farmers, flames, and forests: historical ecology of pastoral fire use

Forest Ecology and Management xxx (2013) xxx–xxx

Contents lists available at ScienceDirect

Forest Ecology and Management

journal homepage: www.elsevier .com/locate / foreco

Farmers, flames, and forests: Historical ecology of pastoral fire useand landscape change in the French Western Pyrenees, 1830–2011

0378-1127/$ - see front matter � 2013 Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.foreco.2013.10.021

⇑ Tel.: +1 7065426160.E-mail address: [email protected]

Please cite this article in press as: Coughlan, M.R. Farmers, flames, and forests: Historical ecology of pastoral fire use and landscape change in theWestern Pyrenees, 1830–2011. Forest Ecol. Manage. (2013), http://dx.doi.org/10.1016/j.foreco.2013.10.021

Michael R. Coughlan ⇑Postdoctoral Research Associate, Coweeta LTER, Sustainable Human Ecosystems Laboratory, University of Georgia, 250 Baldwin Hall, 355 S. Jackson Street, Athens,Georgia 30602-1619, USA

a r t i c l e i n f o a b s t r a c t

Article history:Received 14 August 2013Received in revised form 9 October 2013Accepted 14 October 2013Available online xxxx

Keywords:Historical ecologyLand use intensityLandscape transitionPastoral fire usePyrenees MountainsSocioecological interaction

The human use of fire is a major disturbance factor shaping the long term composition and patterning oftemperate forest landscapes. Yet, knowledge of the role of human agency in the historical dynamics offire in temperate forests remains vague. This paper presents a cross-scale Bayesian Weights of Evidenceanalysis of change in the spatial patterns of fire use over the last 180 years for a village territory in theBasque portion of the French Pyrenees. Research investigated the historical relationships between socialinstitutions that control land use, the spatial patterning of fire use, and landscape change. Analysis con-sidered the spatial contexts within which humans use and manage land: the household institution andthe parcel unit of land management. Bayesian methods established statistically significant associationsbetween social and ecological factors driving fire use and landscape change. These associations suggestthat social institutions differentially affected fire use patterns through inherited constraints. The resultingsocioecological legacies helped to explain the spatial patterns of landscape change. Uncertainty high-lighted in the modeling process suggests that we need a better understanding of the historical ecologicaldynamics of household institutions and land use change in order to better explain relationships betweenvariability in land use intensity and the fire regime.

� 2013 Elsevier B.V. All rights reserved.

1. Introduction

The human use of fire is a major disturbance factor shaping thelong term composition and patterning of temperate forest land-scapes (Delcourt et al., 1998; Foster et al., 2002; Tinner et al.,2005; Vanniere et al., 2008). Historical and ecological implicationsof fire use patterns are especially notable in the mesic, broadleaf-dominated forest landscapes of the western portion of thePyrenees Mountain range, where non-anthropogenic fires are rareand farmers continue to use pastoral fire (Métailié, 2006; Riuset al., 2009). In order to maintain ecological integrity of forest land-scapes, conservation efforts face the specter of reconstructing his-torical agrarian disturbance regimes in absence of the socialsystems that formerly drove them (Egan and Howell, 2001; Gimmiet al., 2008; Bürgi et al., 2013). This translates to a need to betterunderstand the ecological role of traditional management practicessuch as fire use in shaping and maintaining landscapes (Anderson,1996; Berkes et al., 2000; Peter and Shebitz, 2006; Agnoletti, 2007;Hobbs, 2009; Bugalho et al., 2011). Further, understanding the his-torical processes that contributed to landscape change is para-mount for designing policies that encourage sustainable

socioecological systems (Lambin et al., 2001; De Aranzabal et al.,2008).

Previous research on the historical ecology of human driven fireregimes focuses attention on changes in fire frequency, correlatingthese with shifts in human population densities or broadly-definedsociocultural attributes (Delcourt et al., 1998; Tinner et al., 1999;Guyette et al., 2002, 2006; Rius et al., 2009; Colombaroli et al.,2010). However, despite these efforts, knowledge of the role of hu-man agency in the historical dynamics of fire in temperate forestsremains vague. This is due, in part, to methodologies and researchdesigns that lack analytical reference to the levels of social andecological organization that link humans, fire use, and landscape.In the Pyrenees household social institutions use and manage thelandscape at the parcel-level.

This paper presents a cross-scale Bayesian Weights of Evidence(WoE) analysis of change in the spatial patterns of fire use over thelast 180 years for a village territory in the Basque portion of theFrench Pyrenees. Research investigated the historical relationshipsbetween social institutions that control land use, the spatial pat-terning of fire use, and landscape change. The research draws onan historical ecology approach that seeks to understand how pasthuman-environment interaction shapes contemporary landscapes(Crumley, 1994; Gragson, 2005).

Research in historical ecology often requires inference fromdiverse and indirect forms of evidence in order to link social and

French

http://dx.doi.org/10.1016/j.foreco.2013.10.021

mailto:[email protected]


http://www.sciencedirect.com/science/journal/03781127

http://www.elsevier.com/locate/foreco


2 M.R. Coughlan / Forest Ecology and Management xxx (2013) xxx–xxx

ecological parameters through time and space (Russell, 1997). Thisanalysis quantified spatial associations between current fire usepatterns, topography, land use change, and historical householdland use strategies. Associations between variables contribute toa spatially explicit understanding of how institutionally structuredland use strategies influenced fire use patterns and how, in turn,those patterns influenced landscape change. WoE is well suitedto historical ecological analyses because the method is quantita-tive, spatially explicit, data driven, and capable of incorporatingdiverse categorical data. While others have used WoE formodeling the spatial patterning of disturbance events (Dicksonet al., 2006; Poli and Sterlacchini, 2007; Romero-Calcerrada et al.,2008; Dilts et al., 2009), I use WoE to establish probabilitiesof associations between factors influencing the processes oflandscape change.

1.1. Farmers, flames, and landscape change

Pastoral fires in the Pyrenees are relatively small (mean of 10hectares), low severity, running surface fires. Livestock raisingfarmers use these fires to maintain pasture size and quality of for-age. Farmers set fires in late winter and early spring during fireweather windows when fuel moistures remain high in non-pastureland use but are sufficiently dry to permit the incineration of win-ter-cured grasses, shrubs, and dead wood in pastures (Coughlan,2013).

The practice of using fire to maintain pasture is thought to haveoriginated in the Levant (Naveh, 1975), spreading to WesternEurope in association with a suite of agropastoral practices, includ-ing slash and burn techniques of crop field and pasture creation(Kuhnholtz-Lordat, 1939; Sigaut, 1975; Trabaud, 1981; Métailié,2006). Paleoecological records from the Pyrenees suggest thatsince at least ca. 3000 before present (BP), human land and fireuse strongly dictated the regional fire regime (Rius et al., 2009,2012; Bal et al., 2011). By the Early Medieval period (ca. 1400BP), these same records show a sharp decline in forest clearanceand a probable transition to a fire regime dominated by pastoralfire use. Around the same time, the development of social organi-zation in the Pyrenees centered on autonomous household farmunits, most of which were established before ca. 800 BP (Cursente,1998; Bortoli and Palu, 2009).

From the Late Medieval period (ca. 15th century) until the 19thcentury, historical and paleoecological archives evidence a gradualexpansion of pastures at the expense of woodlands throughout thePyrenees (Métailié, 2006; Rius et al., 2009). While fire use is notimplicated as a cause of pasture expansion, such expansions likelyincreased the surface area under pastoral fire management. The1827 passage of the Code Forestier placed communal forests underthe jurisdiction of the French Administration des Eaux et Forets(French National Forest Service) which banned fire use within200 m of the forest edge. However, the regulation appears to havehad little actual impact on fire use in the study area since farmerscontinued to act autonomously with respect to land use andmanagement (Coughlan, 2013). Over the last 50 years, demo-graphic and socioeconomic changes resulted in regionally variableagricultural extensification and abandonment (Mottet et al.,2006). Analyses of these land use and management changes inthe eastern and southern portions of the Pyrenean range, whereagricultural abandonments occurred earliest, show increases inshrub and forest cover at the expense of cultivated lands(Vicente-Serrano et al., 2004) and decreases in landscapeheterogeneity (Roura-Pascual et al., 2005). Declining use offire is likely a proximal driver of landscape changes (Métailié,2006), but the specific relationships between changes inthe spatial patterning of fire use and the landscape remainunexplored.

Please cite this article in press as: Coughlan, M.R. Farmers, flames, and forests:Western Pyrenees, 1830–2011. Forest Ecol. Manage. (2013), http://dx.doi.org/1

1.2. The social context of farmers’ flames

Farmers cyclically initiate pastoral fires as part of a land use andmanagement regime that ultimately serves dynamic social andeconomic demands. Rule-based pastoral fire frequencies have beeninferred for particular biogeographic vegetation associations(Métailié, 1981) and specific land uses (Métailié, 2006). However,these do not address variability in fire management with regardsto the social processes and patterns driving fire use.

In the Western Pyrenees, as in many other agropastoral land-scapes, the household institution represents the principal unitof economic production and decision making over land use(Gómez-Ibáñez, 1975; Ott, 1993; Arrizabalaga, 1997). Pyreneanfarming households have usufruct rights to and decision makingpowers over a biophysically heterogeneous and often discontiguousset of land plots called parcels. Parcels are physically delineatedby natural or manmade boundaries and defined through their useand management, e.g. crop field, hay meadow, pasture, woodland.The spatial patterning of fire use emerges through a cross-scaleinteraction between households and parcels. However the longterm relationships between individual farming households andlandscape-level fire regime have never been modeled with empir-ical evidence.

As members of households, farmers make parcel level land usedecisions within the context of specific sociocultural institutionsand arrangements that determine patterns of ownership, access,and inheritance of land, capital, and other productive assets (Cole,1969, 1973; Netting, 1974, 1993; Barlett, 1976, 1980; Durrenberger,1980). Changes in the patterns of land use are driven by house-holder decision making constrained by the intersection of spatiallyheterogeneous social and biophysical contexts (Mottet et al., 2006).In this sense, a household’s land use strategy can be measured asthe cumulative outcome of parcel level land use decisions. Whenchanges in a household’s land use strategy cause changes in house-holder preferences for land use and parcel ownership, land use‘‘intensity’’ also changes. Modeling the spatially explicit relation-ships between household level land use intensity and parcel levelchange in land use and fire management holds promise for a morecomplete understanding of how human history is inscribed inlandscape patterns and processes.

1.3. Research questions

Despite ample documentation of the relationship between landuse change and landscape transformation, changes in the spatialpatterning of fire use remain largely undescribed. Further the dif-ferential effects of the social institutions that structure land useand fire management have not been established. This analysis ad-dressed two specific questions: (1) What relationships exist be-tween the biophysical landscape template, fire use practices, andland use change? (2) How does the historical institutional contextof land use and ownership (e.g. private household, pastoral collec-tives, and communal) account for spatial variability of changes inlandscape and fire use?

2. Materials and methods

2.1. Biogeographical setting

The Pyrenees are an east–west trending mountain range divid-ing the Iberian Peninsula from the rest of Europe and forming theborder between France and Spain. The western portion of the rangeis characterized by a humid, oceanic climate, with mild tempera-tures and relatively high amounts of precipitation (Gómez-Ibáñez,1975). Forests at lower elevations (up to 900 msl) are dominated

Historical ecology of pastoral fire use and landscape change in the French0.1016/j.foreco.2013.10.021


Fig. 1. Map of project area. Cartography, Michael R. Coughlan. Imagery courtesy of ESRI, Inc. under creative commons license CC By-NC-SA 3.0.

M.R. Coughlan / Forest Ecology and Management xxx (2013) xxx–xxx 3

by oak, transitioning to a mixture of beech and fir (800–1300 msl),while upper elevations (above 1300 msl) tend to be dominated byalpine and subalpine grasslands and heaths with patches of mixedconifer and pine (Gómez-Ibáñez, 1975; Ninot et al., 2007). Due tothe temperate climate and the east–west orientation of the moun-tains, biogeographic differences exist between cooler, wetter,north-facing slopes and dryer, warmer, south-facing slopes (Ricaand Recoder, 1990).

2.2. Field site

I selected a field site in the Basque portion of the FrenchWestern Pyrenees where farmers continue to use pastoral fires(Fig. 1). The field site covers approximately 125 square km and islocated in the upper Soule Valley with elevations ranging from300 to 2000 m above sea level. Rainfall averages approximately1700 mm per year.1 The local population continues to embracethe Basque ethno-linguistic identity, with the majority claimingSoulitine, the local Basque dialect, as its first language (Peaucelle,1977; Ott, 1993).

Prior to changes initiated in the 1980s, farming centered ondairy sheep and cheese production, but kitchen gardens, graincrops, cows, and pigs were also important sources of sustenanceand income (Lefèbvre, 1933; Peaucelle, 1977; Ott, 1993). Farmerspracticed seasonal transhumance, shepherding animals tocommunally owned, high elevation pastures (estives, French) insummer (Lefèbvre, 1928; Cavailles, 1931; Gómez-Ibáñez, 1975).In winter, farmers kept animals in infield fallows and barns (borda,Basque) where they bedded on bracken fern and ate hay and leaffodder harvested from both private and communal lands. Farmersused mid-elevation pastures, located in the forest matrix, for tran-

1 Data: Meteo France 1956–2011.


sitional forage in spring and fall (Palu, 1992). Because of their loca-tion in the forest matrix, mid-elevation pastures and the loweredge of the estives constitute the zone in which farmers use run-ning fires to prepare pastures for the spring.

The Pyrenean householder system was anchored in privatelyowned crop and hay infields surrounding the physical infrastruc-ture of house and barns (Palu, 1992). Spatially contiguous as wellas ‘‘islet’’ outfield hay and bracken meadows, pastures, and wood-lands also formed a part of the estate. Usufruct relationships ofspecific households to specific parcels have been systematicallypreserved in the historical record since 1830 through officialcadastral parcelization (Mottet et al., 2006; Bortoli and Palu,2009). Selective ground truthing, historical aerial photos, and theinherent constraints of mountain land use suggest that recordedchanges in parcel ownership and use represent reliable and spa-tially explicit proxies for mapping both parcel level land use andhousehold level land use strategy.

Importantly, household membership also conferred inheritableusufruct rights to communal property adjacent to the homesiteand to higher elevation communal pastures. The most significantof rights were access to pasturage, mast forage (for pigs andhumans alike), and collection of leaf fodder and small wood for fueland utensils (Palu, 1992; Cunchinabe et al., 2011; Coughlan, 2013).Rights to communal areas were not exclusively held by individualhouseholds, but shared through participation in inter-householdcooperative labor networks: the aizoak (Basque), first neighborinstitution and the olha (Basque) or cayolar (French), collective pas-toral institution (Ott, 1993). Households proportionally influencedland use on communal parcels through these networks. In theory,the aizoak was based in dyadic reciprocity (Ott, 1993), thus, neigh-boring households had equal rights to and the responsibility toequally share the resources of adjacent commons. Olha member-ship was formalized through a share system (txotx, Basque). Thetxotx system obliged the share owner to provide a set number of



Table 1Household extensification trends for households occupied between 1830 and 2010, standard deviations shown as (10).

1830a 1862 1931b 1963b 1977c 1990c 2010c

Actively farming households 102 �111d 70 60 43 31 24Average household landholdings (ha) 13.83 (11.01) 21.5 (10.02) 19.28 (8.12) 28.75 (9.83) 32.70 (15.31)Average household landholdings, extant in 2010 (ha) 17.68 (10.23) 21.74 (8.33) 31.5 (9.34) 32.70 (15.31)Average number of sheep per household 56 77 (61) 82 (73) 165 (118)Average number of cattle per household 9 14 (15) 17 (13) 36 (21)

a Landowning households only, figure does not include sharecropping households. Source: Cadastre Napoléonien.b All actively farming households. Source: Liste Electorale, Chambre Départemental d’Agriculture and Recensement de l’Agriculture et du Betail.c All actively farming households. Source: French agricultural subsidy records.d Estimate based on the sum of 1830 farming households and difference between the total number of house structures built and the number demolished between 1830 and

1862, multiplied by the percentage of total houses (173) to farming households (102) in 1830.


sheep (approximately 40) for summer cheese production in the es-tives. This ensured that at the end of the summer, cheese producedat the Olha could be divided according to share. A txotx could besplit to allow for the participation of less wealthy households(Ott, 1993), but wealthy households were required to obtain moreshares in order to increase sheep contribution and cheese take.Thus, household use of communal pasture was limited by the num-ber of txotx each owned.

Population in the commune declined steadily following a peakin the mid-nineteenth century (Peaucelle, 1977). Over this sametime period, farming households in the village substantially de-creased in number and size while remaining households increasedtheir landholdings by absorbing those abandoned by neighbors andrelatives (Table 1). This process of household land use transforma-tion is consistent with other communes in the Western Pyreneesand appears to be most drastic for the period following the1960s (Mottet et al., 2006; Welch-Devine, 2010). By the 1980sfarmers adopted tractors and many transitioned from raising dairysheep to raising cattle and horses for meat. The 1980s also markedthe end of crop cultivation outside of kitchen gardens with farmersconverting formerly plowed fields into mechanically cut haymeadows, pastures, or ‘‘abandoned’’ fallow.

2.3. Data collection and transformation

2.3.1. Fire useI collected fire use permit requests and authorizations for the

years 1969–2010 from communal archives. Permit requests con-tained parcel level spatial information corresponding with the2003 cadastral survey. While permit information is incompletefor any given year, farmers do not request permits for parcels thatare never burned (Coughlan, 2013). Thus, the permit archive pro-vided a reliable sample of burned parcels. I entered this informa-tion into a database and linked the spatial information to adigitized version of the 2003 survey. I also observed, photo-graphed, and point located pastoral fire use on 35 separate daysduring the 2011 burning season. I used photographs, GPS points,and field notes on paper topographic maps to create a GIS layerof the 2011 burned area. In order to create a map layer of fireuse for the years 1969 through 2011, I combined the 2011 burnarea map with fire use permit map. This augmented the parcelsample since several parcels burned in 2011 were not in the permitarchive.

2.3.2. TopographyI used ArcGIS Spatial Analyst to construct elevation, slope steep-

ness (slope), topographic roughness (standard deviation of slope),and slope aspect (aspect) raster layers from a 50 m resolution dig-ital elevation model (DEM) of the study area.


2.3.3. Land use changeLand survey maps from 1830 and 2003 provided parcel level

spatial data with land use attributes. Maps from 1830 and 2003were scanned, digitized into shapefiles using ArcGIS, and linkedto attribute data. The parcel ‘‘nature’’ attribute is part of a fiscaltaxation system that assigns each parcel a stratified tax valuebased on the surface area by hectare (ha) and a predefined landuse typology. Beginning with the Napoleonic Cadastre of 1830,there were 41 fiscal land use designations for the study area. How-ever, in order to simplify and match 2003 land use categories, thesewere consolidated into 10 classes (Table 2). This classification wasfurther consolidated into four ‘‘change analysis’’ groups in order toensure the capture of landscape changes that are significant for fireuse patterning. For each map, parcels were merged into featureclasses based this land use classification to create the 1830 and2003 land use maps.

I created a map showing areas of land use change between thetwo time periods by overlaying the 1830 and 2003 land use mapsand intersecting them (Fig. 2). Land use change categories there-fore include categories of no change, i.e. pasture to pasture. I veri-fied and edited land use change categories using aerial photos from2003 in order to ensure validity of change classes.

2.3.4. Institutional context of 1830: Household land use strategyLand uses form a gradient of intensity in terms of labor and

nutrient inputs and biomass outputs (Mottet et al., 2006). Forexample, crop fields are more intensive than hay meadows andhay meadows are more intensive than pastures. The cadastral landuse tax captured this variability in order to more accurately tax thepotential income generating output of land. Consequently, I calcu-lated the average per ha tax value by land use category for 1830.For each farming household, I summed the area of each land useowned and multiplied this by the average tax value for the corre-sponding land use category. I operationalized the intensity of afarming household’s land use strategy as a function of that house-hold’s total land use tax divided by the total surface area owned.This provided an ordinal index of land use intensity: householdswith mostly high value crop fields were ranked as high intensitywhile households with mostly extensive low value pasture hold-ings were ranked as low intensity. The index provides a proxy forland use strategy since high intensity households likely specializedin crop production and low intensity household likely specializedin livestock production. Index values were assigned as an attri-butes to the GIS features representing the spatial footprints ofthe household landholdings. Parcels owned by non-farming house-holds, including some parcels owned by large landholders, wereexcluded from the analysis.

Due to the difficulty in linking households with specific plots ofcommunal land, I grouped communal land into one feature class. Isimilarly grouped multi-owner private parcels (‘‘indivisible’’ prop-erties) associated with olha institutions. These parcels occur in the



Table 21830 and 2003 Cadastral land use and land use change analysis categories.

Land use categories (French) Explanation (English) Change analysis categories

1830 2003

Bois B (Bois) General woodland WoodlandBois futaie BF (Bois Futaie) Forest (timber production) WoodlandBois and haut taillis, chataigneraie BT (Bois Taillis) Copice and pollard woodlands, Chestnut grove WoodlandBroussaille NA Shrubland OtherJardin, verger J,VE (Jardin, verger) Garden, orchard OtherLabour T (Terre) Plowed crop field FieldPré P (Pré) Cut hay meadow FieldPâture L (Lande) Pasture PastureTerre vague, vaine L (Lande) Waste land, low quality pasture PastureBâtiment,a cour et sol, canal S (Sol) Structural footprint, including modern roads and paved areas Other

a Bâtiment = maison, grange, cabane, and moulin.

1 20.50KM

Land Use Change Categories

Woodland to Woodland

Woodland to Pasture

Pasture to Woodland

Pasture to Pasture

Field to Field

Fig. 2. Land use change by analysis categories. Cutout for illustration of detail only.


mid-mountain area and the estives and are often partly owned byhouseholds from other villages. Communal and indivisible landsare both collectively owned and represent two types of inter-household land use strategies associated with extensive land use.

2.4. WoE

2.4.1. Analysis overviewWoE uses the known spatial distribution of dependent variable

occurrence, e.g. fire use locations, to create a conditional probabilitymap of occurrence given spatial associations between the depen-dent variable and any number of evidence maps (Bonham-Carteret al., 1989). For this analysis, I used a GIS application, ArcSDM(Sawatzky et al., 2009), to generate probability maps of fire useand land use change, given evidence derived from fire use, topo-graphic, and cadastral maps of the study area. WoE is a multivariateanalysis method that uses known locations of particular occur-rences to derive conditional probabilities of association betweenthe dependent variable and conditionally independent evidence(Bonham-Carter et al., 1989). The WoE use Baye’s Theorem to calcu-late posterior probabilities of conditional association:


PðDjBÞ ¼ PðBjDÞPðDÞ=PðBÞ

The method is well tested and described for a variety of spatialanalysis applications (Agterberg, 1992; Bonham-Carter, 1994;Mensing et al., 2000; Dickson et al., 2006; Poli and Sterlacchini,2007).

I used a three step approach to answer the research questions(Fig. 3). The stepped analysis built progressively on spatial associ-ations between layers. Therefore, I used a consistent sample space(n = 138,918, 30 m2 units) for all three steps.

Step 1 provides a probability map for backcasting fire use in thelandscape for any given time period. The research assumes unifor-mitarianism in ‘‘bottom up’’ controls on fire (Heyerdahl et al.,2001), such that topographic conditions conducive to fire use in2003 are likely to be the same for 1830. I chose topographic char-acteristics as independent conditions because they represent a ma-jor biophysical constraint on both mixed mountain agriculture(Netting, 1972) and fire ecology (Métailié, 1981). In addition,topography remains spatially fixed at the human time scale.

Step 2 uses the Step 1 probability map assess associations be-tween fire and land use changes from 1830 to 2003. Interpretation



Step 1 Step 3Step 2

Fire use locations

Topography

P(D)

P(B)

P(D|B)Topographic fire use probability

ResultsMap

WoE1

Topographic fire use probability

Land use change categories

PtoWWtoP PtoP WtoW

TFU1, TFU2,...TFU10

WoE2a

Associations between fire use and

land use change

WoE2b WoE2d

WoE2c

P(D)

P(B)

P(D|B)

Household land use strategies

FUPtoWFUWtoP

FUPtoPFUWtoW

Land use changein high fire use zones

HHLUStrat1, ...HHLUStrat 4, Communal, Indivsible

WoE3a

WoE3biWoE2d

WoE2c

Probability of land use change being associated with fire use given house-

hold land use strategy

P (D

|B) =

P (B

|D) P

(D) /

P(B

)

Fig. 3. Diagram of analysis steps showing the dependent variable (D), independent variables (B), posterior probability parameter (D|B), and results maps (WoEi).

Table 3WoE steps and analyses.

Step Analysis Training points Evidence

1 WoE1 Fire use 1969–2011 Topography2 WoE2a PtoW WoE12 WoE2b WtoP WoE12 WoE2c PtoP WoE12 WoE2d WtoW WoE13 WoE3a FU PtoW HH land use strategy3 WoE3b FU WtoP HH land use strategy3 WoE3c FU PtoP HH land use strategy3 WoE3d FU WtoW HH land use strategy


of the results of this step relies on logical consistencies with eco-logical theories of disturbance and successional processes (Whiteand Pickett, 1985). The results provide probability maps forchanges in the spatiotemporal patterns of fire use.

Step 3 uses the significant associations found in Step 2 to estab-lish probable relationships between the institutional context of1830 and the inferred changes in fire use patterns. The results ofStep 3 provide maps indicative of the historical legacy of socialinstitutions on changes in the fire management regime at the levelof individual land management units.

2.4.2. ProcedureThe ArcSDM GIS application uses a set of ‘‘training’’ points repre-

senting a sample of known occurrences or events and a set of the-matic evidence raster maps of potential predictive conditions. ForStep 1, I extracted training points from the 1969 to 2011 fire uselocation map. Parcels in which fires occur average about 4 ha butrange from 95 to 0.006 ha. In order to create a point layer withoutlosing spatial significance of the parcel area, I transformed the fireuse polygon layer into a 30 m spaced point layer (n = 6089). I thentook a 10% random sample of the original points for use as the train-ing point layer, thinning points to ensure 1 per 30 m2 sample unit.

For topographic evidence layers, I divided elevation, slope, andtopographic roughness into categorical ranges using the Jenks nat-ural breaks method: 3 categories each for elevation and slope and 4categories for roughness. The Jenks method minimizes variancewithin classes and maximizes variance between them (Jenks,1967). I divided aspect into 8 categories representing aspectranges located between the 4 cardinal and 4 intercardinal points(e.g. NNE, ENE, ESE, etc.). Slope, topographic roughness, and eleva-tion were combined in the final analysis to ensure conditionalindependence between layers: e.g. slope (SLb)+ roughness (Rb)+elevation (ELb) = SL1R1EL1, E1S1TR2, etc. This resulted in 44 binarytopographic evidence layers (36 SLbRbELb and 8 aspect layers).I used ArcSDM to test spatial associations between fire use trainingpoints and the topographic layers derived from the DEM. Theanalysis created a raster map (WoE1) of the conditional probabilityof fire use, given topography.

In Step 2, I used the Step 1 map (WoE1) to analyze the probableassociation between fire use probability and the 1830–2003 landuse change categories. In order to assess the significance of the


WoE1 map for each land use change category, I conducted fourseparate analyses (Table 3). I derived four sets of training pointsfrom the Pasture to Woodland (PtoW), Woodland to Pasture(WtoP), Pasture to Pasture (PtoP), and Woodland to Woodland(WtoW) land use change categories. I converted each land usechange polygon into a 30 m spaced point layer and randomly se-lected 1000 points from each layer. I then thinned the point layerto ensure one point per unit area (30 m2)

I selected these four categories of land use change because theymost clearly represent potential change and persistence in fire usepatterns in terms of spatial distribution and fire return interval(FRI): e.g. PtoW = longer FRI, WtoP = shorter FRI, PtoP = short FRI,low variability, WtoW = long FRI, low variability. Other land usechange categories are more likely to have experienced multiplechanges over the time period (Mottet et al., 2006), while pastureand woodland land uses offer more concrete evidence of distur-bance frequency and severity.

For the binary evidence layers, I used Jenks natural breaks tocreate 10 categorical fire use probability maps the from WoE1 ras-ter layer. These topographic fire use layers (TFU) range from lowfire use probability to high fire use probability (TFU1, TFU2, . . . ,TFU10). The Step 2 analysis produced four probability maps(WoE2a, WoE2b, WoE2c, WoE2d) corresponding with associationsbetween each land use change category and the TFU categories. Itthus provided a potential measure of the relative importance ofchanges in fire management to land use change processes.

In Step 3 I analyzed the relationships between household levelland use strategies from 1830 and land use changes associated




with fire use or disuse. I derived four sets of training points fromthe Step 2 training point sets. With the exception of WtoP points,I selected points with associated WoE probability values abovethresholds determined by the value at which a fitted trend linemodeling the weighted association between the land use changeand TFU evidence classes is equal to 0 (Fig. 4). This provided pointsfrom WoE2 probability classes that were positively weighted.WtoP points did not display a linear association with fire use clas-ses. Consequently, for WoE3b1, I selected WtoP training points inWoE2 locations above the 1st standard deviation above the meanprobability of association with fire use classes. This provided train-ing points most strongly associated with TFU classes, but thesepoints were associated with both low and high fire use probabili-ties, potentially conflating relationships. For WoE3b2, I selected asecond WtoP training point set from points that only intersectedwith TFU7 (WoE1 0.32–0.37 probability). This provided trainingpoints for WtoP that were significantly associated with higherprobability for fire use, excluding the rest. In the results section Icompare the associations derived from both WtoP training pointsets.

I derived binary evidence layers from the map of 1830 house-hold land use intensity. I used Jenks natural breaks to divide house-hold land use intensity into 4 ordinal classes ranging from high tolow intensity. Communal and indivisible parcels were also in-cluded as evidence classes, resulting in a total of 6 classes. The re-sults of Step 3 (WoE3a, WoE3b, WoE3c, WoE3d) provideprobability maps for changes in fire and land use localized aroundland use intensity at a scale relevant to social processes. It thusdemonstrates associations between linked land-fire use changesand specific household and inter-household land use strategies.

2.5. Significance

I evaluated importance and significance of each evidence classto WoE model by examining the contrast (C) positive and negativeweights (Bonham-Carter, 1994). A positive C indicates that pres-ence of the evidence layer increases chances of training pointoccurrence, while a negative C indicates the inverse conclusion.Importance of the evidence class to the model increases as positiveand negative C values move away from 0. Contrast significance isdetermined with a ‘‘studentized’’ test of significance. Evidence lay-ers whose studentized C value falls outside the 95% confidenceinterval (a studentized C value of < 1.64) do not contribute to theprobability raster map.

ArcSDM also uses the studentized value of the posterior proba-bility to generate a confidence map. I used this map to identify spa-tial locations with values below 1.64 (95% confidence interval) forall WoE analyses, to assess the potential effects of this uncertaintyon the results, and to ensure that uncertainties did not confoundresults of successive analyses.

Fig. 4. Evidential weights for land use change categories. X-axis: WoE1 probabilitycategories. Y-axis: weights assigned in WoE2 analyses. Significance of R2 values notcalculated due to small sample size.


Lastly, WoE assumes conditional independence (CI) of the evi-dence layers with respect to the training points. Although someconditional dependence is expected, it should be minimized to en-sure that probabilities are not inflated (Bonham-Carter, 1994). Iused the Agterberg–Cheng test in ArcSDM to assess the degree ofCI (Agterberg and Cheng, 2002). This test uses the training points,posterior probability map, and a map of standard deviation of theposterior probability to measure the significance of the differencebetween the number of expected sum of all posterior probabilities(T) and the actual number of training points used (n). Conditionaldependence is present when a one-tailed test finds T to be signifi-cantly greater than n. Conditional dependence is generally miti-gated by dropping some layers and combining others (Agterbergand Cheng, 2002; Dickson et al., 2006). I used tests for conditionaldependence on preliminary analyses to inform the 3 step design ofthe final analysis. Specifically, I dropped the communal land useclass in WoE3a and WoE3b, and I combined the elevation, slope,and topographic roughness evidence layers for WoE1.

3. Results

3.1. Step 1

WoE1 results produced a CI value of 90.4%. eight out of 44 topo-graphic evidence layers were significant (Table 4). Confidence inthe posterior probability was high with studentized valueof > 3.2. High elevation, flat and even areas displayed highestcontrast and are, therefore, the most important topographic char-acteristics associated with fire use. Southern aspects weigh in nextas positively associated, followed by northern aspects and low ele-vation, flat areas as negatively associated. The final posterior prob-ability map shows that fire use is largely consistent withtopographic features of the landscape (Fig. 5).

3.2. Step 2

Land use changed on less than 12% of the study area betweenfrom 1830 to 2003. Most land use change concerned a transitionto woodland. PtoW represented 73% of land use change used inStep 2 while WtoP represented only 27%. PtoP represented 86%of 1830 pasture land use and WtoW represented 91% of 1830woodland land use.

All of the land use change categories were significantly associ-ated with two or more WoE1 evidence classes (Table 5). WoE2a(PtoW) and WoE2c (PtoP) displayed a gradient in strength of asso-ciations in directions consistent with the known effects of fire dis-turbance on land cover (Fig. 4). For example, the high topographicfire use probability classes were strongly and positively associatedwith pasture persistence (PtoP). These same topographic fire useprobability classes were strongly negatively associated with pas-ture to woodland transition (PtoW). PtoW was positively associ-ated with topographic fire use probability class below 0.37, butlevels below 0.2 were more strongly associated. WoE2b (WtoP)and WoE2d (WtoW) did not follow a pattern consistent with whatwe might expect in terms of fire management (Table 5). However,inconsistencies found for WoE2d include TFU1, 2, and 4. Thesewere located in lower elevation, flat areas with higher intensityland uses, including most of the major roads and the main clusterof village houses. Once these values were removed, WoE2d dis-plays negative associations at higher topographic fire use probabil-ity classes and positive associations at lower probability classes(Fig. 4). Inconsistencies with WoE2b were positive at low probabil-ity classes, negative at moderate classes, positive again, then neg-ative at high classes. I discuss the relevance of theseinconsistencies in Step 3 results.



Table 4Fire use WoE1 results for significantly associated topographic classes. Positive weight and contrast values indicate strength of positive association while negative values indicatestrength of negative association. Significance is defined as a studentized contrast value > ±1.64, 95% confidence envelope.

Evidence class Area (Ha) Training Points W+ W� Contrast Student C⁄

SL1R1EL3 464 69 1.002 �0.048 1.051 6.322SSW 1481 188 0.733 �0.12 0.853 8.426SSE 1340 158 0.616 �0.088 0.703 6.589ESE 1351 122 0.0228 �0.03 0.039 2.275WNW 1286 77 �0.301 0.031 �0.332 �2.523NNE 2304 119 �0.478 0.092 �0.57 �5.303NWN 1921 75 -0.802 0.111 �0.913 �7.079SL1R1EL1 482 16 �0.982 0.029 �1.011 �3.801

2.5 50 1.25KM

WoE Fire Use Probability

Low : 0.0406045

High : 0.660444

Fig. 5. WoE fire use probability.

Table 5WoE2 contrasts for significantly associated topographic fire use (TFU) classes. Positive values indicate strength of positive association while negative values indicate strength ofnegative association. Significance is defined as a studentized contrast value > ±1.64, 95% confidence envelope. NS = not significant.

Evidence class Probability range WoE2a (PtoW) WoE2b (WtoP) WoE2c (PtoP) WoE2d (WtoW)

TFU1 0.07–0.1 0.9638 NS �1.2377 �0.7018*

TFU2 0.1–.12 1.3892 0.7045 NS �1.2035*

TFU3 0.12–0.14 NS 0.3509 �0.5735 0.9817TFU4 0.14–0.2 0.9684 NS NS �1.7143*

TFU5 0.2–0.25 NS �0.2004 �0.575 0.4888TFU6 0.25–0.32 0.2095 NS NS NSTFU7 0.32–0.37 0.2515 0.2923 0.2485 �0.4742TFU8 0.37–0.45 �0.4023 NS 0.7608 �1.2083TFU9 0.45–0.66 �0.8653 �0.5845 0.5316 �1.0265TFU10 >0.66 �2.0419 NS 1.4803 NS

* Values not used in calculating probability threshold for WoE3d training points.


Confidence in the posterior probability maps was high forWoE2b and WoE2c. WoE2a and WoE2d both showed small areasof high uncertainty (studentized posterior probabilities < 1.64),found along parcel edges. This uncertainty appears to be linkedwith inaccuracies in data transformation, for example in the over-lay and intersection of the 1830 and 2003 land use maps andthrough rasterization of land use change polygons.

3.3. Step 3

1830 Land use intensity is significantly associated with land-fire use change associations (Table 6). WoE3a (Low fire use, PtoW)was most strongly and positively associated with households inthe highest land use intensity class. The strength of associationgradually diminishes as land use intensity lessens, transitioningto a weak, negative association for communal and indivisible


properties. The communal class adds to conditional dependence ofthe model and is removed from the final probability map (WoE3a).

WoE3b2 is significantly and negatively associated with commu-nal lands and positively associated with indivisible lands andhousehold land use intensity class 2. As with WoE3a, communallands added to conditional dependence between variables. WoE3b1

is similarly positively associated with indivisible lands, but signif-icantly negatively associated with household land use intensity le-vel 3. Communal lands were not significant. In comparison, giventhat WoE3b2 uses training points significantly associated with onlyhigh fire use probabilities, WoE3b2 provides a stronger and moreconsistent link between land use strategy and changes in land-fireuse associations. However, the WoE3b1 confidence map showedvery low uncertainties at all probability levels, while uncertaintiesfor WoE3b2 probabilities were high (studentized probabilitiesof < 1.64) for the significant negative associations with household



Table 6WoE3 weight contrasts for significantly associated fire-land use change classes. Positive values indicate strength of positive association while negative values indicate strength ofnegative association. Significance is defined as a studentized contrast value > ±1.64, 95% confidence envelope. NS = not significant.

Evidence Class WoE3a (;TFU, PtoW) WoE3b1 (±TFU, WtoP) WoE3b2 ("TFU, WtoP) WoE3c ("TFU, PtoP) WoE3d (;TFU, WtoW)

Communal �1.2071a NS �0.6248a 0.5623 0.3444Indivisible �0.4389 1.0255 1.0329 0.5097 �1.7406HHLUStrat1 1.0738 NS NS NS NSHHLUStrat2 1.4579 NS 0.6394 �0.6592 �0.7725HHLUStrat3 1.4799 �1.7961 NS �1.2502 NSHHLUStrat4 1.5481 NS NS NS NS

NS = not significant.HHLUStrat = Household land use strategy (intensity categories 1–4)."TFU = Association with high topographic fire use probability.;TFU = Association with low topographic fire use probability.±TFU = Nondirectional association with fire use probability.

a Removed from final analysis due to conditional dependence between variables.


land use intensity. Thus, woodland to pasture transitions linkedwith fire use are most likely associated with lower land use inten-sities (indivisible lands), but uncertainty exists for this change cat-egory on all other land use strategies.

WoE3c shows that pasture persistence coupled with high TFU issignificantly associated with inter-household land use strategieson communal and indivisible lands. Pasture persistence is nega-tively associated with higher intensity household land use strate-gies. WoE3d shows a similar result for woodland persistence,which is positive on communal lands, but negative for the one sig-nificant household land use strategy class. However, unlike pasturepersistence, woodland persistence is negatively associated withindivisible properties.

4. Discussion

Consideration of the spatial contexts within which humans useand manage land proved fruitful for modeling the effects of socialinstitutions and fire use practices on landscape change. Analysisestablished statistically significant associations between the spa-tiotemporal patterning land use change, fire use, and historical so-cial institutions. These associations suggest that the institutionalcontext strongly determined the relationship between fire useand landscape change. For example, households differentially af-fected fire use patterns through inherited constraints concerningthe flexibility of their land use strategies. As each generationpasses the farm’s parcels onto the next, they pass on the con-straints and possibilities associated with that set of parcels. Fur-ther, shifting household land use preferences are not only subjectto the spatial constraints of socially controlled access, but alsothe topographically defined flammability of the landscape. Thesecombined factors helped to define the spatial patterning of fireand land use through time. Uncertainty highlighted in the model-ing process suggests that we need a better understanding of thehistorical ecological dynamics of individual households and landuse change in order to explain the relationship between the inten-sity of land use and the variability in fire use. Four key aspects ofthe analysis bear further explanation.

Firstly, topography appears to provide a reliable template forbackcasting the potential distribution of fire use. While topographydoes not ‘‘control’’ fire use patterning per se, it does furnish a sig-nificant constraint on both fire and land use. Some topographicconstraints are inherent in the functioning of mixed mountainagricultural land use across mountain ranges and cultures (Rhoadesand Thompson, 1975), e.g. higher elevations are not suitable forinfield cultivations and are better suited to extensive pastoraluses. However, farmers show a clear preference for fire use man-agement on southern aspects. This preference is consistent withphysiographic controls on vegetation (Ninot et al., 2007) and with


vegetation-fire dynamics in areas as diverse as Corsica (Mouillotet al., 2003), Appalachia (Flatley et al., 2011) and the NorthwesternUS (Heyerdahl et al., 2001).

Increased insolation and exposure to warm, southern windsduring the burning season provides farmers with significantlymore ignition opportunities on south-facing slopes. As an indexfor ignition opportunities, topographic fire use probabilities mayalso represent an index of fire frequency since the most ‘‘flamma-ble’’ areas provide for the most efficient use of time allocation inmaintaining pastures. These south-facing pastures burn more uni-formly, require less ignition points, and exhibit more predictablefire behavior. The conversion of gorse shrublands to grass-domi-nant systems is linked with higher fires frequency through the po-sitive feedback effects of grassy fuels (Santana et al., 2012).

Over the short term, patch level interactions between vegeta-tion dynamics and grazing likely help determine the frequencieswith which farmers burn pastures (Kerby et al., 2007). Yet, fordomestic animals, grazing pressure is a function of household graz-ing strategies which differentially affects vegetation through bothshepherd decision processes and relative numbers and proportionsof sheep and cattle owned by a household. For example, García-González et al. (1990) found that in the Western Pyrenees, sheepgrazing patterns were predominantly determined by shepherdpasturing strategies while cattle grazing patterns were determinedby forage abundance and availability. Positive feedbacks betweenpasture preference, grazing strategies, and topographic constraintson fire use may encourage a higher shrub-to-grass ratio on north-facing slopes and a lower shrub to grass ratio on south-facingslopes. Field observations from 2011 certainly support this notion.Thus, in labor limited situations, north-facing slopes provide lowerquality pastures.

Secondly, given topographic constraints on fire use, land usepersistence and change was largely directionally consistent withpreferences for more efficient use of land under the rationale of firemanagement. For example, pastures less conducive to fire use weremore likely to convert to woodlands. Long term pasture persis-tence (PtoP) was more likely in areas conducive to efficient firemanagement, i.e. slopes with southern aspects. Inversely, wood-lands persisted on north-facing slopes, in part because ignitionopportunities are rare. The exclusion of fire from northerly aspectsis facilitated by the microclimatic reality of the location. The oneexception to this rule was woodlands that converted to pasture(WtoP). These changes were less clearly linked with fire use prefer-ences, indicating stronger influence of social contexts (discussedbelow).

Thirdly, the importance of topographic constraints on fire use isrelative to a gradient of land use strategies structured through so-cial institutions. Land use strategies ranged from ‘‘extensive’’ oncommunal and indivisible lands where topography was more




important to persistence of land use to ‘‘intensive’’ on privatehousehold lands where topography proved more important to landuse change. The 1830 land management institutions provide a leg-acy of social constraint that resonates through time by modulatingthe effects of topography on land use decisions.

The dynamics of agricultural abandonment were such thathousehold estates that persisted absorbed the lands of abandonedhouseholds. This transition entailed fewer workers per ha; hencethe socioecological dynamic shifted toward a less labor intensiveland use regime. During the early 20th century, birthrates declined,exacerbating the deficit of laborers to farm surface area. House-holds preferentially adjusted the intensity of land use within theconstraints of the surface area and diversity of land they could ac-cess. The type and amount of land that a household decided to letafforest or actively reclaim depended, in part, on these particularratios. The importance of the legacy of land use strategies ca.1830 is reflected in the associations between households andchanges in coupled land-fire use patterns.

For example, a household with suboptimal pasture may obtainnew pastures through purchase or inheritance. However, limitedaccess to summer communal lands (txotx ownership) still con-strains the household’s ability to increase the number of animalsit manages. Thus, the household may stop burning the suboptimalpasture and allow it to afforest, perhaps meeting demands for fire-wood or other woodland services. Oral histories collected in 2011suggest that farmers planted chestnut trees (Castanea sativa) forleaf fodder and mast on some north-facing pastures ca. 1900(Coughlan, unpublished data).

Fourthly, inter-household institutions supporting collectiveland use strategies insulated parcels from extensification pro-cesses. As Netting pointed out, in the Alps, such institutions existto promote, ‘‘an efficiency of utilization that would be threatenedby fragmented private ownership; the potential for maintainingyields by enforced conservation; the equitable sharing of necessaryresources by all group members’’ (Netting, 1981:64). However, pri-vately owned indivisible parcels were associated with some wood-land to pasture changes. This intensification of land use is likely along term result of the proximity and function of the parcel to theolha shepherd cabin. Enclosure of communal woodlands by theFrench state in 1827 resulted in the imposition of a strict rationingof fire wood to the olhas. Private lands were not subject to thisstricture and likely faced the pressures of increased wood collec-tion. Wood collection in combination with grazing and fire useprobably converted the olha’s woodlands to pastures.

5. Conclusions

The changes in coupled land and fire use patterns analyzedabove highlight a give-and-take relationship where fire use issituated between household economic demand, historically con-tingent social constraints, and the ecological template. As thenumber of households decreased and access to land opened up,decreasing land use intensity resulted in land use strategiesmore in line with topographic constraints. This implies thattopographic constraints on fire use played a role in householdstrategies concerning the maintenance or transformation of land usein specific locations. However, equally important to determiningfire use patterns were the social constraints imposed by thehistorical institutional context governing the means of productionand distribution.

The implications of this historical ecological dynamic are thatcertain areas were more likely to have cycles of use, abandonment,and reclamation in response to socioeconomic changes at largerscales. Human influences on temperate forests landscapes werenever monolithic, but involved the dynamic and complex interplay


of human decision making, institutionally defined land use strate-gies, and social and ecological constraints. This conclusion has sig-nificant implications for disturbance-based land management.Beyond the obvious point that forest planning should seek compat-ibility with topographically defined fire management, stakeholdersranging from home owners to commune officials to national levelland managers should build collaborative partnerships that over-come land use incompatibilities introduced through institutionallegacies. For example, officials might consider implementing firemanagement plans that allow for fire to traverse agriculturallyabandoned private lands (where topographically appropriate) asa means to mitigate potential wildfires. At the same time, the levelof variability in land use intensity introduced by historical institu-tions might serve as a means to justify trade-offs between stake-holders demanding different types land uses. For example, hikersand hunters may desire more forest cover while farmers requirepasture. Managers can reach compromises by adjusting for desiredpasture to forest ratios using the historical range of institutionalvariability in land use intensity. They can further ensure land usecompatibility by leaving areas with moderate fire use probabilitiesopen to this type of negotiation, thus reserving high and low prob-ability areas for most compatible uses.

While much work remains to be done, the findings presentedhere suggest that research on the history and evolution of hu-man-fire-landscape interaction should scrutinize the levels of so-cial and ecological organization at which human-fire-landscapeinteraction occurs. For regions where sufficient information existsfor modeling the historical and social contexts of fire use, themethods presented in this paper hold promise for reconstructinghistorical processes responsible for fire use patterns.

Acknowledgements

This paper was developed under STAR Fellowship AssistanceAgreement No. FP917243 awarded by the U.S. Environmental Pro-tection Agency (EPA). It has not been formally reviewed by EPA.The views expressed in this paper are solely those of the author.Partial support was provided by the Laboratoire GEODE, Universitéde Toulouse – Le Mirail; the National Science Foundation throughan award to the Coweeta LTER Program (DEB-0823293); the Labo-ratoire ITEM, Université de Pau et des Pays de l’Adour; and a FACE-Partner University Fund award to the University of Georgia and tothe Université de Pau et des Pays de l’Adour. I am grateful to theMayor’s office at the study site, my informants, professors TedGragson, Dolorès De Bortoli, and Pascal Palu who provided crucialguidance, support, and access to data. Dominique Cunchinabe, SaraDe La Torre Berón, and John Chamblee assisted with data collec-tion, entry, and management, respectively. This manuscript wasgreatly improved by comments from the aforementioned profes-sors, Stephen Kowalewski, Albert J. Parker, Bram Tucker, and twoanonymous reviewers.

References

Agnoletti, M., 2007. The degradation of traditional landscape in a mountain area ofTuscany during the 19th and 20th centuries: Implications for biodiversity andsustainable management. Forest Ecology and Management 249, 5–17.

Agterberg, F.P., 1992. Combining indicator patterns in weights of evidence modelingfor resource evaluation. Natural Resources Research 1, 39–50.

Agterberg, F.P., Cheng, Q., 2002. Conditional independence test for weights-of-evidence modeling. Natural Resources Research 11, 249–255.

Anderson, M.K., 1996. Tending the wilderness. Restoration and Management Notes14, 154–166.

Arrizabalaga, M.-P., 1997. The stem family in the French Basque country: sare in thenineteenth century. Journal of Family History 22, 50–69.

Bal, M.-C., Pelachs, A., Perez-Obiol, R., Julia, R., Cunill, R., 2011. Fire history andhuman activities during the last 3300cal yr BP in Spain’s Central Pyrenees: thecase of the Estany de Burg. Palaeogeography, Palaeoclimatology, Palaeoecology300, 179–190.


http://refhub.elsevier.com/S0378-1127(13)00688-9/h0005

















Barlett, P.F., 1976. Labor efficiency and the mechanism of agricultural evolution.Journal of Anthropological Research 32, 124–140.

Barlett, P.F., 1980. Adaptive strategies in peasant agricultural production. AnnualReview of Anthropology 9, 545–573.

Berkes, F., Colding, J., Folke, C., 2000. Rediscovery of traditional ecologicalknowledge as adaptive management. Ecological Applications 10, 1251–1262.

Bonham-Carter, G.F., 1994. Geographic Information Systems for Geoscientists:Modelling with GIS. Pergamon, Tarrytown, New York.

Bonham-Carter, G.F., Agterberg, F.P., Wright, D.F., 1989. Weights of evidencemodelling: a new approach to mapping mineral potential. In: Agterber, F.P.,Bonham-Carter, G.F. (Eds.), Statistical Applications in the Earth Sciences.Geological Survey of Canada, Ottawa, Canada.

Bortoli, D.d., Palu, P., 2009. Le système maison comme déterminant de la pérennitéorganisationnelle. Revue française de gestion 192, 141–150.

Bugalho, M.N., Caldeira, M.C., Pereira, J.S., Aronson, J., Pausas, J.G., 2011.Mediterranean cork oak savannas require human use to sustain biodiversityand ecosystem services. Frontiers in Ecology and the Environment 9, 278–286.

Bürgi, M., Gimmi, U., Stuber, M., 2013. Assessing traditional knowledge on forestuses to understand forest ecosystem dynamics. Forest Ecology andManagement 289, 115–122.

Cavailles, H., 1931. La transhumance pyrénéenne et la circulation des troupeauxdans les plaines de Gascogne. Armand Colin.

Cole, J.W., 1969. Economic alternatives in the upper nonsberg. AnthropologicalQuarterly 42, 186–213.

Cole, J.W., 1973. Social process in the Italian Alps. American Anthropologist 75,765–786.

Colombaroli, D., Henne, P.D., Kaltenrieder, P., Gobet, E., Tinner, W., 2010. Speciesresponses to fire, climate and human impact at tree line in the Alps asevidenced by palaeo-environmental records and a dynamic simulation model.Journal of Ecology 98, 1346–1357.

Coughlan, M.R., 2013. Errakina: pastoral fire use and landscape memory in theBasque region of the French Western Pyrenees. Journal of Ethnobiology 33, 86–104.

Crumley, C.L., 1994. Historical Ecology: Cultural Knowledge and ChangingLandscapes. School of American Research Press; Distributed by the Universityof Washington Press.

Cunchinabe, D., Palu, P., De Bortoli, D., Lavergne, M.-P., 2011. Histoire del’écosystème cultivé et de la biodiversité à Larrau: Analyse de l’impactagropastoral dans la gestion du milieu. In: Rapport de recherche, ITEM(Equipe d’Accueil no. 3002), UPPA, Pau, France.

Cursente, B., 1998. Des maisons et des hommes: la Gascogne médiévale (XIe-XVesiècle). Presses universitaires du Mirail, Toulouse.

De Aranzabal, I., Schmitz, M.F., Aguilera, P., Pineda, F.D., 2008. Modeling oflandscape changes derived from the dynamics of socio-ecological systems: acase of study in a semiarid Mediterranean landscape. Ecological Indicators 8,672–685.

Delcourt, P.A., Delcourt, H.R., Ison, C.R., Sharp, W.E., Gremillion, K.J., 1998.Prehistoric human use of fire, the eastern agricultural complex, andAppalachian oak-chestnut forests: paleoecology of Cliff Palace pond,Kentucky. American Antiquity 63, 263–278.

Dickson, B., Prather, J., Xu, Y., Hampton, H., Aumack, E., Sisk, T., 2006. Mapping theprobability of large fire occurrence in northern Arizona, USA. Landscape Ecology21, 747–761.

Dilts, T.E., Sibold, J.S., Biondi, F., 2009. A weights-of-evidence model for mapping theprobability of fire occurrence in Lincoln county, Nevada. Annals of theAssociation of American Geographers 99, 712–727.

Durrenberger, E.P., 1980. Chayanov’s economic analysis in anthropology. Journal ofAnthropological Research 36, 133–148.

Egan, D., Howell, E.A., 2001. The Historical Ecology Handbook: A Restorationist’sGuide to Reference Ecosystems. Island Press, Washington, DC.

Flatley, W.T., Lafon, C.W., Grissino-Mayer, H.D., 2011. Climatic and topographiccontrols on patterns of fire in the southern and central Appalachian Mountains,USA. Landscape Ecology 26, 195–209.

Foster, D.R., Clayden, S., Orwig, D.A., Hall, B., 2002. Oak, chestnut and fire: climaticand cultural controls of long-term forest dynamics in New England, USA.Journal of Biogeography 29, 1359–1379.

García-González, R., Hidalgo, R., Montserrrat, C., 1990. Patterns of livestock use intime and space in the summer ranges of the western Pyrenees: a case study inthe Aragon valley. Mountain Research and Development 10, 241–255.

Gimmi, U., Burgi, M., Stuber, M., 2008. Reconstructing anthropogenic disturbanceregimes in forest ecosystems: a case study from the Swiss Rhone valley.Ecosystems 11, 113–124.

Gómez-Ibáñez, D.A., 1975. The Western Pyrenees: Differential Evolution of theFrench and Spanish Borderland. Clarendon Press, Oxford [Eng.].

Gragson, T.L., 2005. Time in service to historical ecology. Ecological andEnvironmental Anthropology 1, 2–9.

Guyette, R.P., Muzika, R.M., Dey, D.C., 2002. Dynamics of an anthropogenic fireregime. Ecosystems 5, 472–486.

Guyette, R.P., Spetich, M.A., Stambaugh, M.C., 2006. Historic fire regime dynamicsand forcing factors in the Boston Mountains, Arkansas, USA. Forest Ecology andManagement 234, 293–304.

Heyerdahl, E.K., Brubaker, L.B., Agee, J.K., 2001. Spatial controls of historical fireregimes: a multiscale example from the interior west, USA. Ecology 82, 660–678.


Hobbs, R., 2009. Woodland restoration in Scotland: ecology, history, culture,economics, politics and change. Journal of Environmental Management 90,2857–2865.

Jenks, G.F., 1967. The data model concept in statistical mapping. InternationalYearbook of Cartography 7, 186–190.

Kerby, J., Fuhlendorf, S., Engle, D., 2007. Landscape heterogeneity and fire behavior:scale-dependent feedback between fire and grazing processes. LandscapeEcology 22, 507–516.

Kuhnholtz-Lordat, L., 1939. La terre incendiée: essai d’agronomie comparée.Bruguier, Nimes, France.

Lambin, E.F., Turner, B.L., Geist, H.J., Agbola, S.B., Angelsen, A., Bruce, J.W., Coomes,O.T., Dirzo, R., Fischer, G., Folke, C., George, P.S., Homewood, K., Imbernon, J.,Leemans, R., Li, X., Moran, E.F., Mortimore, M., Ramakrishnan, P.S., Richards, J.F.,Skånes, H., Steffen, W., Stone, G.D., Svedin, U., Veldkamp, T.A., Vogel, C., Xu, J.,2001. The causes of land-use and land-cover change: moving beyond the myths.Global Environmental Change 11, 261–269.

Lefèbvre, T., 1928. La transhumance dans les Basses-Pyrénées. Annales deGéographie, 35–60.

Lefèbvre, T., 1933. Les Modes de Vie dans les Pyrénées Atlantiques Orientales.Armand Colin, Paris.

Mensing, S.A., Elston, R.G., Raines, G.L., Tausch, R.J., Nowak, C.L., 2000. A GIS modelto predict the location of fossil packrat (Neotoma) middens in central Nevada.Western North American Naturalist 60, 111–120.

Métailié, J.-P., 1981. Le feu pastoral dans les Pyrénées centrales: Barousse, Oueil,Larboust. Editions du C.N.R.S., Paris.

Métailié, J.-P., 2006. Mountain landscape, pastoral management and traditionalpractices in the northern Pyrenees (France). In: Agnoletti, M. (Ed.), TheConservation of Cultural Landscapes. CAB International.

Mottet, A., Ladet, S., Coqué, N., Gibon, A., 2006. Agricultural land-use change and itsdrivers in mountain landscapes: a case study in the Pyrenees. Agriculture,Ecosystems and Environment 114, 296–310.

Mouillot, F., Ratte, J.-P., Joffre, R., Moreno, J.M., Rambal, S., 2003. Some determinantsof the spatio-temporal fire cycle in a Mediterranean landscape (Corsica, France).Landscape Ecology 18, 665–674.

Naveh, Z., 1975. Evolutionary significance of fire in Mediterranean region. IsraelJournal of Botany 24, 51–52.

Netting, R.M., 1972. Of men and meadows: strategies of alpine land use.Anthropological Quarterly 45, 132–144.

Netting, R.M., 1974. Agrarian ecology. Annual Review of Anthropology 3, 21–56.

Netting, R.M., 1993. Smallholders, Householders: Farm Families and the Ecology ofIntensive, Sustainable Agriculture. Stanford University Press, Stanford,California.

Ninot, J.M., Carrillo, e., Font, X., Carreras, J., Ferre, A., Marsalles, R.M., Soriano, I., Vigo,J., 2007. Altitude zonation in the Pyrenees. A geobotanic interpretation.Phytocoenologia 37, 371–398.

Ott, S., 1993. The Circle of Mountains: A Basque Shepherding Community.University of Nevada Press, Reno, USA.

Palu, P., 1992. Rapports entre organisation sociale et écosytème dans la sociétépastorale souletine. Sociétés contemporaines, 239–264.

Peaucelle, D., 1977. Analyse d’une communauté rurale basque, Larrau en HauteSoule. In: Département d’Etudes Rurales. Université de Paris 7, Paris.

Peter, D., Shebitz, D., 2006. Historic anthropogenically maintained bear grasssavannas of the southeastern Olympic Peninsula. Restoration Ecology 14, 605–615.

Poli, S., Sterlacchini, S., 2007. Landslide representation strategies in susceptibilitystudies using weights-of-evidence modeling technique. Natural ResourcesResearch 16, 121–134.

Rhoades, R.E., Thompson, S.I., 1975. Adaptive strategies in alpine environments:beyond ecological particularism. American Ethnologist 2, 535–551.

Rica, J.P.M., Recoder, P.M., 1990. Biogeographic features of the Pyrenean range.Mountain Research and Development 10, 235–240.

Rius, D., Vanniere, B., Galop, D., 2009. Fire frequency and landscape management inthe northwestern Pyrenean piedmont, France, since the early Neolithic (8000cal. BP). Holocene 19, 847–859.

Rius, D., Vannière, B., Galop, D., 2012. Holocene history of fire, vegetation and landuse from the central Pyrenees (France). Quaternary Research 77, 54–64.

Romero-Calcerrada, R., Novillo, C., Millington, J., Gomez-Jimenez, I., 2008. GISanalysis of spatial patterns of human-caused wildfire ignition risk in the SW ofMadrid (Central Spain). Landscape Ecology 23, 341–354.

Roura-Pascual, N., Pons, P., Etienne, M., Lambert, B., 2005. Transformation of a rurallandscape in the Eastern Pyrenees between 1953 and 2000. Mountain Researchand Development 25, 252–261.

Russell, E.W.B., 1997. People and Land through Time: Linking Ecology and History.Yale University Press, New Haven.

Santana, V.M., Baeza, M.J., Marrs, R.H., 2012. Response of woody and herbaceousfuel to repeated fires in Mediterranean gorse Shrublands. International Journalof Wildland Fire 22, 508–514.

Sawatzky, D.L., Raines, G.L., Bonham-Carter, G.F., Looney, C.G., 2009. Spatial DataModeller (SDM): ArcMAP 9.3 Geoprocessing Tools for Spatial Data Modellingusing Weights of Evidence, Logistic Regression, Fuzzy Logic and NeuralNetworks. <http://arcscripts.esri.com/details.asp?dbid=15341>.

Sigaut, F., 1975. L’agriculture et le feu: rôle et place du feu dans les techniques depréparartion du champ de l’ancienne agriculture Européenne. Mouton and Co.,Paris.























































































































































http://arcscripts.esri.com/details.asp?dbid=15341






Tinner, W., Hubschmid, P., Wehrli, M., Ammann, B., Conedera, M., 1999. Long-termforest fire ecology and dynamics in southern Switzerland. Journal of Ecology 87,273–289.

Tinner, W., Conedera, M., Ammann, B., Lotter, A.F., 2005. Fire ecology north andsouth of the Alps since the last ice age. Holocene 15, 1214–1226.

Trabaud, L., 1981. Effects of past and present fire on the vegetation of the FrenchMediterranean region. In: Symposium on Dynamics and Management ofMediterranean-type Ecosystems. U.S. Department of Agriculture, San Diego,California.

Vanniere, B., Colombaroli, D., Chapron, E., Leroux, A., Tinner, W., Magny, M., 2008.Climate versus human-driven fire regimes in Mediterranean landscapes: the


Holocene record of Lago dell’Accesa (Tuscany, Italy). Quaternary ScienceReviews 27, 1181–1196.

Vicente-Serrano, S.M., Lasanta, T., Romo, A., 2004. Analysis of spatial and temporalevolution of vegetation cover in the Spanish Central Pyrenees: role of humanmanagement. Environmental Management 34, 802–818.

Welch-Devine, M., 2010. Local places, global influences: Pastoralism in Xiberoa andEU regulation. In: Vaccaro, I., Beltran, O. (Eds.), Social and Ecological History ofthe Pyrenees: State, Market, and Landscape. Left Coast Press, Walnut Creek,California.

White, P., Pickett, S., 1985. Natural Disturbance and Patch Dynamics: AnIntroduction. Academic Press, San Diego, California.





















farmers, flames, and forests: historical ecology of pastoral fire use

Documents