average 12.3% over the whole basin) observed ona very high resolution panchromatic satellite image (World-View image, February 2008, 0.5m resolution). As we didnot have measurements of LAI for shade trees, we consid-ered this LAI in the order of magnitude of coffee LAI on acrown-projected basis, and therefore we multiplied the ac-tual coffee LAI measured on transects by 1.123, to estimatethe ecosystem LAI (tree and coffee). In order to monitorthe time-course of ecosystem LAI at the basin scale, wecombined these ground measurements with time series ofremotely-sensed images. We used time series of NormalizedDifference Vegetation Index (NDVI) from Moderate Reso-lution Imaging Spectroradiometer (MODIS) data productsMOD13Q1 and MYD13Q1 (16-Day composite data, 250mresolution). NDVI is known to be correlated with the greenLAI if it is low, for most ecosystems (Rouse et al., 1974).Twenty-three MODIS pixels covering the experimental basinwere selected, and their NDVI time series were downloaded.We filtered the raw NDVI time series according to qualitycriterion given in the MODIS products, and we adjusted asmooth spline function on it as in Marsden et al. (2010).Then, a linear regression between the smoothed NDVI of thepixel including the flux tower and ground values of actualecosystem LAI was calibrated (R2 = 0.69). This regressionwas used on other pixels of the basin, and averaged to havethe annual time-course of actual LAI.Water table level: four piezometric wells measuring up
to 4m depth were built in the three main transects of study.They were equipped with pressure transducers (Mini-Divers,Schlumberger Water Services) that measure and record thewater table level every 30min.Period of measurement, data gaps and gap-filling: the
recording information is given in Table 1 for the five hydro-logic variables. The frequency of measurement varies, but
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
374 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
is finally calculated at the 30min time step (except for soilwater content that is a non-continuous measurement). Whengaps are present in the measurements, a gap filling methodwas applied.
3 Hydro-SVAT lumped modelWe designed a lumped, five-reservoir-layer eco-hydrologicalmodel in order to predict the water balance (WB) partition-ing (stocks and fluxes) at the scale of the entire basin. It isbased on the water balance models developed by Moussa etal. (2007a,b) and Granier et al. (1999), and built to reproducethe main hydrological processes measured at the experimen-tal basin, which will be presented in Sect. 4. The model ofMoussa et al. (2007a,b) works at the basin scale and simu-lates the ecosystem evapotranspiration rather roughly, whilethe one of Granier et al. (1999) works at the plot scale and to-tally ignores the lateral water fluxes through the soil and therole of the basin aquifers. The main novelties of the Hydro-SVAT model with respect to the model structure of Moussaet al. (2007a,b) are the inclusion of a land cover reservoir toseparate the intercepted rainfall from the combined through-fall/stemflow component, and the partition of non-saturatedsoil into two reservoirs, one with and one without roots ofplants and trees. The first of these innovations intends totake into account the non-negligible interception loss in cof-fee AF systems, as reported by Jimenez (1986), Harmand etal. (2007) and Siles (2007). The second new addition to themodel is to better represent the water dynamics in the non-saturated soil, given that only its upper layer will lose humid-ity by root extraction. The water balance model of Granieret al. (1999) is incorporated in this superficial reservoir butin a simplified form, so that both, the root distribution andthe soil porosity, are homogeneous through the vertical, non-saturated profile. Hence, the water content in this reservoir isthe state variable linking our two parent models.The modelling hypotheses governing the model architec-
ture were: (a) the interception loss component is not negligi-ble in the WB and is a function of rainfall intensity, (b) in-filtration is a function of the soil water content in the non-saturated reservoirs, (c) evapotranspiration is a significantcomponent in the WB and is best described using a SVATmodel that couples evapotranspiration to root water extrac-tion from the soil, (d) the aquifer has a higher discharge rateabove a threshold level, (e) any lateral inputs to the basinssystem are considered negligible in comparison to rainfallinputs, and (f) there is a net water outflow from the system asdeep percolation.The model was implemented using Matlab® V. R2007a
(The MathWorks Inc., USA).
3.1 Model structureThe model structure is presented in Fig. 3. The next threesections will describe the model structure according to itsthree major routines and five layers. The first layer is called“land cover reservoir” and separates the total rainfall intoan intercepted loss and a joint throughfall/stemflow com-ponent. The second layer or “surface reservoir” regulatesthe surface runoff. The infiltration process from the secondlayer is controlled by the joint water content at the third andfourth layers, called “non-saturated root reservoir” and “non-saturated non-root reservoir”, respectively. The evapotran-spiration flux is calculated at the “non-saturated root reser-voir”, while both non-saturated layers control the drainage,the percolation and the non-saturated runoff processes. Thefifth and last layer is the “aquifer reservoir”, which deter-mines the baseflow and the deep percolation. Finally, wewill explain the sum of the total runoff and baseflow com-ponents and the routing procedure to generate the modelledstreamflow. Let A(t), B(t), C(t), D(t) and E(t) [L] be thewater levels at time t in the five reservoirs A, B, C,D and E,respectively (or land cover reservoir, surface reservoir, nonsaturated root reservoir, non-saturated no-root reservoir andaquifer reservoir). Let AX, BX, CX, DX and EX [L] be thewater levels corresponding to the maximum holding capaci-ties for the five reservoirs.3.1.1 Infiltration and actual evapotranspiration(a) InfiltrationThe infiltration process i [LT−1] occurs from the secondlayer (surface reservoir) to the third one (non-saturated rootreservoir), and eventually to the fourth one (non-saturatednon-root reservoir) when i fills the third one. The infiltra-tion capacity fi(t) [LT−1] is a state variable that depends onthe water level in the non-saturated root reservoir, given byC(t). In addition to CX (maximum) we define CF as the wa-ter level in the root reservoir at field capacity. Then, fi(t) iscalculated as (see Fig. 4a):If C(t) < CF then fi (t) = f0 + (fc − f0) C(t) C−1F (1)If C(t) ≥ CF then fi (t) = fc (2)where f0 [LT−1] is the maximum infiltration capacity (f0=α fc) and fc [LT−1] is the infiltration rate at field capacity.The infiltration i both modifies and depends on B ′(t), whichis the water availability in the second reservoir before i isextracted, according to:If B ′(t) 1t−1 < fi (t) then i = B ′(t) 1t−1 (3)and B(t) = 0If B ′(t) 1t−1 ≥ fi (t) then i = fi (t) (4)and B(t) = B ′(t) − fi (t) 1t
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 375
Fig. 3. The lumped conceptual hydrological model proposed for the experimental basin.
The infiltration module calculates the infiltration i as outputvariable, using the state variables B(t), C(t) andfi(t). Fourparameters (CX, CF, fc and α) are demanded.(b) EvapotranspirationThe evapotranspiration component ET [LT−1] acts directlyon the third layer (non-saturated root reservoir) and is thesum of Eu [LT−1] the understory and soil evaporation, andof T [LT−1] the transpirational water uptake by roots.ET = Eu + T (5)According to Shuttleworth and Wallace (1985), the frac-tion of total evapotranspiration originating from the plants isclose to 100% of the total evapotranspiration of the ecosys-tem when LAI> 3 an when the soil is not saturated at itssurface, which was always the case in our study. We thus as-sumed for simplicity that Eu, the evaporation from the soil,was nil.
Transpiration T is obtained by solving T from the slightlymodified ratio: r = T ET−10 [dimensionless] proposed byGranier et al. (1999). We substituted the original Penmanpotential evapotranspiration PET in that ratio by the Penman-Monteith reference evapotranspiration ET0 [LT−1] (Allen etal., 1998). While ET0 was calculated at each time step 1t ,we estimated r as a function of the relative extractable waterREW(t) [dimensionless], a state variable given by Granier etal. (1999) as:REW(t) = C(t) C−1F (6)The REW(t) is linked to the soil water content according to:
REW(t) =θ (t) − θrθf − θr
(7)
with θ(t): volumetric soil water content [L3L−3] at time t , θr:residual soil water content [L3L−3] and θf: soil water contentat field capacity [L3L−3].
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
376 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
The parameter REWc [dimensionless] is the criticalREW(t) below which the transpiration of the system beginsto decrease. Figure 4b shows an example of some r curvesas a function of REW(t). Each curve can be defined only byREWc and the rmLAI, a maximum value for the ratio r thatdepends on the LAI of the system as:rmLAI = LAI LAI−1X rm (8)where LAIX is the maximum measured LAI during the mod-elling period and rm is a parameter indicating the maximumratio T ET−10 that can be found in this system. Then:If REW(t) < REWc then r = rmLAI REW(t) REW−1c (9)If REW (t) ≥ REWc then r = rmLAI (10)Finally, we find the transpiration as T = rET0. The total mod-elled evapotranspiration including the interception loss, canbe calculated as: ETRm =Eu + T +RIn, with RIn [LT−1] be-ing the intercepted/evaporated rainfall loss that will be ex-plained in the next section. Hence, ETRm can be directlycompared to the evapotranspiration that we measured at theflux tower.This module provides the evapotranspiration ET as a func-
tion of the state variable C(t), two input variables (LAI andET0) and three parameters (CX, REWc and rm).3.1.2 Water balance in the model reservoirs(a) Land cover reservoirThe first layer of the model, denoted “land cover reservoir”,represents the soil cover in the basin and controls the parti-tion of the total incident rainfall R [LT−1] into intercepted(then evaporated) rainfall loss RIn [LT−1] and the combinedtroughfall/stemflow RTS [LT−1]. A simple water balance ofthis reservoir is established to calculate a proxy A′(t) [L]of the final water level A(t) for each time step t , by addingthe incident rainfall R and subtracting the Penman potentialevapotranspiration PET [LT−1] from the existing land coverhumidity level A(t−1):A′(t) = A(t − 1) + R − PET (11)We calculated the water level A(t) in this reservoir as well asRTS and RIn by differentiating three cases:If A′(t) ≤ 0 then A(t) = 0 and (12)RIn = A(t − 1) + R and RTS = 0If 0 < A′(t) < AX then A(t) = A′(t) and (13)RIn = PET and RTS = 0If A′(t) ≥ AX then A(t) = AX and (14)RIn = PET and RTS = A′(t) − AXThe land cover module calculates at each time t the waterlevel A(t) as a state variable, demanding two input variables(R and PET) and one parameter (AX). It yields the partitionof R into RIn and RTS.
Fig. 4. (a) Infiltration from surface reservoir as a function of soilwater content in the non-saturated non-root reservoir. (b) The ratior = T ET−10 as a function of relative extractable water REW and fordifferent values of LAI (here, LAI1>LAI2> ...>LAIi ). (c) Sur-face runoff from the surface reservoir as a function of its water con-tent. (d) Baseflow from the aquifer reservoir as a function of itswater content.
(b) Surface reservoirThe second layer is called “surface reservoir” and acts as asheet top soil with a given roughness and surface runoff de-laying properties. The water balance in this surface reservoirfor a given interval 1t is:B(t) = B(t − 1) + RTS − QB1 − QB2 − i (15)where RTS [LT−1] is the combined throughfall/stemflowcomponent from the previous layer andQB1 andQB2 [LT−1]are the non-immediate and immediate surface runoffs calcu-lated as:QB1 = kB B(t) (16)where kB [T−1] is a discharge parameter, and:If B(t) ≤ BX thenQB2 = 0 (17)If B(t) > BX thenQB2 = [B(t) − BX] 1t−1 (18)If QB2 > 0 then the water level B(t) is reset to BX. Theinfiltration i [LT−1] is a function of the water content in thethird layer and it is the last component to be evaluated in thesurface reservoir.This surface reservoir module calculates the water level
B(t) as a state variable and demands one input variable(RTS), and two parameters for the reservoir (BX and kB).It produces three output variables: i, QB1 and QB2. Thetwo latter variables constitute the surface runoff in the basin(Fig. 4c).
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 377
(c) Non-saturated root reservoirThe “non-saturated root reservoir” is the third layer of themodel and it represents a soil layer with presence of root sys-tems from trees and plants. The water balance here is:C(t) = C(t − 1) + i − ET − d1 − d2 − QC (19)where C(t) [L] is the state variable of the water level at agiven time t , i is the infiltration from the second layer, ET[LT−1] is the evapotranspiration, d1 and d2 [LT−1] are thenon-immediate and immediate drainages to the fourth layer,respectively andQC [LT−1] is the non-saturated runoff fromthe root reservoir.There will be immediate drainage d2 if at anytime the
RTS component fills the reservoir above CX. Then d2 =[C(t)−CX] 1t−1 goes to the fourth layer and C(t) is resetto CX. Both non-immediate drainage d1 and non-saturatedrunoff QC occur whenever C(t) is higher than the field ca-pacity threshold CF [L] according to:ρ = [C(t) − CF] kC (20)where ρ [LT−1] is the total outflow capacity in this reservoirand kC [T−1] a discharge parameter. The partition of ρ ind1 and QC depends on a parameter β [dimensionless], with0< β < 1. Then:d1 = (1 − β) ρ andQC = β ρ (21)The root soil module calculates the water level C(t) as statevariable using two input variables (i and ET) and four pa-rameters (CX, CF, kC and β). It provides three outputs (d1,d2 andQC).(d) Non-saturated non-root reservoirThe fourth layer of the model is denoted “non-saturated non-root reservoir” and represents a soil layer with total absenceof root systems and hence, of root water extraction. The wa-ter balance here is given by:D(t) = D(t − 1) + d1 + d2 − QD − g1 − g2 (22)where D(t) [L] is the state variable of the water level ata given time t , d1 and d2 [LT−1] are the non-immediateand immediate drainages from the third layer, respectively;QD [LT−1] is the non-saturated runoff from the non-rootreservoir and g2 and g1 [LT−1] are the immediate and non-immediate percolation to the fifth model layer, respectively.Immediate percolation g2 will be produced if at anytime
the drainage (d2 and/or d1) fills the reservoir aboveDX. Theng2 = [D(t)−DX] 1t−1 moves to the aquifer reservoir andD(t) is reset to DX. Both non-immediate percolation g1 andnon-saturated runoffQD occur wheneverD(t) is higher thanthe field capacity threshold DF [L]:η = [D(t) − DF] kD (23)
where η [LT−1] is the total outflow capacity of this reservoirand kD [T−1] a discharge parameter. The partition of η in g1andQD depends on the parameter β [dimensionless]. Then:g1 = (1 − β) η andQD =β η (24)The non-root soil module calculates the water level D(t) asstate variable using two input variables (d1 and d2) and fourparameters (DX, DF, kD and β). It provides three outputs(g1, g2 andQD).(e) Aquifer reservoirA fifth layer called “aquifer reservoir” represents the ground-water system and controls baseflow and deep percolation.The reservoir is composed by a shallow aquifer that actswhenever the water level in the reservoir is higher than EX,and by a deep aquifer with a permanent contribution. Thewater balance here is:E(t) = E(t − 1) + g1 + g2 − QE1 − QE2 − DP (25)where E(t) [L] is the state variable of the water level ata given time t , g1 and g2 [LT−1] are respectively the non-immediate and immediate percolation from the fourth layer,QE1 andQE2 [LT−1] are the baseflow from deep and shallowaquifers respectively (Fig. 4d), and DP [LT−1] is the deeppercolation.If E(t) ≤ EX thenQE1 = kE1 E(t) andQE2 = 0 (26)If E(t) > EX thenQE1 = kE1 EX and (27)QE2 = kE2 [E(t) − EX]DP = kE3 E (t) (28)where kE1, kE2, and kE3 are discharge parameters controllingdeep/shallow aquifers and deep percolation, respectively.This module calculates the water level E(t) as state vari-
able using two input variables (g2 and g1) and four param-eters (EX, kE1, kE2 and kE3), to provide three outputs (QE1,QE2 and DP).3.1.3 Total runoff, baseflow and streamflowThe components of surface runoff, non-saturated runoff andbaseflow are added to obtain the total runoffQT [LT−1]:QT = QB + QC + QD + QE (29)As explained in Moussa and Chahinian (2009) the stream-flow Q [LT−1] at the outlet of the basin is obtained by therouting of QT using a transfer function (to take into accountthe water travel time). The Hayami (1951) kernel function(an approximation of the diffusive wave equation) is devel-oped to obtain a unit hydrograph linear model for this pur-pose. That is:
Q(t) =
t∫
0QT(τ ) H (t − τ) d τ (30)
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
378 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
H(t) is the Hayami kernel function, equal to:
H (t) =(w zF
π
)12 ezF (2 − t
w− w
t )
t3/2 (31)
and∞
∫
0H (t) dt = 1
where w [T] is a time parameter that represents the centreof gravity of the unit hydrograph (or the travel time) and zF[dimensionless] a form parameter. Q in [LT−1] units can betransformed to volume units [L3T−1] multiplying it by thebasin area [L2].3.2 Model parameterization, calibration and validationSummarizing, this Hydro-SVAT model uses four input vari-ables: rainfall R, Penman-Monteith ET0, Penman PET andleaf area index LAI to generate five main output variables:interception RIn, infiltration i, evapotranspiration ET, dis-charge componentsQ=QB+QC+QD+QE (from surface,non-saturated and aquifer reservoirs, respectively) and deeppercolation DP.Five state variables are calculated for every time step, the
water levels in the five reservoirs: A(t), B(t), C(t),D(t) andE(t). A coupled discharge QCD for the two non-saturatedreservoirs is obtained by addingQC andQD.In our experimental basin we applied the model for a one-
year period (2009), a time step 1t = 30min (1800 s) and abasin area equal to 0.886 km2.This model contains 20 parameters that are used to cal-
culate infiltration (CX, DX, CF, DF, fc and α), evapotran-spiration (REWc and rm), the exchange between reservoirs(AX, BX, kB, kC, kD, β, EX, kE1, kE2 and kE3) and the basintransfer function (w and zF).Four out of these twenty parameters (CX, DX, CF, DF)
were estimated using field data. For instance, two excava-tion experiments down to 3.5m showed that very few rootswere present below 1.5m, where the andisol layer turns intoa more clayey, compact and stony deposit. Then, the depthof the non-saturated root soil layer was fixed at CH = 1.6m(for simplicity, equal to the length of our FDR probe tubes).The depth of the non-root layer was estimated inDH = 1.0m.Following the relationships CX = (θs−θr) CH and DX =(θs−θr) DH, the levels for maximum water holding capac-ities in the non-saturated reservoirs can be calculated, aswell as the levels for field capacities, using the equationsCF = (θf−θr) CH and DF = (θf−θr) DH. The volumet-ric soil water contents θ were estimated as θr = 0.37: theresidual water content equal to the minimum θ observedin the basin during the study period; θs = 0.63: the θ atsaturation for a typical andisol, according to Hodnett andTomasella (2002); and θf = 0.43: the θ at field capacityequal to the average van Genuchten value for a matric po-tential =−10 kPa (Hodnett and Tomasella, 2002).
Three parameters were taken from literature reviews andexpert criteria (AX, REWc and rm). We set the surface reser-voir maximum storage capacityAX for our coffee AF systemequal to 4× 10−4m using data from Siles et al. .(2010b). Thetwo parameters of the evapotranspiration routine were takenas: REWc= 0.4 from Granier et al. (1999) and rm= 0.8 fromfield measurements of T ET−10 (data not shown).One parameter (BX) was obtained at the end of the op-
timization process in order to fit the maximum observedstreamflow peak (this parameter is very sensitive as it actsdirectly on the highest peaks). Two other parameters (w andzF) were separately estimated by a trial and error procedure,given the low sensitivity of the model to their variation.The remaining 10 empirical parameters (fc, α, kB, kC, kD,
β, EX, kE1, kE2 and kE3) were simultaneously optimized. Forthis purpose we used the Nelder-Mead (Nelder and Mead,1965) simplex algorithm included in Matlab (following La-garias et al., 1998) on 17 520 semi-hourly time steps (oneyear). Convergence was reached within 1000 runs and thestabilization of all parameter values by the end of the itera-tion process was checked. A two-step calibration procedurewas applied: (a) selection of an initial value for each param-eter, falling within the respective range (fourth column, Ta-ble 2), and (b) simultaneous estimation of parameter valuesthat maximize an objective function (sixth column of Table 2,identified as M1), in this case the Nash and Sutcliffe (1970)efficiency coefficient.We calibrated the model using one year of streamflow Q,
and then validated it using the remaining three measuredvariables: evapotranspiration ETR, water content in the non-saturated zone θ and water table level z. For ETR wegrouped the measured and modelled values (given in thesame units) at the daily time scale, which is the originaltime scale in Granier’s model, and then we excluded thegap-filled values. To calculate the modelled θ from thewater level in the root reservoir C(t) we used the relationθ =C(t) C−1X (θs−θr)+ θr, while the observed values wereobtained as the average of the 20 FDR point-measurementsthroughout the basin. To validate z we proposed an effectiveporosity of the aquifer nA= 0.39, to be able to directly linkpiezometric measurements z with the modelled water levelin our aquifer reservoir E(t), according to z=E(t)/nA.We also carried out an independent validation ofQ by run-
ning the calibrated Hydro-SVATmodel over 8 months of datafrom 2010 and verifying the model performance.In addition, we performed an analysis of model residuals:
zero expectancy, normality, homoscedasticity and standard-ized residuals.3.3 Uncertainty and sensitivity analysisIn order to investigate the uncertainty in model predictionswe performed a Monte-Carlo approach on a restricted subsetof parameter combinations, as suggested by Helton (1999).For each of the 10 parameters a range was created with
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 379
Table 2. Parameter description, range for optimization, optimized value and range for sensitivity analysis. M1: original 10-parameter model,M2: simplified 7-parameter model.
Parameter Description Units Range for Reference Optimum Optimumoptimization valueM1 valueM2
β Vertical/lateral split coefficient to divide fraction [0–0.84] Moussa and Chahinian 0.032 0.031outputs from non-saturated reservoirs into (2009)non-saturated runoff and vertical flows
kB Discharge rate for surface reservoir s−1 [0–1]× 10−4 Empirical parameter 2.12× 10−5 2.11× 10−5fc Infiltration rate at field capacity m s−1 [0–1]× 10−5 Minimum steady state infiltrab. in the 7.45× 10−6 –
experim. basin (Kinoshita,personal communication, 2009)
α Coefficient to calculate the maximum dimensionless [1–70] Moussa and Chahinian 102 –infiltration capacity from field capacity (2009)
kC Discharge coefficient, total outputs from s−1 [0–1]× 10−4 Empirical parameter 1.02× 10−4 1.06× 10−4non-saturated root reservoir
kD Discharge coefficient for total outputs from s−1 [0–1]× 10−4 Empirical parameter 6.65× 10−5 –non-saturated non-root reservoir
EX Threshold level in the aquifer reservoir, above m [0–1] Empirical parameter 0.341 0.339which a shallow-aquifer outlet is found
kE2 Discharge coefficient for baseflow from s−1 [0–2.4]× 10−6 Charlier et al. (2008) 1.02× 10−6 9.81× 10−7shallow aquifer reservoir
kE1 Discharge coefficient for baseflow from deep s−1 [0–2.1]× 10−6 Charlier et al. (2008) 1.58× 10−7 1.50× 10−7aquifer reservoir
kE3 Discharge coefficient for deep percolation s−1 [0–1]× 10−7 Empirical parameter 4.36× 10−8 4.29× 10−8from the aquifer reservoir
a deviation of ±30% around the optimum value found inthe calibration process (like in White et al., 2000; Ines andDroogers, 2002; Lenhart et al., 2002; Zaehle et al., 2005;Droogers et al., 2008), following the “one-at-a-time” methoddescribed by Hamby (1994, 1995) and Frey and Patil (2002).Then, we assumed a uniform distribution for all the parame-ters, and used the Latin Hypercube function of Simlab 2.2(http://simlab.jrc.ec.europa.eu) in order to produce a sam-ple from a joint probability distribution, of size equal to tentimes the number of parameters as recommended by Sim-Lab developers, i.e., 100 parameter combinations. Thesecombinations were introduced into the calibrated Hydro-SVAT model and we retrieved 100 streamflow output seriesover the 17 520 semi-hourly time steps. Then we generated17 520 empirical confidence intervals (95% and 99%) fromthe respective frequency distributions over the 100 parametercombinations.A sensitivity analysis (SA) was carried out to determine
which of the input variables contributed significantly tothis modelling uncertainty. Two separate assessments wereproduced. First, a summary (through the time) index ofmodel performance was studied: the Nash-Sutcliffe coef-ficient (NS), being evaluated by four sensitivity indexes:Pearson, Spearman, standardized regression and standard-ized rank regression coefficients (Hamby, 1994, 1995). Then,a second approach of sensitivity analysis was tested to try tofollow the behaviour of the Spearman coefficient for each of
the 10 parameters included in the SA through all time steps,as in Helton (1999).The sensitivity indexes are calculated departing from the
joint probability sample matrix xij of size m× n, where m
is the sample size and n the number of independent vari-ables (here our 10 parameters) to study. The Monte Carloevaluation of xij in the model produces the result vector yi ,configuring the matrix system [yi :xij ]. Then, the Pearsonproduct moment correlation (PEAR) for a given parameter jis the linear correlation coefficient between the variables xij
and yi over the m samples. To account for non-linear re-lationships that can be hidden by indicators like PEAR, asimple rank transformation is applied, replacing the originalx−y series with their corresponding ranks R(x) and R(y).Then, the Spearman coefficient (SPEA) is obtained by cal-culating the correlation on the transformed data, as SPEA(x,y) = PEAR[R(x), R(y)]. The standardized regression coeffi-cients (SRC) are the result of a linear regression analysis per-formed on [yi :xij ] but previously standardizing all the vari-ables. This is useful to evaluate the effect of the independentvariables on the dependent one, without regarding their unitsof measurement, and can be computed by standard statisticalmethods. Finally, the standardized rank regression coeffi-cients are obtained as SRRC(x, y) = SRC[R(x), R(y)].
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
380 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
4 ResultsIn the 0.9 km2 micro-basin of Mejıas creek, within the Cafe-talera Aquiares AF coffee farm, we obtained one full year(2009) of comprehensive experimental results of streamflow,evapotranspiration, soil water content and piezometry thatwe used to calibrate and validate the Hydro-SVAT model.First we present the hydrological behaviour of the basin, thenthe ecophysiological behaviour, and finally the water bal-ance.4.1 Hydrological behaviour of the basinThe time series of streamflow Q and rainfall R are given ata semi-hourly time-step in Fig. 5a for 2009. Rainfall (Fig. 2)was quite evenly distributed (no marked dry spell), althoughthe period from January to June clearly received less rain(later named the “drier season”, in opposition to the “wet-ter season”). From the total R (3208mm), the measured Q
at the outlet was 2048mm, yielding an annual streamflow co-efficient of 0.64. TheQ hydrograph (Fig. 5a) displays a con-tinuous baseflow with episodes of groundwater recharge af-ter rainfall events, followed by marked recessions controlledby the baseflow. Q peaks reached an annual maximum of0.84m3 s−1, i.e. 28% of the nominal capacity of the flume,indicating that the size chosen for the flume was adequate. Itcan be observed that similar rainfall events resulted in higherQ peaks during the wetter season. The lower Q peaks in re-sponse to rainfall events during the drier season could thusbe interpreted as the consequence of higher infiltration rateswhen belowground was less saturated. Soil water contentin the 0–1.6m layer (Fig. 5d) remained above 37% all-yearround and rose up to a maximum of 47% during the transitionbetween the drier and the wetter season. In order to under-stand the large baseflow shaping the hydrograph of Q, fourpiezometers were installed in early June 2009 (Fig. 5e) andshowed the existence of a permanent aquifer at levels vary-ing between 0.7 and 3.2m deep, according to their respectivedistance to water channels. Their behaviour was extremelyvariable, as expected, from responsive (piezos #1 and #3) toconservative behaviours (piezos #2 and #4). The continuousbaseflow observed by the flume originated mainly from animportant aquifer, covering a large (although undefined) areawithin the basin.4.2 Ecophysiological behaviour of the basinThe ecosystem LAI (Fig. 5b) changed seasonally quiteseverely from 2.8 (in March), as a result of coffee pruningduring the drier season and leaf shedding by E. poeppigiana,coffee flowering and new leafing just after the beginning ofthe wetter season, to a maximum of 4.8 (in September), thencoffee leaf shedding during the main coffee-berry harvest (inOctober).
The actual evapotranspiration (ETR) obtained by eddy-covariance (Fig. 5c) accounted for the sum of coffee andshade-tree transpiration, understory evaporation (mainly baresoil), and rainfall interception loss. The total ETR in 2009amounted to 818mm (25% of R), fluctuating daily accord-ing to atmospheric demand and seasonally according to LAIand canopy conductance. It always stayed below 4.5mmd−1and, in average, around 60% of reference ET0, clearly invali-dating the use of ET0 as a reliable indicator of ETR in coffeesystem eco-hydrological models. The crop coefficient (theETR ET−10 ratio) was clearly lower during the drier season,as a consequence mainly of a lower LAI. We did not observea period during which the relative extractable water (REW)of the soil drop below the critical value (REWc) of 0.4, con-firming that the coffee plants probably encountered no sea-sonal water stress.4.3 Water balance partitioning and closureFigure 6 shows the water balance partition and closure for2009, as obtained by the water flows measured by indepen-dent experimental methods, rainfall (R), streamflow (Q), andevapotranspiration (ETR) (including the rainfall interceptionloss). It was observed on cumulative values that the sum ofQ+ETR was 11% lower than annual R. However,Q+ETRmatched or exceeded R once at the end of April, just afterthree episodes of lower rainfall, which confirmed the occur-rence of an important storage in the basin, namely the soiland aquifer reservoirs, creating a seasonal hysteresis betweenR and Q. On an annual basis, measured Q represented 64%of R and measured ETR amounted 25%. The remaining11% was attributed to deep percolation, measurement errorsand/or inter-annual changes in soil and aquifer water stocks.Figure 7a shows the result of the streamflow modelling
for the year 2009, after calibrating the 10 parameters fromTable 2. The model yielded a NS coefficient of 0.89 andR2 = 0.88 for N = 17 520 semi-hourly time-steps. Baseflowand peakflow events appeared to be satisfactorily repre-sented for the whole time series. The modelled partition-ing of streamflow into surface runoff, non-saturated runoffand baseflow is presented in Fig. 7b. It indicated a promi-nent contribution of baseflow, as already inferred from vi-sual inspection of theQ time series. Hillslope surface runoffand non saturated runoff were a minor part of Q. The wa-ter stored in the 4 lowest reservoirs of the model is shownin Fig. 7c to f. The water level at the surface reservoirwas rarely greater than zero, displaying some few eventsof surface storage beyond the modelling time step (30min).The non-saturated reservoirs were replenished during stormevents and depleted by non-saturated runoff and evapotran-spitation (this latter acting only in the root reservoir). Finally,the aquifer reservoir displayed the largest magnitude of varia-tion, fluctuating seasonally by a factor of almost 4 (this factoris reduced to 1.3 when the aquifer effective porosity is takeninto account).
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 381
Fig. 5. Measurements in the experimental basin for 2009: (a) rainfall R and streamflow Q, 1t = 30min, (b) leaf area index (LAI) within1 std. dev. confidence bands (grey), (c) reference (ET0) and measured (ETR) daily evapotranspiration, (d) soil water content θ with onestandard deviation bars and (e) water table level z in the four piezometers throughout the basin. Graphs on the right side correspond to atwo-month period (June–July 2009) for better illustration.
5 DiscussionIn the next sections we first discuss the validation, uncer-tainty and sensitivity analysis of the model, second somemodel simplification attempts, third the main hydrologicalprocesses that we observed, and finally we examine the hy-drological services in our coffee AF basin.
5.1 Validation, uncertainty and sensitivity analysis forthe Hydro-SVAT model
5.1.1 Model validationModel validation was done by direct comparison of modeloutput variables with three field measurements: actual
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
382 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
Fig. 6. Water balance in the experimental basin for 2009. Onlymeasured values are presented here.
evapotranspiration ETR, soil water content θ and aquifer wa-ter level z. For ETR, the determination coefficient betweenthe daily sum of observed and modelled ETR was R2 = 0.79(Fig. 8a). Concerning θ , the observed and modelled time se-ries appeared to be consistent, but divergent during the driestseason (Fig. 8b), reaching a rather low R2 = 0.35. However,considering that this global θ was obtained as the arithmeticaverage of only 20 observations for the whole basin, the un-certainty of these measured values is high. The large amountof rocks hindering the tubes might also affect the FDR read-ings. Finally, we obtained a good approximation for the be-haviour of the aquifer (Fig. 8c), with R2 = 0.84. The useof the two piezometers that displayed the highest stability(probably representing the larger and more relevant aquifersystems) out of the four piezometers installed was crucial atthis step. We considered to be recording two different pro-cesses, the typical gradual aquifer response (piezos #2 and #4in Fig. 1c) and the local quick-varying shallow-water accu-mulations (piezos #1 and #3 in Fig. 1c), which might be as-sociated with rapid changes in soil water contents at smallerand less representative aquifer units. Though the represen-tativeness of these few piezometers of the behaviour of themain basin-aquifer could be questioned, the very high simi-larity that we found between the modelled values and the av-erage measurements from piezometers #2 and #4 (located intwo opposite ends of the basin), supported the idea of a cor-rect performance of both, the field method for aquifer moni-toring and the corresponding model routine that is based ona linear reservoir.Concerning the precision of our model, the NS= 0.89 on
streamflow seems to be good enough for a semi-hourly timestep, considering the detail with which the hydrological pro-cesses need to be described. Some studies that have eval-uated hydrological models using this coefficient for differ-ent time scales have shown the decline in NS values as themodelling time step is shortened. For instance, at calibra-tion stages, Notter et al. (2007) achieved maximum NS val-ues from 0.8 to 0.69 for decadal to daily time steps (basinarea = 87 km2), while Bormann (2006) obtained 0.8, 0.9,
0.85 and 0.73 for annual, monthly, weekly and daily timesteps, respectively (basin area = 63 km2). Garcıa et al. (2008)reached NS coefficients of 0.93, 0.91 and 0.61 for quarterly,monthly and daily modelling in a 162 km2 basin.In addition, the streamflow validation over an independent
eight-month test period (January–August 2010) produced aNS= 0.75, which is considered satisfactory (Fig. 8d). Duringthis first eight months of 2010 the total rainfall amounted to1978mm (compared to 2222mm in the first eight monthsof 2009), the total streamflow summed 1057mm (against1445mm in 2009) and 11 storm events exceeded the thresh-old of 0.3m3 s−1 (compared to 13 events in 2009).To study possible systematic errors in the Hydro-SVAT
model, we examined the distribution of residuals betweenmeasured and modelled streamflows (since this was the op-timized variable). A t-test with a confidence level of 95%indicated that we cannot conclude that the mean of our resid-uals (equal to −1× 10−4m3 s−1) is significantly differentfrom zero (p= 0.36). At our short time step it is common tofind highly autocorrelated residuals, so we found significantpartial autocorrelations up to the seventeen time lag (8.5 h).The Kolmogorov-Smirnov test revealed that the distributionof residuals is not normal, though it is highly symmetricalaround zero. Studying residuals as functions of time, rain-fall R, streamflow Q, soil water content θ and water tablelevel z (data not shown), we did not find any trends, but no-ticeable changes in their variability make clear that the ho-moscedasticity condition was not properly fulfilled. How-ever, it is accepted that these ordinary least squares assump-tions are often not satisfied in streamflow modelling (Xu andSingh, 1998).5.1.2 Uncertainty analysisIf we accept that the model reproduces efficiently the actualstreamflow at the outlet of the experimental basin, the uncer-tainty in the modelled values needs to be known. A Monte-Carlo (MC) uncertainty analysis was produced as detailed inSect. 3.3 to yield the empirical 95% and 99% confidence lim-its (CL) (the 95% CL are presented in Fig. 9, along with mea-sured streamflow). 82% of the measured values fell withinthe 95% CL produced by our model, while 87% of them fellwithin the 99% CL. The ranges of the 95% confidence in-tervals (CI) along each of the time steps varied from 20%to 131% of the MC mean value, while for the 99% CI theranges were from 24% up to 157% of the MC mean. Fromthese analyses it is possible to state that the model is efficient(high NS coefficient), precise (relatively small confidence in-tervals, Fig. 9) and accurate (given the high percentages ofmeasured Q values falling within the 95% and 99% confi-dence intervals).
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 383
Fig. 7. Model results: (a)measured vs. modelled streamflow and (b) components of modelled streamflow. Water level in the model reservoirs:(c) surface reservoir, (d) non-saturated root reservoir (offset by θrCH to show the absolute water content in the soil layer), (e) non-saturatednon-root reservoir (offset by θrDH) and (f) aquifer reservoir. Graphs on the right side correspond to a two-month period (June–July 2009)for better illustration.
5.1.3 Sensitivity analysis
The sensitivity analysis was carried out to determine the re-sponsiveness of the model predictions to variations in ourmain parameters. The first assessment consisted in testingthe statistical significance of the relationships between eachof our 10 calibration parameters and the NS coefficient, us-ing four dimensionless sensitivity indexes: Pearson (PEAR),Spearman (SPEA), standardized regression and standardized
rank regression coefficients (SRC and SRRC, respectively).Table 3 presents the results of these tests revealing that, withthe exception of the discharge coefficients from the root non-saturated and shallow-aquifer reservoirs (kC and kE2), andof the shallow-aquifer threshold (EX), none of the parame-ters seemed to significantly influence the global model ef-ficiency. However, these results may be misleading giventhe enormously variable conditions under which the modelworks through time. Therefore, in a second approach we
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
384 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
Fig. 8. Validation of: (a) daily evapotranspiration ETRm, (b) soil water content θ at the 1.6m root-reservoir, (c) water table level z and(d) streamflowQ for eight months of 2010. (R2: determination coefficient, RRMSE: relative root mean squared error).
selected the Spearman test (a simple rank transformation in-dex that can identify non linear relationships) to assess the in-fluence of each of these 10 parameters on streamflow, at eachtime step. Figure 10 presents the results, displaying the di-mensionless Spearman index in the vertical axis and the timein the horizontal axis. A positive Spearman index reveals aproportional influence of the parameter on the model result,
while negative values indicate inverse proportionality. Fig-ure 10a shows the time series for all the parameters, reveal-ing an alternation in their influence on streamflow over timeand model state. The two black horizontal lines represent the95% confidence limits above (or below) which the correla-tion is significantly different from zero. In Fig. 10b to k thesignificant values are plotted as black points in contrast to
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 385
Table 3. Sensitivity indexes for each of the 10 calibration parame-ters vs. the NS coefficient. PEAR: Pearson product moment corre-lation coefficient, SPEA: Spearman coefficient, SRC: Standardizedregression coefficient and SRRC: Standardized rank regression co-efficients. Index values in bold are statistically significant at 95%confidence level.
Sensitivity indexParameter PEAR SPEA SRC SRRCβ 0.08 0.07 0.09 0.08kB −0.04 −0.04 −0.06 −0.05fc −0.06 −0.10 −0.05 −0.07α −0.13 −0.11 −0.04 −0.06kC −0.20 −0.24 −0.13 −0.16kD −0.07 −0.11 −0.08 −0.10EX 0.18 0.05 0.22 0.11kE2 −0.29 −0.29 −0.27 −0.24kE1 0.10 0.09 0.06 0.05kE3 0.02 −0.06 −0.01 −0.06
non-significant in grey, for each parameter separately. Fig-ure 10h suggests that the parameter EX is again one of themost influential, because it is permanently displaying highpositive or negative correlations, together with the dischargecoefficient (DC) for the deep-aquifer kE1, which most of thetime has a proportional influence on model outputs (Fig. 10j).These two parameters reach the highest Spearman indexesduring sustained periods (not only during storm events) andare clearly relevant during long streamflow recessions, whenthe modelled water table level z is around or below EX. TheDC for the surface reservoir kB (Fig. 10c) is noticeably in-fluencing model outputs during each individual storm event,while the vertical/lateral split coefficient β (Fig. 10b) is rel-evant only during the main ones. The DC for the deep per-colation kE3 (Fig. 10k) is significant over recession periods,while DC for root reservoir kC is significant when its watercontent exceeds field capacity (Fig. 10f). The DC for theshallow aquifer kE2 (Fig. 10i) has proportional influence onstreamflow when z is above EX, or else, inverse influenceduring the driest part of the recessions. Finally, the infiltra-tion rate at field capacity (fc), the coefficient for maximuminfiltration rate (α) and the DC for non-root reservoir (kD)seem to have no relevant effects on the streamflow modelling(Fig. 10d, e and g, respectively).5.2 Model simplificationA model simplification was carried out by assuming that thethree less sensitive parameters could be considered as mod-elling constants. The original 10-parameter model (identifiedasM1) was converted to a 7-parameter model (calledM2) inthe search for model parsimony and faster convergence. Theparameters fc, α and kD were fixed according to the optimal
Fig. 9. Measured streamflow against empirical 95% confidence in-terval.
values from the original model. Hence, we assumed thatfc = 7.45× 10−6ms−1, α = 102 and kD = 6.65× 10−5 s−1.This simplified model M2 converged after 1524 runs andreached a NS= 0.88, a performance that is almost as goodas that of M1. The optimal values for M2 are presented inthe seventh column of Table 2, and are very close to the op-timal forM1. Given the algorithmic configuration of Nelder-Mead procedure, the convergence of M2 to the same para-metric optimal than M1 is not assured (since the new modelstructure defines a completely different simplex problem, ona smaller 7-dimensional space). From this result, an almostidentical water balance is obtained fromM2. In any case,M2should be validated under different conditions (other basinsand spatio-temporal scales), in order to corroborate that thenumerical values that we assigned to the three fixed param-eters can be exported to different contexts, and can alwaysbe treated like modelling constants. The simplification hereundertaken also proves the effectiveness of the time-varyingsensitivity analysis here proposed, as a tool to identify themain parametric sources of modelling uncertainty and, there-fore, the potentially redundant model parameters.An additional simplification step was carried out by sepa-
rately removing two lateral flowpaths that could be unnec-essary in the model structure (QB2 and QE2), given thatreservoirs B (surface) and E (aquifer) already have primaryflowpaths (QB1 and QE1). However, while suppressing QB2affects the simulation of the main peakflow (that decreasesfrom 0.84m3 s−1 to 0.52m3 s−1), the exclusion ofQE2 enor-mously worsens the model’s ability to shape the hydrographrecessions (and therefore the aquifer recharge/discharge pro-cesses), yielding a NS= 0.71 for the optimized model. Fromthis exercise we conclude that these two flowpaths (and theirrespective parameters) play a clearly identifiable and relevantrole in the model structure.
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
386 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
Fig. 10. Spearman indexes between Q and: (a) all the parameters: black horizontal lines indicate the 95% confidence interval out ofwhich the Spearman is significant, (b) partition parameter β, c) surface discharge kB, (d) infiltration rate at field capacity fc, (e) maximuminfiltration α, (f) root-reservoir discharge kC, (g) non-root discharge kD, (h) threshold for shallow aquifer EX, (i) shallow aquifer dischargekE2, (j) deep aquifer discharge kE1 and (k) deep percolation discharge kE3. The black dots indicate the time steps for which the Spearman issignificant at the 95% confidence level, while grey dots indicate non-significant correlations.
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 387
5.3 Hydrological processes in the experimental basinThe main hydrological processes and components observedusing this measuring/modelling approach are presented in thefollowing four sub-sections.5.3.1 Interception, throughfall, stemflow and surface
runoffA review of water balance (WB) partitioning in comparablesituations is proposed in Table 4 (interception loss, evapo-transpiration, surface/non-saturated runoff, baseflow, changein soil water content and deep percolation). In our basin,the adjusted interception loss (RIn) equalled 4% of inputrainfall (R). This value is the same as found by Imbachet al. (1989) in a WB experiment under similar coffee andE. poeppigiana land cover, but is lower than other reportsobtained by direct measurements at plot scale in Costa Rica:Jimenez (1986) found 16% under the same AF system; Har-mand et al. (2007) found 15% for coffee and Eucalyptusdeglupta (but this RIn could be lower because stemflow wasnot separately measured); Siles (2007) found 11–15% undercoffee and Inga densiflora. Interception loss can be greatlyaffected by the local LAI of both layers, the specific archi-tecture of the coffee and trees and the rainfall regime.From the throughfall/stemflow component in our model
(equal to 96% of R), 4% of R came out of the basin assurface runoff QB. This value is not far from other re-ports at plot scale, like those by Avila et al. (2004): 1–9%(coffee and E. deglupta); Harmand et al. (2007): 2% andSiles (2007): 3–6% (coffee and I. densiflora). At basin scale,Fujieda et al. (1997) measured 5% on a 0.56 km2 basin un-der a three layer forest in Brazil, Lesack (1993) modelled3% in a rain-forest basin in Brazil (area: 0.23 km2), Kinneret al. (2004) modelled 4% in a Panamanian tropical-forestbasin (0.10 km2) and Charlier et al. (2008) modelled 10%in a banana-plantation basin in Guadeloupe (0.18 km2). Amajor source of surface runoff and discrepancy between plotand basin scale studies could be the presence of roads. Inour experimental basin the total length of roads is 10 km andit represents 4.5% of the basin area. Then, assuming a di-rect road-runoff of 80% (Ziegler et al., 2004), this componentcould represent up to 95% of the total basin surface runoff.5.3.2 InfiltrationThe non-intercepted and non-runoff fraction of incident rain-fall was infiltrated (i = 92% of R). This large i/R ra-tio and a QB/i ratio close to 4% give indication of veryhigh infiltration and drainage capacities, which are typicalof andic-type volcanic soils (Poulenard et al., 2001; Cat-tan et al., 2006) and are further enhanced for perennialcrops in the absence of tillage and in presence of substan-tial macroporosity (Dorel et al., 2000). It was indicated bypreliminary measurements carried out in our experimental
basin with a Cornell infiltrometer (Ogden et al., 1997)that steady state infiltrability values could be as high as4.7× 10−5ms−1 (168mmh−1) (Kinoshita, personal com-munication, 2009). Recent experiments found hydraulic con-ductivity values as high as 3.4× 10−5ms−1 (122mmh−1)and 2.1× 10−5ms−1 (75mmh−1) for andisols in Costa Rica(Cannavo et al., 2010) and Guadeloupe (Charlier et al.,2008), respectively. From our infiltration capacity (that wasmodelled as a function of soil water content in the non-saturated reservoirs), we calculated the infiltration/rainfall(i/R) and the runoff/infiltration (QB/i) ratios at the storm-event scale, analyzing 78 events with cumulative rain higherthan 10mm. No significant changes were found in any ofthese two ratios as a function of time (or season), and onlyslight reductions were detected for increasing storm cumu-lative rainfall or soil water content. This fact, added to thepermanently high i/R ratios for each individual event (65–98%), are explained by the constantly high infiltration capac-ity of andisols and the stable behaviour of soil water contentthroughout the year. We therefore observed very efficient soiland aquifer recharge mechanisms (Fig. 7d to f). According toDorel et al. (2000) the soil properties of non-tilled perennialcrops in andisols are mainly determined by wetting-dryingcycles and by biological activity in the soil. Those are rela-tively stable factors in our experimental basin, given the ab-sence of a well-defined dry season on this side of the coun-try (Caribbean influence), and the permanently high organicmatter content in these soils.5.3.3 Evaporation and transpirationTranspiration of coffee plants and trees T accounts for20% of R (645mmy−1, obtained from modelling). Highertranspiration values were reported by van Kanten andVaast (2006) in a coffee AF system under E. poeppi-giana (29%, 897mmyr−1), and also by Siles (2007), whomeasured values ranging from 31% (1008mmyr−1) to34% (905mmyr−1) (see Table 4). In other experimentalplots in Costa Rica, estimations of T ranged from 42%(811mmyr−1) to 53% (750mmyr−1) (Imbach et al., 1989;Jimenez, 1986, respectively). T can be highly dependent onlocal ET0, on the effect of drought on stomatal closure, andon LAI. For a better site comparison, we computed a sim-ple “normalized transpiration” index NT= T (ET0 LAI)−1 inTable 4 and found that our value (0.16) is very close to therespective ratio in the study of Siles (2007).As mentioned earlier, we measured the actual evapotran-
spiration (ETR= T +Eu +RIn) in our experimental basin bythe eddy-covariance method and it represented 25% of R
(804mmyr−1). This is the smallest value reported in com-parison to the other studies on coffee AF systems (Table 4).The closest value (38%, 956mm in nine months) was mea-sured by Harmand et al. (2007), but it is about 1.5 times ourETR; while a maximum of 69% (985mmyr−1) was reportedby Jimenez (1986). It must be stressed that our study is the
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
388 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
Table 4. Comparison of annual water balance components (as % of rainfall) for different studies at plot and basin scales, in tropical regions.Source Location Climate Basin area Land cover LAI Rainfall RaIn T a Total ETa0 NTd QaB QaCD QaE Total Qa 1Sa Da DP,
(km2) (m2m−2) R (mm) ETRa SOa
Jimenez Costa Rica Humid plot scale Coffee and – 2642b 16 53 69 – – – – – – 9 22 –(1986) tropical E. poeppigianaImbach et al. Costa Rica Humid plot scale Coffee and – 1919 4 42 46 60 – – – – – – 54 –(1989) tropical E. poeppigianaHarmand et al. Costa Rica Humid plot scale Coffee and 3.5 2622b 15 23 38 – – 2–3 – – – 6 54 –(2007) tropical E. degluptaSiles (2007), Costa Rica Humid plot scale Coffee and 5.0–6.0 2684– 11–15 28–34 41–46 39–44 0.14 3–6 – – – −1–1 44–55 –Cannavo et al. tropical I. densiflora 3245(2010)Lesack Brazil Humid 0.23 Rain forest – 2870 – – 39 – – 3 – – 57 2 – 1(1993) tropicalFujieda et al. Brazil Humid 0.56 Three layer – 2319 15 15 30 32 – 5 6 59 70 – – –(1997) subtrop. forestKinner and Panama Humid 0.10 Tropical – 2400 – – 53 56 – 4 17 20 41 6c – 6cStallard (2004) tropical forestGenereux et Costa Rica Humid 0.26 Tropical rain - 4974 – – 32–46 – – – – – 54–68 – – –al. (2005) tropical forestCharlier et Guadeloupe Maritime 0.18 Banana – 4229 – – 31 31 – 10 – 17 27 0 59 42al. (2008) hum. trop.This study Costa Rica Humid 0.90 Coffee and 3.8 3208 4 20 25 32 0.16 4 4 56 64 0 69 11–12
tropical E. poeppigiana
* Numbers in bold are measured quantities.a Given as % of R: RIn: interception loss, T : transpiration, ETR: evapotranspiration, ET0: reference evapotranspiration, QB: surface runoff, QCD: non-saturated runoff (subsurfaceflow), QE: baseflow, Q: streamflow, D: drainage, 1S: change in soil water content, DP: deep percolation, SO: subsurface outflow.b Annual estimation for experiments conducted in short term periods (less than a year).c The authors ignore whether this fraction is being stored in soil, or it became subsurface outflow downstream the gauging site.d NT=normalized transpiration equal to T (ET0 LAI)−1
only one that brings independent validation of ETR throughenergy balance closure, whereas many plot studies mightcarry errors due to the calibration of sapflow, the model ofEu or the sampling of RIn. At the basin scale, many authorsmodelled ETR values around 30% and 46% in tropical ex-perimental basins. In absolute terms these values are: 1120,554–682, 1961–2345 and 1300mmyr−1 (Lesack, 1993; Fu-jieda et al., 1997; Genereux et al., 2005; Charlier et al., 2008,respectively).5.3.4 Drainage, deep percolation and streamflowAnother component of interest at plot scale is drainage (ver-tical flow beyond root reservoir), which we modelled as 69%of R and seems, on average, slightly higher than the valuesreported in the literature (Table 4). The modelled deep per-colation (or subsurface outflow downstream the basin outlet)was 12%, much lower than the 42% reported by Charlier etal. (2008), but similar to the 6% encountered by Kinner andStallard (2004). As deep percolation is calculated from waterbalance closure, its accuracy is enhanced when Q and ETRare precisely measured.The first (or the second) highest WB basin output is usu-
ally streamflow Q, which in our experimental basin wasrecorded and modelled as 64% of R. It seems to be veryclose to similar measurements in the tropics (Table 4). Thebaseflow from the aquifer accounted for 87% of total Q,
while surface runoff was only 6% and non-saturated runoff7%.5.4 The coffee agroforestry basin and hydrological
servicesModelling the hydrological behaviour of this experimental,coffee AF basin gave some insights on the provision of HESby this system. This is particularly relevant in the Costa Ri-can context where HES payments have already been imple-mented as national environmental protection policies (Pagi-ola, 2008). Two main services related to water quality canbe recognized, both linked to the observed high infiltration i
(92% of R) and low surface runoff QB (4% of R).At first, the low QB in the basin is closely associated to
low surface displacement of fertilizers, pesticides and sed-iments (Cattan et al., 2006, 2009; Leonard and Andrieux,1998; Bruijnzeel, 2004). We found very constant QB/i ra-tios through the time, or under different rainfall intensities,which may come from the expected stability in soil hydraulicproperties (e.g. high infiltration capacity) and the absenceof either a marked dry season that controls soil desiccation(Park and Cameron, 2008; Dorel et al., 2000) or mechanizedagricultural practices like tillage affecting soil compaction,surface roughness, continuity of pores, macroporosity, soilcover and organic matter content (Le Bissonnais et al., 2005;Chahinian et al., 2006). However, the high drainage capacity
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 389
of these andisols might be a disadvantage in terms of per-colation and groundwater contamination by agrochemicals(Cattan et al., 2007; Saison et al., 2008) given our model es-timates for groundwater recharge of around 67% of R. Inaddition, the modelled surface runoff for the experimentalbasin includes possible discharges from unpaved roads andditches, which needs to be controlled to avoid excessive wa-ter, sediment and contaminant flux concentrations.A second HES might be the streamflow regulation
function provided by this AF basin through aquiferrecharge/discharge mechanisms. With a measured evapo-transpiration close to 25% of R (which is presumably muchlower than the equivalent for forests), soil and aquifer waterdepletion seems unlikely under the observed hydrogeologicaland climatic conditions, favouring water availability duringdry seasons (Robinson et al., 2003; Bruijnzeel, 2004). Onthe other hand, during intense rainfalls and tropical stormsthe aquifer is efficiently recharged, as we have observed inour piezometric measurements. The result is a homogeneousseasonal distribution of streamflow, with a high rainfall recu-peration fraction of 64% (Q/R).
6 ConclusionsThis paper gives some insights into the assessment of Hydro-logical Environmental Services (HES) by studying the hy-drological processes in a particular micro-basin. The waterbalance partition is proposed as a baseline for analyses andnegotiations leading to the payment for HES, for which mea-suring and modelling approaches are complementary. Theunderstanding of water dynamics supplied a better knowl-edge about the main services provided by the studied ecosys-tem, as well as some potential vulnerabilities.The general behaviour of the coffee AF basin (1 km2) on
andisols can be summarized by the fact that 92% ofR was in-filtrated through the highly permeable andisol, 64% of R wasmeasured as streamflow, 25% of R was measured as evapo-transpiration, no major seasonal variation was detected in thesoil water stock, and a large aquifer contribution to stream-flow was observed in the shape of baseflow (Fig. 7b). Theseare characteristics of a system prone to generate importantHES at basin scale, which is a result infrequently reportedfor coffee systems.We proposed an original modelling approach coupling a
hydrological and a SVAT model, calibrated using the stream-flow at the outlet of the basin, but validated by independentand direct measurements of streamflow (test period), evapo-transpiration, soil water content and water table level.We presented a standard uncertainty analysis in order to
built simulation confidence intervals around our modelledstreamflow values, as well as a sensitivity analysis to inves-tigate the source of such uncertainty. The first parameter-ization of the model is considered adequate, though modelsimplification could be attempted centred on the three less
sensitive parameters. Special attention needs to be given todirect measurement of a representative field capacity and theassociated probability distribution function.The conceptual nature of our Hydro-SVAT model allows a
wide time/space domain of application, conditional only onknowledge of some general properties of the basin of interestand on the acquisition of basic hydrological data. Differ-ent environments can be configured in terms of climate, landcover, soils and hydrogeology, and further applications un-der different conditions are desired to test the generality ofthe model. Complementary studies like hillslope and chan-nel surface runoff, basin water losses through roads, temporalvariation in soil and ecophysiological properties and groundwater dynamics and composition are expected.
Acknowledgements. This work was funded by the Centre deCooperation Internationale en Recherche Agronomique pourle Developpement (CIRAD, France), the European projectCAFNET (EuropAid/121998/C/G) and the Cafetalera Aquiaresfarm (http://www.cafeaquiares.com). CATIE (Centro AgronomicoTropical de Investigacion y Ensenanza) provided collaborationand infrastructure. We gratefully thank the Cafetalera Aquiaresfarm for allowing us to install the experimental basin and forproviding us research facilities. Rintaro Kinoshita contributed tosaturated infiltrability observations and Alvaro Barquero and hisfamily backed the field work. Discussions with Lars Gottschalk,Irina Krasovskaia, Sadı Laporte and Rafael Chacon provided usefulinsight. We would also like to thank the two anonymous reviewersfor their constructive and effective comments. Federico Gomez-Delgado was supported by a CIRAD fellowship during his PhD.Edited by: H. Cloke
ReferencesAllen, R. G., Pereira, L. S., Raes, D., and Smith, M.: Crop evap-
otranspiration: Guidelines for computing crop water require-ments, Food and Agriculture Organization, Rome, 301 pp., 1998.
Ataroff, M. and Monasterio, M.: Soil erosion under different man-agement of coffee plantations in the Venezuelan Andes, SoilTechnol., 11, 95–108, 1997.
Avila, H., Harmand, J. M., Dambrine, E., Jimenez, F., Beer, J.,and Oliver, R.: Dinamica del nitrogeno en el sistema agroforestalCoffea arabica con Eucalyptus deglupta en la Zona Sur de CostaRica, Agroforesterıa en las Americas, 41–42, 83–91, 2004.
Baldocchi, D. and Meyers, T.: On using eco-physiological, mi-crometeorological and biogeochemical theory to evaluate carbondioxide, water vapor and trace gas fluxes over vegetation: a per-spective, Agr. Forest Meteorol., 90, 1–25, 1998.
Bergstrom, S.: The HBV model, in: Computer Models of Water-shed Hydrology, edited by: Singh, V. P., Water Resources Publi-cations, Colorado, USA, 1995.
Bormann, H.: Impact of spatial data resolution on simulatedcatchment water balances and model performance of the multi-scale TOPLATS model, Hydrol. Earth Syst. Sci., 10, 165–179,doi:10.5194/hess-10-165-2006, 2006.
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
390 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
Bruijnzeel, L. A.: Hydrological functions of tropical forests: notseeing the soil for the trees?, Agr. Ecosyst. Environ., 104, 185–228, 2004.
Cannavo, P., Sansoulet, J., Harmand, J. M., Siles, P., Dreyer, E.,and Vaast, P.: Agroforestry associating coffee and Inga densi-flora results in complementarity for water uptake and decreasesdeep drainage in Costa Rica, Agr. Ecosyst. Environ., in press,doi:10.1016/j.agee.2010.11.005, 2010.
Cattan, P., Cabidoche, Y. M., Lacas, J. G., and Voltz, M.: Effectsof tillage and mulching on runoff under banana (Musa spp.) on atropical Andosol, Soil Till. Res., 86, 38–51, 2006.
Cattan, P., Voltz, M., Cabidoche, Y. M., Lacas, J. G., and Sansoulet,J.: Spatial and temporal variations in percolation fluxes in a trop-ical Andosol influenced by banana cropping patterns, J. Hydrol.,335, 157–169, 2007.
Cattan, P., Ruy, S. M., Cabidoche, Y. M., Findeling, A., Desbois,P., and Charlier, J. B.: Effect on runoff of rainfall redistribu-tion by the impluvium-shaped canopy of banana cultivated on anAndosol with a high infiltration rate, J. Hydrol., 368, 251–261,2009.
Ceballos, A. and Schnabel, S.: Hydrological behaviour of asmall catchment in the dehesa landuse system (Extremadura,SW Spain), J. Hydrol., 210, 146–160, 1998.
Chahinian, N., Moussa, R., Andrieux, P., and Voltz, M.: Compar-ison of infiltration models to simulate flood events at the fieldscale, J. Hydrol., 306, 191–214, 2005.
Chahinian, N., Moussa, R., Andrieux, P., and Voltz, M.: Accountingfor temporal variation in soil hydrological properties when sim-ulating surface runoff on tilled plots, J. Hydrol., 326, 135–152,2006.
Charlier, J. B., Cattan, P., Moussa, R., and Voltz, M.: Hydrologicalbehaviour and modelling of a volcanic tropical cultivated catch-ment, Hydrol. Process., 22, 4355–4370, 2008.
Cormary, Y. and Guilbot, A.: Relations pluie-debit sur le bassin dela Sioule, Rapport D. G. R. S. T. No. 30, Universite des Scienceset Techniques, Montpellier, France, 1969.
Crawford, N. H. and Linsley, R. K.: Digital simulation in hydrol-ogy: Stanford watershed model IV, Technical Report 39, Depart-ment of Civil Engineering, Stanford University, California, USA,210, 1966.
Dauzat, J., Rapidel, B., and Berger, A.: Simulation of leaf tran-spiration and sap flow in virtual plants: model description andapplication to a coffee plantation in Costa Rica, Agr. Forest Me-teorol., 109, 143–160, 2001.
Dawes, W. R., Zhang, L., Hatton, T. J., Reece, P. H., Beale, G. T.H., and Packer, I.: Evaluation of a distributed parameter ecohy-drological model (TOPOG IRM) on a small cropping rotationcatchment, J. Hydrol., 191, 64–86, 1997.
Diskin, M. H. and Nazimov, N.: Linear reservoir with feedbackregulated inlet as a model for the infiltration process, J. Hydrol.,172, 313–330, 1995.
Donigan, A., Bicknell, B., and Imhoff, J. C.: Hydrological simula-tion program - Fortran (HSPF), in: Computer Models of Water-shed Hydrology, edited by: Singh, V. P., Water Resources Publi-cations, Colorado, USA, 1995.
Dorel, M., Roger-Estrade, J., Manichon, H., and Delvaux, B.:Porosity and soil water properties of Caribbean volcanic ashsoils, Soil Use Manage., 16, 133–140, 2000.
Droogers, P., Van Loon, A., and Immerzeel, W. W.: Quantifying theimpact of model inaccuracy in climate change impact assessmentstudies using an agro-hydrological model, Hydrol. Earth Syst.Sci., 12, 669–678, doi:10.5194/hess-12-669-2008, 2008.
Duan, Q., Sorooshian, S., and Gupta, V.: Effective and efficientglobal optimization for conceptual rainfall-runoff models, WaterResour. Res., 28, 1015–1031, doi:10.1029/91WR02985, 1992.
Falge, E., Baldocchi, D., Olson, R., Anthoni, P., Aubinet, M., Bern-hofer, C., Burba, G., Ceulemans, R., Clement, R., Dolman, H.,Granier, A., Gross, P., Grunwald, T., Hollinger, D., Jensen, N.-O., Katul, G., Keronen, P., Kowalski, A., Ta Lai, C., Law, B. E.,Meyers, T., Moncrieff, J., Moors, E., William Munger, J., Pile-gaard, K., Rannik, U., Rebmann, C., Suyker, A., Tenhunen, J.,Tu, K., Verma, S., Vesala, T., Wilson, K., and Wofsy, S.: Gapfilling strategies for long term energy flux data sets, Agr. ForestMeteorol., 107, 71–77, 2001.
Fleming, G.: Computer Simulation Techniques in Hydrology, En-vironmental Science Series, Elsevier, New York, USA, 1975.
Frey, H. C. and Patil, S. R.: Identification and Review of SensitivityAnalysis Methods, Risk Anal., 22, 553–578, 2002.
Fujieda, M., Kudoh, T., de Cicco, V., and de Calvarcho, J. L.: Hy-drological processes at two subtropical forest catchments: theSerra do Mar, Sao Paulo, Brazil, J. Hydrol., 196, 26–46, 1997.
Garcıa, A., Sainz, A., Revilla, J. A., Alvarez, C., Juanes, J. A.,and Puente, A.: Surface water resources assessment in scarcelygauged basins in the north of Spain, J. Hydrol., 356, 312–326,2008.
Genereux, D. P., Jordan, M. T., and Carbonell, D.: A paired-watershed budget study to quantify interbasin groundwater flowin a lowland rain forest, Costa Rica, Water Resour. Res., 41,W04011, doi:dx.doi.org/10.1029/2004WR003635, 2005.
Granier, A., Breda, N., Biron, P., and Villette, S.: A lumped wa-ter balance model to evaluate duration and intensity of droughtconstraints in forest stands, Ecol. Model., 116, 269–283, 1999.
Gutierrez, M. V., Meinzer, F. C., and Grantz, D. A.: Regulation oftranspiration in coffee hedgerows: covariation of environmentalvariables and apparent responses of stomata to wind and humid-ity, Plant Cell Environ., 17, 1305–1313, 1994.
Hamby, D. M.: A review of techniques for parameter sensitivityanalysis of environmental models, Environ. Monit. Assess., 32,135–154, 1994.
Hamby, D. M.: A Comparison of Sensitivity Analysis Techniques,Health Phys., 68, 195–204, 1995.
Harmand, J.-M., Avila, H., Dambrine, E., Skiba, U., de Miguel, S.,Renderos, R., Oliver, R., Jimenez, F., and Beer, J.: Nitrogen dy-namics and soil nitrate retention in a Coffea arabica – Eucalyptusdeglupta agroforestry system in Southern Costa Rica, Biogeo-chemistry, 85, 125–139, 2007.
Havnø, K., Madsen, M. N., and Dørge, J.: MIKE 11 - a generalizedriver modelling package, in: Computer Models ofWatershed Hy-drology, First edition, edited by: Singh, V. P., Water ResourcesPublications, Colorado, USA, 1995.
Hayami, S.: On the propagation of flood waves, Disaster Prev. Res.Inst. Bull., 1, 1–16, 1951.
Helton, J. C.: Uncertainty and sensitivity analysis in performanceassessment for the Waste Isolation Pilot Plant, Comput. Phys.Commun., 117, 156–180, 1999.
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/
F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica 391
Hodnett, M. G. and Tomasella, J.: Marked differences betweenvan Genuchten soil water-retention parameters for temperate andtropical soils: a new water-retention pedo-transfer functions de-veloped for tropical soils, Geoderma, 108, 155–180, 2002.
ICE: Plan de Expansion de la Generacion Electrica, Perıodo 2010-2021, Instituto Costarricense de Electricidad, San Jose, CostaRica, 2009.
Imbach, A. C., Fassbender, H. W., Beer, J., Borel, R., and Bonne-mann, A.: Sistemas agroforestales de cafe (Coffea arabica) conlaurel (Cordia alliodora) y cafe con poro (Erythrina poeppigiana)en Turrialba, Costa Rica, VI. Balances hıdricos e ingreso conlluvias y lixiviacion de elementos nutritivos, Turrialba, 39, 400–414, 1989.
Ines, A. V. M. and Droogers, P.: Inverse modelling in estimatingsoil hydraulic functions: a Genetic Algorithm approach, Hydrol.Earth Syst. Sci., 6, 49–66, doi:10.5194/hess-6-49-2002, 2002.
Jimenez, F.: Balance hıdrico con enfasis en percolacion de dos sis-temas agroforestales: cafe-poro y cafe-laurel, en Turrialba, CostaRica, MSc Thesis, CATIE, Turrialba, Costa Rica, 104 pp., 1986.
Kinner, D. A. and Stallard, R. F.: Identifying storm flow pathwaysin a rainforest catchment using hydrological and geochemicalmodelling, Hydrol. Process., 18, 2851–2875, 2004.
Lagarias, J. C., Reeds, J. A., Wright, M. H., and Wright, P. E.:Convergence Properties of the Nelder-Mead Simplex Method inLow Dimensions, SIAM J. Optimiz., 9, 112–147, 1998.
Le Bissonnais, Y., Cerdan, O., Lecomte, V., Benkhadra, H.,Souchere, V., and Martin, P.: Variability of soil surface char-acteristics influencing runoff and interrill erosion, Catena, 62,111–124, 2005.
Lenhart, T., Eckhardt, K., Fohrer, N., and Frede, H. G.: Comparisonof two different approaches of sensitivity analysis, Phys. Chem.Earth, 27, 645–654, 2002.
Leonard, J. and Andrieux, P.: Infiltration characteristics of soils inMediterranean vineyards in Southern France, Catena, 32, 209–223, 1998.
Lesack, L. F. W.: Water Balance and Hydrologic Characteristics ofa Rain Forest Catchment in the Central Amazon Basin, WaterResour. Res., 29, 759–773, doi:10.1029/92WR02371 1993.
Maeda, K., Tanaka, T., Park, H., and Hattori, S.: Spatial distributionof soil structure in a suburban forest catchment and its effect onspatio-temporal soil moisture and runoff fluctuations, J. Hydrol.,321, 232–256, 2006.
Marsden, C., le Maire, G., Stape, J.-L., Seen, D. L., Roupsard, O.,Cabral, O., Epron, D., Lima, A. M. N., and Nouvellon, Y.: Re-lating MODIS vegetation index time-series with structure, lightabsorption and stem production of fast-growing Eucalyptus plan-tations, Forest Ecol. Manag., 259, 1741–1753, 2010.
MEA: Ecosystems and HumanWell-Being: Synthesis., MillenniumEcosystem Assessment, Island Press, Washington, USA, 155,2005.
MINAE-RECOPE: Mapa geologico de Costa Rica, Scale1:200 000, San Jose, Costa Rica, 1991.
Mora-Chinchilla, R.: Geomorfologa de la cuenca del Rıo Turrialba,unpublished, Universidad de Costa Rica, San Jose, Costa Rica,2000.
Moussa, R. and Chahinian, N.: Comparison of different multi-objective calibration criteria using a conceptual rainfall-runoffmodel of flood events, Hydrol. Earth Syst. Sci., 13, 519–535,doi:10.5194/hess-13-519-2009, 2009.
Moussa, R., Chahinian, N., and Bocquillon, C.: Distributed hy-drological modelling of a Mediterranean mountainous catchment– Model construction and multi-site validation, J. Hydrol., 337,35–51, 2007a.
Moussa, R., Chahinian, N., and Bocquillon, C.: Erratum to “Dis-tributed hydrological modelling of a Mediterranean mountain-ous catchment - Model construction and multi-site validation”[J. Hydrol. 337 (2007) 35–51], J. Hydrol., 345, 254–254, 2007b.
Nash, J. E. and Sutcliffe, J. V.: River flow forecasting through con-ceptual models part I – A discussion of principles, J. Hydrol., 10,282–290, 1970.
Nelder, J. A. and Mead, R.: A Simplex Method for Function Min-imization, Comput. J., 7, 308–313, doi:10.1093/comjnl/7.4.308,1965.
Nilson, T.: A theoritical analysis of the frequency of gaps in plantstands, Agr. Meteorol., 8, 25–38, 1971.
Notter, B., MacMillan, L., Viviroli, D., Weingartner, R., and Lin-iger, H.-P.: Impacts of environmental change on water resourcesin the Mt. Kenya region, J. Hydrol., 343, 266–278, 2007.
Ogden, C. B., van Es, H. M., and Schindelbeck, R.R.: Miniature Rain Simulator for Field Measurement ofSoil Infiltration, Soil Sci. Soc. Am. J., 61, 1041–1043,doi:10.2136/sssaj1997.03615995006100040008x, 1997.
Pagiola, S.: Payments for environmental services in Costa Rica,Ecol. Econ., 65, 712–724, 2008.
Park, A. and Cameron, J. L.: The influence of canopy traits onthroughfall and stemflow in five tropical trees growing in a Pana-manian plantation, Forest Ecol. Manag., 255, 1915–1925, 2008.
Peel, M. C., Finlayson, B. L., and McMahon, T. A.: Updatedworld map of the Koppen-Geiger climate classification, Hy-drol. Earth Syst. Sci., 11, 1633–1644, doi:10.5194/hess-11-1633-2007, 2007.
Poulenard, J., Podwojewski, P., Janeau, J.-L., and Collinet, J.:Runoff and soil erosion under rainfall simulation of Andisolsfrom the Ecuadorian Paramo: effect of tillage and burning,Catena, 45, 185–207, 2001.
Roberts, G. and Harding, R. J.: The use of simple process-basedmodels in the estimate of water balances for mixed land usecatchments in East Africa, J. Hydrol., 180, 251–266, 1996.
Robinson, M., Cognard-Plancq, A. L., Cosandey, C., David, J., Du-rand, P., Fuhrer, H. W., Hall, R., Hendriques, M. O., Marc, V.,McCarthy, R., McDonnell, M., Martin, C., Nisbet, T., O’Dea, P.,Rodgers, M., and Zollner, A.: Studies of the impact of forests onpeak flows and baseflows: a European perspective, Forest Ecol.Manag., 186, 85–97, 2003.
Roupsard, O., Bonnefond, J.-M., Irvine, M., Berbigier, P., Nouvel-lon, Y., Dauzat, J., Taga, S., Hamel, O., Jourdan, C., Saint-Andre,L., Mialet-Serra, I., Labouisse, J.-P., Epron, D., Joffre, R., Bra-connier, S., Rouziere, A., Navarro, M., and Bouillet, J.-P.: Parti-tioning energy and evapo-transpiration above and below a tropi-cal palm canopy, Agr. Forest Meteorol., 139, 252–268, 2006.
Roupsard, O., Dauzat, J., Nouvellon, Y., Deveau, A., Feintrenie,L., Saint-Andre, L., Mialet-Serra, I., Braconnier, S., Bonnefond,J.-M., Berbigier, P., Epron, D., Jourdan, C., Navarro, M., andBouillet, J.-P.: Cross-validating Sun-shade and 3D models oflight absorption by a tree-crop canopy, Agr. Forest Meteorol.,148, 549–564, 2008.
www.hydrol-earth-syst-sci.net/15/369/2011/ Hydrol. Earth Syst. Sci., 15, 369–392, 2011
392 F. Gomez-Delgado et al.: Modelling hydrological behaviour of coffee agroforestry basin in Costa Rica
Rouse, J. W., Haas, R. H., Schell, J. A., and Deering, D. W.: Mon-itoring vegetation systems in the Great Plains with ERTS, Proc.Third Earth Resources Tech. Satellite-1 Symp. Vol. 1: TechnicalPresentations, Sec. A, NASA SP-351, Washington, USA, 1974.
Ruiz, L., Varma, M. R. R., Kumar, M. S. M., Sekhar, M., Marechal,J.-C., Descloitres, M., Riotte, J., Kumar, S., Kumar, C., andBraun, J.-J.: Water balance modelling in a tropical watershedunder deciduous forest (Mule Hole, India): Regolith matric stor-age buffers the groundwater recharge process, J. Hydrol., 380,460–472, 2010.
Saison, C., Cattan, P., Louchart, X., and Voltz, M.: Effect of Spa-tial Heterogeneities of Water Fluxes and Application Pattern onCadusafos Fate on Banana-Cultivated Andosols, J. Agr. FoodChem., 56, 11947–11955, 2008.
SEPSA: Boletın Estadıstico Agropecuario 19, Secretarıa Ejecutivade Planificacion Sectorial Agropecuaria, San Jose, Costa Rica,2009.
Shuttleworth, W. J. and Wallace, J. S.: Evaporation from sparsecrops-an energy combination theory, Q. J. Roy. Meteor. Soc.,111, 839–855, 1985.
Siles, P.: Hydrological processes (water use and balance) in a coffee(Coffea arabica L.) monoculture and a coffee plantation shadedby Inga densiflora in Costa Rica, PhD Thesis, Universite HenriPoincare, Nancy-I, Nancy, France, 153 pp., 2007.
Siles, P., Harmand, J.-M., and Vaast, P.: Effects of Inga densi-flora on the microclimate of coffee Coffea arabica L.) and overallbiomass under optimal growing conditions in Costa Rica, Agro-forest. Syst., 78, 269–286, 2010a.
Siles, P., Vaast, P., Dreyer, E., and Harmand, J.-M.: Rainfall par-titioning into throughfall, stemflow and interception loss in acoffee (Coffea arabica L.) monoculture compared to an agro-forestry system with Inga densiflora, J. Hydrol., 395, 39–48,doi:10.1016/j.jhydrol.2010.10.005, 2010b.
Singh, V. P.: Computer Models of Watershed Hydrology, First edi-tion, Water Resources Publications, Colorado, USA, 1995.
USDA: Soil Taxonomy, A Basic System of Soil Classification forMaking and Interpreting Soil Surveys, Agriculture HandbookNumber 436, Washington, USA, 1999.
Vaast, P., Beer, J., Harvey, C., and Harmand, J. M.: Environmen-tal services of coffee agroforestry systems in Central America:a promising potential to improve the livelihoods of coffee farm-ers communities, in: Intregrated Management of EnvironmentalServices in Human-Dominated Tropical Landscapes, IV, editedby: Wallace, H. A., Inter-American Scientific Conference Series,CATIE, Turrialba, Costa Rica, 2005.
van Kanten, R. and Vaast, P.: Transpiration of Arabica Coffeeand Associated Shade Tree Species in Sub-optimal, Low-altitudeConditions of Costa Rica, Agroforest. Syst., 67, 187–202, 2006.
Vega, F. E. and Rosenquist, E.: The coffee berry borer and coffeeresearch at the United States Department of Agriculture, Proc.First ICO World Coffee Conference, London, UK, 2001.
Wahl, T. L., Clemmens, A. J., Replogle, J. A., and Bos, M. G.: Win-Flume –Windows-based software for the design of long-throatedmeasuring flumes, Fourth Decennial National Irrigation Sympo-sium, Phoenix, Arizona, 2000.
White, M. A., Thornton, P. E., Running, S. W., and Ne-mani, R. R.: Parameterization and Sensitivity Analysis of theBIOME-BGC Terrestrial Ecosystem Model: Net Primary Pro-duction Controls, Earth Interact., 4, 1–85, doi:10.1175/1087-3562(2000)004<0003:PASAOT>2.0.CO;2, 2000.
Wilson, K. B., Hanson, P. J., Mulholland, P. J., Baldocchi, D. D.,and Wullschleger, S. D.: A comparison of methods for deter-mining forest evapotranspiration and its components: sap-flow,soil water budget, eddy covariance and catchment water balance,Agr. Forest Meteorol., 106, 153–168, 2001.
Xu, C. Y. and Singh, V. P.: A Review on Monthly Water Bal-ance Models for Water Resources Investigations, Water Resour.Manag., 12, 31–50, 1998.
Zaehle, S., Sitch, S., Smith, B., and Hatterman, F.: Ef-fects of parameter uncertainties on the modeling of terrestrialbiosphere dynamics, Global Biogeochem. Cy., 19, GB3020,doi:dx.doi.org/10.1029/2004GB002395, 2005.
Ziegler, A. D., Giambelluca, T. W., Sutherland, R. A., Nullet, M.A., Yarnasarn, S., Pinthong, J., Preechapanya, P., and Jaiaree, S.:Toward understanding the cumulative impacts of roads in uplandagricultural watersheds of northern Thailand, Agr. Ecosyst. Env-iron., 104, 145–158, 2004.
Hydrol. Earth Syst. Sci., 15, 369–392, 2011 www.hydrol-earth-syst-sci.net/15/369/2011/