world distribution of land cover changes during pre- and

World distribution of land cover changes during Pre- and Protohistoric Times and estimation of induced

carbon releases

Répartition mondiale des espaces défrichés au cours de la Pré-et Protohistoire et estimation des rejets de carbone induits

Carsten Lemmen*

* GKSS Forschungszentrum Geesthacht GmbH, Institut für Küstenforschung, Max-Planck Straße 1, 21501 Geesthacht, Germany ([email protected]).

AbstractThe role of Pre- and Protohistoric anthropogenic land cover changes needs to be quantified i) to establish a baseline for comparisonwith current human impact on the environment and ii) to separate it from naturally occurring changes in our environment. Results arepresented from the simple, adaptation-driven, spatially explicit Global Land Use and technological Evolution Simulator (GLUES) forpre-Bronze age demographic, technological and economic change. Using scaling parameters from the History Database of the GlobalEnvironment as well as GLUES-simulated population density and subsistence style, the land requirement for growing crops is esti-mated. The intrusion of cropland into potentially forested areas is translated into carbon loss due to deforestation with the dynamicglobal vegetation model VECODE. The land demand in important Prehistoric growth areas – converted from mostly forested areas –led to large-scale regional (country size) deforestation of up to 11% of the potential forest. In total, 29 Gt carbon were lost from glob-al forests between 10000 BC and 2000 BC and were replaced by crops; this value is consistent with other estimates of Prehistoricdeforestation. The generation of realistic (agri-)cultural development trajectories at a regional resolution is a major strength of GLUES.Most of the pre-Bronze age deforestation is simulated in a broad farming belt from Central Europe via India to China. Regional carbonloss is, e.g., 5 Gt in Europe and the Mediterranean, 6 Gt on the Indian subcontinent, 18 Gt in East and Southeast Asia, or 2.3 Gt in sub-saharan Africa.

Key words: land use, land cover change, deforestation, carbon emission, early anthropogenic greenhouse effect.

RésuméLe rôle de l’Homme dans l’évolution des états de surface de la Terre au cours de la Pré- et Protohistoire doit être estimé afin de ré-pondre à deux objectifs : 1) établir une base de données qui permette de comparer les différentes actions anthropiques sur l’environ-nement et 2) bien distinguer les causes « naturelles » des causes d’origine anthropique dans l’évolution des dynamiques environne-mentales. L’utilisation d’un logiciel intitulé « Global Land Use and technological Evolution Simulator (GLUES) » a permis de mo-déliser les changements démographiques, technologiques et économiques antérieurs à l’Âge du Bronze. La démarche a consisté à es-timer le besoin en terres pour répondre à la croissance continue de mises en culture en se basant sur les données de la « History Da-tabase of the Global Environment » et sur la modélisation des densités de population et des types de subsistance estimées d’aprèsGLUES. Les défrichements forestiers se sont traduits par une perte de carbone estimée grâce au modèle VECODE. Le besoin en terresdans les foyers d’expansion préhistoriques - développés principalement dans des secteurs forestiers - a conduit à des déforestationsmassives, représentant jusqu’à 11 % de l’ensemble des forêts potentielles. Sur la période 10 000-2000 av. J.-C., ce sont au total envi-ron 29 Gt de carbone qui ont été perdues en raison de la substitution des champs cultivés aux forêts. GLUES a permis la visualisa-tion d’axes d’expansion agricole à une échelle régionale. La plupart des défrichements observés avant l’Age du Bronze ont affecté unvaste territoire qui s’étend depuis l’Europe centrale jusqu’à la Chine, en passant par l’Inde. Les valeurs régionales de perte en carbo-ne sont estimées à 5 Gt en Europe et en Méditerranée, à 6 Gt en Inde, à 18 Gt en Asie du sud et du sud-est et à 2,3 Gt pour l’Afriquesubsaharienne.

Mots clés : utilisation des sols, changements de couverture végétale, déforestation, émission de carbone, effet de serre anthropique précoce.

Version française abrégée

Le début de l’Anthropocène est placé par certains auteursautour de 1750 (Crutzen et Stoermer, 2000). L’apparition decette période marque le passage d’un environnement préin-dustriel à un environnement industrialisé, caractérisé par desinterventions majeures d’origine anthropique sur les écosys-tèmes. Pour W. Ruddiman (2003), l’Anthropocène se situebien avant le milieu du XVIIIe siècle et débute il y a 8 000 ansquand la diffusion des pratiques agricoles néolithiques se gé-néralise. En effet, après que les forêts aient connu leur exten-sion maximale au début de la dernière période postglaciaire,les pratiques agropastorales du Néolithique ont eu pourconséquence de grandes phases de défrichement afin de dis-poser de surfaces cultivables. Au total, ce sont près de 107 km2

de forêts qui ont été détruits au bénéfice de l’agriculture et dupastoralisme (Williams, 2000). Ces défrichements ont induitune émission de carbone de l’ordre de 100 Gt, soit l’équiva-lent de 15 % des émissions actuelles par la végétation de laplanète. Cette étude a pour but de montrer comment il est pos-sible de cartographier les impacts spatio-temporels de la dé-forestation au cours de la Pré- et Protohistoire.

La modélisation par le logiciel GLUES indique que lespratiques agropastorales de subsistance ont commencé vers8 000 av. J.-C. au Proche-Orient et en Chine. Vers 6 000 av.J.-C., on observe une diffusion importante de l’agriculturedepuis les foyers pionniers vers l’Europe du sud-est et à tra-vers la Chine et le Japon, tandis qu’un nouveau foyer agri-cole fait son apparition dans la partie amont de la vallée del’Indus. Vers 4 000 av. J.-C., certaines régions d’Afrique(Maghreb, Afrique de l’ouest et du sud-est) se convertissentégalement à l’agriculture. Les espaces cultivés se générali-sent durant cette période en s’étendant depuis l’Espagnejusqu’au Japon. Vers 2000 av. J.-C., les techniques agri-coles se développent à partir du centre et du sud-est de l’Eu-rope vers l’Est, l’Inde, l’Indochine, la Chine et le Japon. Acette période, l’Europe du sud-est, l’est de l’Inde, le Viet-nam, le Japon et la Chine présentent les plus fortes densitésde population du globe (3-6 hab/km2). La précision du mo-dèle d’expansion des activités agricoles a été testée en éta-lonnant la durée de la phase de transition à l’aide de data-tions par le radiocarbone réalisées sur des sites néoli-thiques en Eurasie (partie occidentale ; Turney et Brown,2007). Jusqu’à 11 % des terres de la planète sont cultivéesvers 2000 av. J.-C.

L’extension des cultures se superpose de manière satisfai-sante à la carte des densités de population des régions oùles techniques agricoles sont particulièrement avancées. Lesrégions où les cultures sont le plus développées concernentles Balkans, le Proche-Orient, le sud et le nord de la Chineainsi que le Japon. La plupart des régions ayant connu undéveloppement agricole précoce étaient auparavant recou-vertes de forêts ; ces mêmes espaces ont donc subi une dé-forestation massive du fait de la mise en culture. Les pertesestimées les plus importantes en carbone concernent leJapon, la Grèce septentrionale et le sud de la Chine, avecune valeur proche de 36 t/ha. A l’échelle de la planète, laperte en carbone est estimée à 11 Gt vers 7000 av. J.-C. et à

29 Gt vers 2000 av. J.-C. L’importante mutation des pay-sages de la Chine, du sud et du sud-est de l’Asie et de la par-tie occidentale de l’Eurasie a contribué à des pertes totalesde carbone estimées à 90 % ; pour leur part, l’Afrique sub-saharienne et les Amériques n’ont participé seulement qu’àhauteur de 10 % à la perte totale de carbone depuis la pé-riode préhistorique (8000 av. J.-C.).

Introduction

Since when have humans influenced the global carboncycle? And what is the magnitude of anthropogenic perturba-tion in the carbon cycle compared to the natural variability?Commonly, the beginning of the anthropocene is placed at1750 (Crutzen and Stoermer, 2000), a date which is used toseparate the preindustrial from the industrial (anthropogeni-cally perturbed) environment. W. Ruddiman (2003) arguedfor a much earlier start of human interference by carbondioxide (CO2) and methane (CH4) emissions and placed thestart of the anthropocene at 8000 years ago; at this time, thefirst widespread agricultural areas appeared in China, the Me-diterranean, and central Europe (e.g., Childe, 1936). The ad-vent of Neolithic life style was associated with huge cultural,technological and social changes. Foremost the transitionfrom hunting-gathering subsistence to agro-pastoral life stylehad a great impact on our Earth system, or as M. Zeder (2008)stated, ‘domesticates and the agricultural economies basedon them are associated with radical restructuring of humansocieties, worldwide alterations in biodiversity, and signifi-cant changes in the Earth’s land forms and its atmosphere’.Only after the termination of the last glacial, a forest extentwas established which we can use as the baseline for anthro-pogenic perturbation during the Holocene (11600 years ago tothe present); this forest was by no means pristine but had al-ready been shaped by intentional intermittent clearing andburning by Palaeolithic and Mesolithic people (e.g., Sim-mons, 1996). Neolithic agropastoral subsistence for the firsttime required long-term removal of forest to create space forsettlements, crops, or animals; additionally, timber was har-vested for construction and as fuel. For a model NeolithicEuropean village of 30 people, S. Gregg (1988) estimatedthat an area of 6 km2 forested area was necessary for survi-val. In total, about 107 km2 of forest and woodland havebeen lost due to agriculture and pastoralism up to date (e.g.,Williams, 2000). A quick calculation assuming a value foraboveground biomass of 100 t/ha – derived for deciduousbroad-leaved temperate trees by P. Curtis et al. (2002) – andneglecting the possible loss of litter and soil carbon, esti-mates a global carbon loss on the order of 100 Gt. This re-duction amounts to 15% of today’s carbon pool in live ve-getation (650 Gt, Field and Raupach, 2004), although allthese estimates have to be treated with care due to the highuncertainties of estimating the carbon content of the tropicalforest areas (e.g., Houghton, 2005).

The sequence of the transition to agriculture is best visiblein the agricultural records of Europe and the Mediterranean.In forested Europe, Neolithic agriculture had been introdu-ced from the Levant region (today’s Lebanon, Israel, Syria,

304 Géomorphologie : relief, processus, environnement, 2009, n° 4, p. 303-312

Palestine, Jordan and Iraq) by an expansion through the Me-diterranean basin, and by land advance through Anatolia andthe Balkan; this succession is clearly visible in crop ances-tor species and animal bones (e.g., Flannery, 1973; Zeder,2008). The most conclusive evidence for tracing this transi-tion can be found in South East and Central Europe, wherethe appearance of the Körös and Starçevo cultures between6500-5500 BC and the linear pottery culture between 5650-4900 BC mark the onset of intensive horticulture and the in-troduction of cereal agriculture (Bogaard, 2004). To fill spa-tial and temporal gaps in the data obtained from excavationsand to provide a framework for hypothesis testing and pro-cess understanding, model studies have been employed, ran-ging from simple reaction-diffusion models (e.g., Fort et al.,2004) to models with transport of a farming trait (e.g., Ack-land et al., 2007), more complex socio-technological develop-ment in multiple traits (Wirtz and Lemmen, 2003), up to multi-agent simulations (e.g., Axtell et al., 2002). Agent based socio-technological models have, however, only been useful on alocal scale due to the very high computational demand whensimulating single individuals or households. This study in-vestigates regional socio-technological development on aglobal scale and for the entire period of pre-bronze age tran-sition from foraging to farming subsistence. It connectsadaptations in the socio-cultural system to the bioclimaticbackground and interpolates in space and time regional de-mographics, economics, and environmental impact, exem-plified here as deforestation. Calculates are proposed fromthe areal demand for crop land in previously forested areasfor total carbon emission. More importantly than obtainingan exact value of total carbon emission, this study showshow to attribute the emission by Prehistoric deforestation tospecific location and time.

Models and data

Socio-technological model GLUES

The numerical model GLUES (Global Land Use and Tech-nological Evolution Simulator; Wirtz and Lemmen, 2003;https://sourceforge.net/projects/glues/) simulates the transitionfrom hunting-gathering to agropastoral societies. It has beenused to investigate the role of biogeographic proposition andclimatic susceptibility (ibid.), and to assess the role of migra-tion waves in the spread of agriculture (Lemmen and Wirtz,2009). The socio-technological model comprises population,growth, and few characteristic traits whose temporal evolutionare governed by marginal benefits to population growth (adap-tive dynamics, e.g., Wirtz and Eckhardt, 1996; Dieckmann andLaw, 1996). Three dimensionless traits were used to describei) the efficiency gains in production, storing, and use of thefood base (=technology efficiency), ii) division of labour infood production between agropastoral and hunting-gatheringactivities (=farmer share), and iii) economic diversity realizedin different domesticated crops and animals in each regionalpopulation. The value of all traits determines demographicchange; in turn, optimization to optimal growth governsadaptive trait changes. This way, and in accordance with S.

Shennan (2001), the population dynamics is directly cou-pled to cultural evolution. The socio-technological model isembedded in a spatially explicit geographic context. Eachlocal society is tied to a region, characterized by geographicposition, vegetation, and climate fluctuations. These are basedon the IIASA climatology of R. Leemans and W.P. Cramer(1991), the climate variability data base of K. Wirtz et al.(2009), and dynamic vegetation calculations with the Ve-getation Continuous Description model (VECODE; Brov-kin et al., 1997, 2002), assuming Prehistoric carbon dioxidelevels of 280 ppm. To find an optimal parameter set, all pa-rameters were randomly varied within confined ranges. Allvariation sets were scored with the temporal and spatialproximity of simulated agricultural centres to the five globalindependent centres of agriculture (Smith, 1997): a parame-ter set was chosen to generate simulations resembling mostclosely the observed Prehistoric global agriculturalizationpattern. Following G. Ackland et al. (2007), the model pre-dictions should be interpreted as providing a ‘historical nullhypothesis. Its predictions can be taken as requiring no spe-cial explanation, and its failures can be taken as evidence ofrare events that had significant and long-lived conse-quences’.

In GLUES, a socio-technological system, or regionalculture, is defined by the four state variables populationdensity P, technological efficiency T, farmer share (orquota) Q, and economic diversity N. The characteristictraits X ∈ {T, N, Q} are subject to gradient adaptive dyna-mics which optimizes a population’s fitness with the evo-lution equation (dX/dt)=σX(dr/dX), where r=(1/P)·(dP/dt)is the specific growth rate, and σX the variability in trait X.The growth rate r depends on the traits X, on the natural re-sources R, and trait-dependent food procurement S; it iscomposed of growth and loss terms:

where µ, γ, ω and ρ are scaling factors. This growth rateformulation includes a term for instantaneous overexploi-tation (R-γ√TP) of resources based on Ehrlich’s IPAT for-mula, a loss of the work force to administration (1-ωT), anda decrease of density-dependent mortality with technology(-(ρ/T)P). Long-term degradation of natural resources wasnot considered. Subsistence S is derived from either huntingand gathering or from herding and farming. The relativeamount of time and energy spent on either foraging or cul-turing is expressed by the farmer share Q, which defines thetrade-off for the system:

where τ describes temperature limitation. 685 simulation re-gions where defined based on ecological homogeneity witha clustering algorithm by K. Wirtz (pers. comm. 2004) basedon net primary productivity derived from the IIASA databa-se (Leemans and Cramer, 1991). In each simulation region,a socio-technological model is started with identical initialvalues. Interaction of regions occurs via trade and migra-tion. The environmental context is set both by temperaturelimitation τ derived from the number of growing degree

305Géomorphologie : relief, processus, environnement, 2009, n° 4, p. 303-312

(1)

(2)

days above zero in the IIASA climatology (Leemans andCramer, 1991) and resources R, derived from the net prima-ry productivity, calculated with VECODE from the IIASAclimatology. The simulations where performed with thestandard parameters set in GLUES version 1.1.4pre, publi-cy available on sourceforge.net. Knowledge loss or clima-te fluctuations where not considered. Initial values areP0=0.03/km2, T0=1, Q0=0.04, N0=0.25, variability isσT=0.15, and σN=1. The parameter set for this model(equation 1) setup is γ=0.01 km2, ω=0.04, µ=0.0037/a,and ρ=0.00037/a. Conversion from forest to cropland inthe VECODE model (Brovkin et al., 1997) was implemen-ted by subtracting the required area for cropping Ac from theforest part of the land Af and recalculating the live carboncompartments (leafs and structure) assuming grass specificlive carbon content Clive,g for the crops:

Dead carbon (soil and litter) pools were reduced byreducing the specific carbon content Cdead,f by 42% (Guoand Gifford, 2002) in cropped areas, thus:

The mathematical background and an in-depth descriptionof the model are explained in K. Wirtz and C. Lemmen (2003).

Data and configuration for deforestationestimates

GLUES is supplemented withscaling parameters from the His-tory Database of the Global En-vironment (HYDE version 3.1;Klein Goldewijk et al., 2007)and additional information on

forest share and carbon pools (soil and live carbon) fromVECODE. The usage and interaction of the different modelsand databases in the calculation of deforestation are shownin fig. 1. In addition to the model set-up described by K.Wirtz and C. Lemmen (2003), which uses growing degreedays above zero degree (GDD0) and net primary production(NPP) from VECODE, I also use predicted potential forestshare as a base line from which deforestation can be calcu-lated. The mean forest share in all areas where agriculturedeveloped before 2000 BC was 70%, which is close to themaximum attainable value. Cropland and population densi-ties from the HYDE data base at 5’ resolution were regrid-ded on the 685 GLUES regions, the ratio of cropland per ca-pita was calculated and a representative value of 0.02 km2

per person for full-scale agricultural regions (farmershare>90%) was extracted; this value describes the intensi-vely used area in contrast to S. Gregg (1988)’s value of0.2 km2, which characterizes the total are in use (but not ne-cessarily converted) by Neolithic farmers. The cropland-per-person quantity was used to scale from the product offarmer ratio and population density to simulated croplandshare. No representative value for the pasture demand percapita could be derived from the HYDE database; areal de-mand for the forest-pasture conversion was thus neglectedin this study. In the conversion from forest to cropland, thelive carbon in trees is replaced with live carbon representa-tive by grassland (VECODE only distinguishes betweenfour types of land cover: broadleaved forest, needleleavedforest, grassland, and desert). To estimate the loss of soilcarbon, we follow L.B. Guo and R.M. Gifford (2002)’smeta-analysis, who find that reported field site soil carbon


Fig. 1 – Workflow for dynamichindcasting of deforestation basedon databases for cropland (historicaldatabase of the Environment, HYDE)and climate variables (IIASA), the dy-namic global vegetation model VE-CODE, and simulated populationdensity and agropastoral subsistencefraction from GLUES.

Fig. 1 – Cadre schématique généralde la modélisation de la déforesta-tion fondé sur l’utilisation combinéedes bases de données des terres cul-tivées (base de données historiquesde l’Environnement, HYDE) et desvariables climatiques (IIASA), la mo-délisation des dynamiques végé-tales est réalisée grâce à VECODE,les densités de population estiméeset les secteurs de subsistence agro-pastorales sont déterminés avecGLUES.

(3)

(4)

Clive = Clive,f (Af - Ac) + Clive,g (Ag + Ac) .

Cdead = Cdead,f (Af - 0.42Ac) + Cdead,gAg .

loss in the conversion from forest to crop land was on ave-rage 42%. For comparison to studies where only land de-mand and not carbon emission was published, we useP. Curtis et al. (2002)’s value of 100 t/ha carbon in above-ground vegetation derived from North American deciduousbroad-leaved temperate trees to convert from croplandshare. This is a reasonable assumption, when predominant-ly forested areas are used for crop land, and because it re-presents a mean value between 40 t/ha found in more aridsubtropical woodland to up to 180 t/ha found in wet (sub)tro-pical forests. For comparison with the archaeological record,we used the comprehensive data set on the neolithization ofEurope by C.S.M. Turney and H. Brown (2007), whichcontains the dating of 735 western Eurasian sites and their at-tribution to 82 distinct cultural groups and sub-groups.

Results

The global simulation of socio-technological traits revealsat 2000 BC a broad band of technologically advanced cultu-re ranging from central and southeast Europe over the Midd-le East, through India, Indochina, and China to Japan (fig. 2,left panel). High technology levels are also found in theMaghreb and intermediate technologies in Western Africaand in Southeast Africa. Not in all of these regions high po-pulation densities are found (fig. 2, right panel). At 2000 BC,the Southeast European region comprising Italy, the Balkan,Anatolia and the Levant is heavily populated with popula-tion densities of 4-6 inhab/km2. In eastern India and Viet-nam, population density is around 4 inhab/km2, and in Chinamostly 3 inhab/km2. Again heavily populated are the upper


Fig. 2 – Map of simulated technological efficiency (left panel) and population density (right panel) at 2000 BC. Technology ranges bet-ween 1 (the Mesolithic level) and 8 for near Bronze-age technology, population density attains values up to 6 inhab/km2, up from the initialuniform hunter-gatherer density of 0.1 inhab/km2.

Fig. 2 – Carte simulant la diffusion technologique (vignette de gauche) et la densité de population (vignette de droite) à 2 000 BC.Indices de technologie entre 1 (mésolithique) et 8 (Âge du Bronze), les densités de population atteignent des valeurs proches de 6 habitantsau km², alors que celles des communautés de chasseurs/cueilleurs ont des valeurs de 0,1 hab/km².

Fig. 3 – Time slices of the farmer share simulated with GLUES. Colour indicates the range from initially only foraging (light) with shortmixed-strategy transitions to mainly agriculture and herding (dark).

Fig.3 – Différentes phases d’activités fermières simulées avec GLUES. Les couleurs indiquent les différentes possibilités s’échelonnantdes activités fourragères exclusives (en bleu) à une agriculture/pastoralisme principalement (en clair), en passant par une courte périodede transition (en sombre).

Huang He (Yellow River) regionand Japan, where population den-sity is approximately 6 inhab/km2.

The temporal evolution of thetransition from foraging to farming is shown in six timeslices from 8000 BC to 2000 BC in fig. 3. At 8000 BC, nofarming is visible anywhere on Earth, however, inspectionof the data reveals that small-scale farming already begins inthe Levant and in China. By 7000 BC, agriculture is fullyestablished in these founder regions and also in centralJapan. The very fast transition is consistent with estimatesby, e.g., M. Zvelebil and P. Dolukhanov (1991) who notethat only few hundred years are required for a subsistencechange from hunting and gathering to cultivating and her-ding. By 6000 BC, agriculture has expanded from the foun-ding centres: to the north into southeast Europe from the Le-vant, and to the south from the northern Chinese centre; anew agricultural centre appears upstream of the Indus River.1000 years later, the Indian and Chinese crop areas join. At4000 BC, parts of Africa (the Maghreb, western Africa andsoutheast Africa) have converted to full-scale farming. Abroad spatially coherent band of farming reaching fromSpain to Japan is present.

The accuracy of the simulated expansion of farming is tes-ted by comparing the timing of the transition in the model toradiocarbon dates from western Eurasian Neolithic sitescompiled by C.S.M. Turney and H. Brown (2007). The ove-rall agreement of the simulated farming occurrence with thesite data is good; on a per-site basis, the mean absolute dif-ference between model and data is around 500 years (fig. 4).For western Europe, the model simulates the emergence ofagriculture between 7000 and 2500 BC, but the picture isregionally diverse. Neolithic lifestyle is simulated in nor-thern Greece before 7000 BC, and in the entire region of theBalkan and Minor Asia before 6600 BC. By 5700 BC, thewestern and eastern shores of the Black Sea, Italy, and Sou-thern Spain/Northern Morocco have acquired agriculture astheir dominant life style. Central Europe (Austria, Hungary,Slovakia, Southern Poland) and northern Italy show neoli-thization by 5000 BC; the Iberian Peninsula exhibits fullneolithization by 4300 BC; Neolithic life style reaches theBaltic Sea by 4000 BC as well as Belarus. The neolithiza-


Fig. 4 – Time of reaching the Neoli-thic stage (threshold 50% farmingratio, between 8000-4000 BC). Si-mulation results (background sha-ding) compared to the data set ofNeolithic sites by Turney and Brown(2007, shown as circles).

Fig. 4 – Délai pour atteindre laphase néolithique (seuil de 50 %pour les activités fermières, pério-de 8 000-4 000 BC) simulé à partirde GLUES (arrière plan), comparai-son avec la répartition des sites néo-lithiques (points), d’après Turney andBrown (2007).

Fig. 5 – Map of simulated crop fraction (left panel) and crop fraction interpolated to the simulation regions from the HYDE databa-se (right panel).

Fig. 5 – Carte stimulant la part des secteurs cultivés (vignette de gauche) et des secteurs cultivés interpolés à partir de la base dedonnées HYDE (vignette de droite).

tion in southern and northern Germany occurs not before2500 BC; the simulation fails to predict the neolithization ofFrance and Great Britain by 2000 BC. Up to 11% of theregional land surface is subject to cultivation by 2000 BC(fig. 5, left panel). The pattern of fractional crop cover fol-lows the spatial distribution of population density in allareas with fully developed agriculture. Maximum simula-ted crop fraction occurs on the Balkan, in the Levant, insouthern and northern China, and in Japan. For compari-son, fig. 5 (right panel) shows the crop fraction interpolatedfrom the HYDE database (Klein Goldewijk et al., 2007)onto the model regions. On this map, only the Levant andnorthern Italy appear as regions with a large crop fractionof up to 10%. Most of the southern Europe, the Levant, andChina exhibit a crop fraction on the order of 1%. No cropsare apparent in India, the Maghreb, or western and sou-thern Africa.

Using the data from northern American temperate broad-leaved forest as a prototype with ≈10·103 t C/km2 leaf andstem biomass, the simulated cropland share was convertedto carbon loss from deforestation. From 12000 BC to7000 BC, a total of 10 Gt C was lost to be replaced by crops;by 2000 BC, this number increases to 27 Gt C. Potential car-bon density simulated with the DGVM is shown in fig. 6(left panel). Large carbon stock is present in the temperateforests of Europe (250 t C/ha), in subtropical east Asia (300 tC/ha), and the tropics (<350 t C/ha), while the carbon stockin semi-arid areas is below 150 t C/ha: Most regions withearly agriculture are forested, except the river-based areasalong Indus, Euphrates, Tigris and Nile. The carbon releasefrom the potential forest by land use conversion to crops isshown in fig. 6 (right panel). Carbon is released mainly insouthern Europe, and south and east Asia. Largest regionallosses occur in Japan, northern Greece, and south China,with up to 36 t/ha removed carbon. Globally, the dynamicvegetation model simulation gives a total carbon loss of11 Gt by 7000 BC, and 29 Gt by 2000 BC. The heavilyconverted landscapes of China, southeast Asia, south Asia

and western Eurasia contribute around 90% of this carbonloss; the later agricultural areas of subsaharan Africa and theAmericas contribute the remaining 10% (tab. 1). The attri-bution can be further broken down to the country size level.For example, the carbon emission due to land use conver-sion in the western Eurasian agricultural centre (correspon-ding in the model to the region of today’s Israel and Leba-non) is 160 Mt.

Discussion

The simulation results shown are realistic hindcasts of po-pulation density, Prehistoric subsistence economy and resul-ting deforestation, at annual resolution for 685 world regions.It is important to reconsider how this realistic hindcast cameinto existence to assess its utility for advancing our unders-tanding of early human-climate interaction. The model para-meter space is small with only seven parameters; to find va-lues for these parameters, the simulation was run with identi-cal starting conditions and randomly varied parameter values106 times: the temporal and spatial distance of the first occur-rence of simulated farming societies to the undisputed five in-


Fig. 6 – Map of potential carbon density (left panel) and deforested carbon density by replacement with crops, including partial soilloss (right panel). The boxes define selected subcontinental-scale regions, for which the total carbon emission is listed in tab. 1.

Fig. 6 – Cartes de densité potentielle de carbone (vignette de gauche) et de densité de carbone causée par la déforestation et deremplacement par les cultures incluant en partie des pertes de sols (vignette de droite). Les cartons définissent les régions choisiespour la présente étude et pour lesquelles les émissions de carbone sont listées dans le tab. 1.

Region Carbon emission

Americas <1 Gt

Europe & Middle East 5.2 Gt

Indian subcontinent 6.1 Gt

Subsaharan Africa 2.3 Gt

East Asia 11 Gt

Southeast Asia 6.7 Gt

World 29 Gt

Tab. 1 – Total and regional carbon emissions due to land use.See fig. 6 for a definition of the subcontinental-scale regions.

Tab. 1 – Emissions totales et régionales de carbone liées à lamise en valeur agricole des terres. Voir fig. 6 pour la définition deszones régionales (sub-continentales.

dependent centres for domestication by B. Smith (1997) isused to score each simulation run; only the best scoring si-mulation run is selected for further dissemination. Thus, theemergence of the Levant and Chinese regions as early agri-cultural centres in the model is the result of the imposed dataconstraints, and not a model prediction. The added valueprovided by the model is manifold. Foremost, the i) tempo-ral and ii) spatial interpolation delivered by the modelcreates a consistent development trajectory of all world re-gions, despite widespread lack of continuous data. iii) It al-lows to test the sensitivity of simulated histories to externalforcing and hypothesis testing in the archaeological com-munity: K. Wirtz and C. Lemmen (2003) and C. Lemmenand K. Wirtz (2009) employed the model to assess socio-cultural vulnerability to climate fluctuations on interconti-nental and intracontinental scales. In the light of the largeuncertainties with the scaling parameters used to calculatedeforestation in terms of carbon loss, it is not the absolutevalue of global deforestation but the continuous and explicitspatio-temporal attribution of deforestation which is a novelaspect in palaeoclimate reconstruction and modelling.

Even considering the above caveats, the simulated totaldeforestation of 29 Gt (equivalent to a forest/woodland lossof 2·106 km2) is in line with published estimates. The in-crease between 6000 BC and 2000 BC is almost linear, ex-trapolation to preindustrial times would roughly double theprehistoric value to 54 Gt. Already in 1983, E. Matthewspresented based on satellite observations an estimate for lossof woodlands and forests since the beginning of agriculture(Matthews, 1983). She gives a number of 108 km2 of lost fo-rest and woodland (≈100 Gt C using the scaling of thisstudy), which fits well with the estimate presented here,when this is extrapolated to present day. K. Strassmann et al.(2008) estimated a preindustrial value of 45 Gt and relatedthis to a maximum atmospheric CO2 mixing ratio increaseof 3 ppm. A higher estimate for total preindustrial carbon re-lease due to agriculture was obtained by J. Olofsson andT. Hickler (2008). They find with the Lund-Potsdam-Jenavegetation model an emission of 114 Gt carbon from agri-culture, but include land use conversion outside forestedareas in their analysis. Even with their higher estimate, thetotal preindustrial emissions would not be sufficient toconfirm W. Ruddiman (2003)’s early anthropogenic green-house hypothesis. This study adds more support to dispro-ving an early anthropogenic greenhouse, though it is not en-tirely independent of the study by K. Strassmann et al.(2008 based on HYDE 3.0); furthermore, I show that the di-rect emission from deforestation and forest soil loss was toolow to have caused a significant rise of atmospheric carbondioxide. An equally large greenhouse effect may have beencaused by methane release from the conversion of wetlandsand from crop emissions (Ruddiman, 2003), both of whichare not considered in this study. K. Strassmann et al. (2008)found that carbon emission due to land demand for cropsand for pasture is of similar magnitude: the disregard of pas-ture in this study indicates that the calculation likely unde-restimates total carbon emission due to land use. This un-

derestimation, however, may be much less than 50%, be-cause the conversion from forest to pasture releases less car-bon than the conversion to crops (Guo and Gifford, 2002),and pasture may have been predominantly created from na-tural grasslands instead of forests.

In this study: - Only the contribution of removing forest and associated

soil loss is considered; we do not consider, e.g., the carbonrelease from converting peatland to agriculture, whichwould require a reconstruction of palaeosoils which is notyet included in GLUES.

- Interaction of land use with the climate system does notend with the removal of the standing forest and associatedbelowground carbon: important additional agriculturallyinfluenced drivers of the landscape with climate are, e.g.,surface roughness, albedo and latent heat flux. A descriptionof this interaction would require a bi-directionally coupledsystem consisting of a global climate model, a vegetationmodule and the sociotechnological simulator.

- Prior to the calculation employing the DGVM, a simpleconversion using aboveground carbon density of forest of100 t/ha gives a reasonable value for temperate forests butcan be much higher in tropical forest. Globally, a value of120 t/ha may be realistic though even this number is notwell constrained (e.g., Houghton, 2005). Surprisingly, thiscrude ‘back of the envelope’ calculation yielded almostidentical results as my simulation employing a dynamicalvegetation model and differential treatment of live carbonand soil carbon losses. The simple scaling is valid becausemost of the early farming areas are naturally forested areas,as was shown by the simulation of potential forest (fig. 6).

- The disagreement between the crop fraction predictedwith GLUES and the crop fraction derived from the HYDEdata base in most regions can be in part traced back to thesimulated population density which is an order of magnitu-de greater in the GLUES simulation for all those areaswhere farming is established. While in GLUES, the globalpopulation compares well to A. Coale (1974)’s estimate, theHYDE data base relies on interpolation of regional datafrom C. McEvedy and R. Jones (1978); Prehistoric popula-tion estimates differ by a factor of 10 between many studies.In view of the sound evidence for farming in many areaswhere the GLUES simulation shows deforestation (e.g.,Gronenborn, 1999; Huang et al., 2003; Staubwasser et al.,2003; Gupta, 2004; Habu, 2004), and considering the highLevantine fraction of crop also in the HYDE data base, theHYDE crop fraction estimates for southern and eastern Eur-asia are likely too low. The HYDE database was recentlyused by J. Kaplan et al. (2009) in a land use and land coverchange model which includes technological changes. Theirearliest hindcast of European deforestation is at 1000 BC; atthis point in time, their calculated sensitivity of lost forestfraction to the inclusion of technological change is 20% forEurope (Kaplan et al, 2009, their fig 8). This highlights theimportance of explicitly including technology (associated inthe GLUES model with economic change) in the calculationof past land use and not to rely only on population estimates.


- This study does not address the impact of Holocene cli-mate variability on the development of early agricultureand associated deforestation. The proxy data collection byK.W. Wirtz et al. (2009) showed that climate variabilityincreased in the second half of the Holocene in a Panameri-can corridor, where agriculture-based societies developedrather late – relative to western Eurasia and east Asia – des-pite early domestication successes; K. Wirtz and C. Lemmen(2003) demonstrated earlier that regular climate fluctuationsmay have impacted American societies more than Eurasiansocieties. Current preliminary studies with GLUES showthat susceptibility to climate changes in Europe is negligibleif these are not accompanied by loss of technology.

Conclusion

The GLUES model provides a consistent, realistic, andspatio-temporally continuous hindcast of prehistoric popu-lation density and socio-technological change between10000 BC and 2000 BC. In combination with potential fo-rest and carbon pool estimates from a dynamical vegetationmodel, deforestation and carbon release due to land clearingfor crops can be quantified on a global scale with regionalresolution. In total, 29 Gt of carbon were released by an-thropogenic deforestation between 10,000 BC and 2000 BCto give room to growing crops. Most of the release occurredin the heavily populated regions of southeast Europe, theLevant, south Asia and southeast Asia, China, and Japan, allof which are situated in a broad advanced technology beltranging from central Europe via the Middle East, India andIndochina to China and Japan. In contrast to the historicalcropland estimates found in the HYDE database, here cro-plands appear continuous in space throughout this farmingbelt. Small contributions to global deforestation are visiblein several parts of Africa, i.e., the Maghreb, western Africa,and southeast Africa. This study adds more evidence that theconsideration of economic and technological change in landuse calculations is important. Despite existing uncertainties inscaling parameters, the integration of databases, potentialland cover and technology modelling is a promising way toconsistent estimates of land use (change) and associated emis-sions – from the global to the regional level.

AcknowledgementsThe author is supported by the German national science

foundation’s priority program Interdynamik (DFG SPP 1266).V. Brovkin made available the VECODE model. GLUES wasdeveloped together with K. Wirtz; I thank him and two anony-mous reviewers for their helpful comments on an earlier ver-sion of this manuscript, and M. Ghilardi for the translation ofabstract, abbreviated version and captions into French.

References

Ackland G., Signitzer M., Stratford K., Cohen M. (2007) – Culturalhitchhiking on the wave of advance of beneficial technologies. Pro-ceedings of the National Academy of Sciences 104-21, 8714-8419.

Ambrose S.H. (1998) – Late Pleistocene human population bott-lenecks, volcanic winter, and differentiation of modern humans.Journal of Human Evolution 34, 623-651.

Axtell R., Epstein J., Dean J., Gumerman G., Swedlund A.,Harburger J., Chakravarty S., Hammond R., Parker J.,Parker M. (2002) – Population growth and collapse in a mul-tiagent model of the Kayenta Anasazi in Long House Valley.Proceedings of the National Academy of Sciences 99, 7292-7279.

Bogaard A. (2004) – The nature of early farming in central andsouth-east Europe. Documenta Praehistorica 31, 49-58.

Brovkin V., Ganopolski A., Svirezhev Y. (1997) – A continuousclimate-vegetation classification for use in climate-biospherestudies. Ecological Modelling 101, 2-3, 251-261.

Brovkin V., Bendtsen J., Claussen M., Ganopolski A., Kubatz-ki C., Petoukhov V., Andreev A. (2002) – Carbon cycle, vege-tation and climate dynamics in the Holocene: Experiments withthe CLIMBER-2 model. Global Biogeochemical Cycles 16-4,1139.

Childe V.G. (1936) – Man makes himself. Watts, London, 275 p.Coale, A., (1974). The history of the human population. Scientific

American 231-3, 40-51.Crutzen P.J., Stoermer E.F. (2000) – The ‘Anthropocene’. IGBP

Newsletter 41, W. Steffen, National Academy of Sweden, Stock-holm, 17-18.

Curtis P., Hanson P., Bolstad P., Barford C., Randolph J.,Schmid H., Wilson K. (2002) – Biometric and eddy-covariancebased estimates of annual carbon storage in five eastern NorthAmerican deciduous forests. Agricultural and Forest Meteorolo-gy 113, 1-4, 3-19.

Dieckmann U., Law R. (1996) – The dynamical theory of coevo-lution: a derivation from stochastic ecological processes.Journal of Mathematical Biology, 34, 579-612.

Flannery K.V. (1973) – The origins of agriculture. Annual Reviewof Anthropology 2-1, 271-310.

Fort J., Jana D., Humet J. (2004) – Multidelayed random walks:Theory and application to the Neolithic transition in Europe.Physical Review E 70, 3, 867-870.

Gregg S. (1988) – Foragers and farmers: population interactionand agricultural expansion in prehistoric Europe. University ofChicago Press, Chicago, 275 p.

Gronenborn D. (1999) – A variation on a basic theme: The transi-tion to farming in southern central Europe. Journal of WorldPrehistory 13-2, 123-210.

Guo L.B., Gifford R.M. (2002) – Soil carbon stocks and land usechanges: a meta analysis. Global Change Biology 8, 345-360.

Gupta A. (2004) – Origin of agriculture and domestication ofplants and animals linked to early Holocene climate ameliora-tion. Current Science 87-1, 54-59.

Habu J. (2004) – Ancient Jomon of Japan. Cambridge UniversityPress, Cambridge, 332 p.

Houghton R. (2005) – Aboveground forest biomass and the globalcarbon balance. Global Change Biology 11-6, 945-958.

Huang C.C., Zhao S., Pang J., Zhou Q., Chen S., Li P., Mao L.,Din M. (2003) – Climatic aridity and the relocations of the Zhouculture in the southern loess plateau of China. Climatic Change61, 361-378.


Kaplan J., Krumhardt K., Zimmerman N. (2009) – The prehis-toric and preindustrial deforestation of Europe. QuaternaryScience Reviews, 28-27/28, 3016-3034.

Klein Goldewijk K., Van Drecht G., Bouwman A. (2007) – Map-ping contemporary global cropland and grassland distributionson a 5 × 5 minute resolutionn. Journal of Land Use Science 2-3,167-190.

Leemans R., Cramer W.P. (1991) – The IIASA database for meanmonthly values of temperature, precipitation and cloudiness of aglobal terrestrial grid. Research report, International Institute ofApplied Systems Analyses, Laxenburg, Austria, 68 p.

Lemmen C., Wirtz K.W. (2009) – Socio-technological revolu-tions and migration waves re-examining early world history witha mathematical model. In Gronenborn D., Petrasch J. (Ed.): TheSpread of the Neolithic to Central Europe, Mainz, 2006. RGZMTagungen, in press.

Matthews E. (1983) – Global vegetation and land use: New high-resolution data bases for climate studies. Journal of Applied Me-teorology 22-3, 474-487.

McEvedy C., Jones R. (1978) – Atlas of world population histo-ry. Puffin Books, Harmondsworth, Middlesex, 368 pages.

Olofsson J., Hickler T. (2008) – Effects of human land-use on theglobal carbon cycle during the last 6,000 years. Vegetation His-tory and Archaeobotany 17-5, 605-615.

Ruddiman W. (2003) – The anthropogenic greenhouse era beganthousands of years ago. Climatic Change 61-3, 261-293.

Field, .C.B., Raupach, M.R. (2004) – The Global Carbon Cycle.The Global Carbon Cycle: Integrating Humans, Climate, andthe Natural World, Scientific Committee on Problems of theEnvironment Series 62, Island Press, Washington, 526 p.

Shennan S. (2001) – Demography and cultural innovation: amodel and its implications for the emergence of modern humanculture. Cambridge Archaeological Journal 11-1, 5-16.

Simmons I.G., (1996) – The environmental impact of later Meso-lithic cultures: the creation of moorland landscape in Englandand Wales. Edinburgh University Press, Edinburgh, 260 p.

Smith B. (1997) – The initial domestication of Cucurbita pepo inthe Americas 10,000 years ago. Science 276-5314, 932-934.

Staubwasser M., Sirocko F., Grootes P., Segl M. (2003) – Climatechange at the 4.2 ka BP termination of the indus valley civilizationand Holocene south Asian monsoon variability. Geophysical Re-search Letters 30-8, 1425, doi:10.1029/2002GL016822, 2003.

Strassmann K., Joos F., Fischer G. (2008) – Simulating effects ofland use changes on carbon fluxes: past contributions to atmos-pheric CO2 increases and future commitments due to losses ofterrestrial sink capacity. Tellus, B 60-4, 583-603.

Turney C.S.M., Brown H. (2007) – Catastrophic early Holocenesea level rise, human migration and the Neolithic transition inEurope. Quaternary Science Reviews 26, 17-18, 2036-2041.

Williams M. (2000) – Dark ages and dark areas: global deforestationin the deep past. Journal of Historical Geography 26-1, 28-46.

Wirtz K.W., Eckhardt B. (1996) – Effective variables in ecosys-tem models with an application to phytoplankton succession.Ecological Modelling 92, 33-54.

Wirtz K., Lemmen C. (2003) – A global dynamic model for theNeolithic transition. Climatic Change, 59-3, 333-367.

Wirtz K.W., Bernhardt K., Lohmann G., Lemmen C. (2009) –Mid-Holocene regional reorganization of climate variability.Climate of the Past Discussions 5, 287-326.

Zeder M. (2008) – Domestication and early agriculture in the Medi-terranean basin: Origins, diffusion, and impact. Proceedings of theNational Academy of Sciences 105-33, 11597-11604.

Zvelebil M., Dolukhanov P. (1991) – The transition to farming in eas-tern and northern Europe. Journal of World Prehistory 5-3, 233-278.

Article soumis le 8 juin 2009, accepté le 20 octobre 2009


world distribution of land cover changes during pre- and

Documents