20
Ecological Modelling 165 (2003) 107–126 Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France) Raphaël Mathevet a,, François Bousquet b , Christophe Le Page c , Martine Antona c a Centre de Recherche en Géographie et Aménagement Université Lyon 3 & Station Biologique de la Tour du Valat, Le Sambuc, F-13200 Arles, France b IRRI-Cirad, IRRI-Thailand Office Rice Research Institute Building, Kasetsart Campus, Bangkhen, Bangkok P.O. Box 9-159, Chatuchak, Bangkok 10900, Thailand c CIRAD-GREEN, Campus International de Baillarguet, BP 5035, F-34032 Montpellier Cedex, France Received 4 March 2002; received in revised form 29 November 2002; accepted 7 January 2003 Abstract Understanding and predicting how bird populations and land cover respond to natural and anthropogenic changes is a major challenge for environmental planning. Multi-agent modelling enables horizontal relationships (spatial configurations) and vertical relationships (socio-economic organisation) to be integrated. GEMACE is a multi-agent model for simulating farming-hunting-duck interactions in the Camargue (Southern France). For an archetypal region, we simulated land-use conver- sion and ecological change in space and time resulting from the interaction between environment and human drivers. A wintering duck population is simulated and distributed heterogeneously in its habitats. The duck population is affected by various environ- mental factors such as land-use changes, wetland management, hunting harvest, and disturbance. Land-use decisions are made at farmland level by farmers and hunting managers. Important biophysical drivers are water and salt through land relief, land-use history, infrastructure, spatial neighbourhood, and current land use. Important human drivers are the economic conditions of the world agricultural market and land-use strategy. Through three scenario runs, we discuss the implications that can be drawn from this modelling application with regard to the viability of a spatial conservation management alternative and multi-field research on sustainable development. © 2003 Elsevier Science B.V. All rights reserved. Keywords: Agent-based simulation; Land-use changes; Hunting; Rice-farming; Waterbirds; Camargue; Wetland 1. Introduction As a major component of the African flyway, the Camargue (i.e. the Rhone River delta, lying on the Mediterranean coast of France) is a wetland of inter- Corresponding author. E-mail address: [email protected] (R. Mathevet). national importance for waterbirds (Heath and Evans, 2000) as well as one of France’s major waterfowl hunting areas. In terms of total cash income, revenue from leasing of hunting-rights rivals that of some agricultural activities, such as leasing of grazing or reed harvesting rights (Mathevet and Mesléard, 2002). The development of private hunting clubs has become one of the major events in the economic development 0304-3800/03/$ – see front matter © 2003 Elsevier Science B.V. All rights reserved. doi:10.1016/S0304-3800(03)00098-X

Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

Embed Size (px)

Citation preview

Page 1: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

Ecological Modelling 165 (2003) 107–126

Agent-based simulations of interactions between duckpopulation, farming decisions and leasing of hunting

rights in the Camargue (Southern France)

Raphaël Matheveta,∗, François Bousquetb,Christophe Le Pagec, Martine Antonac

a Centre de Recherche en Géographie et Aménagement Université Lyon 3& Station Biologique de la Tour du Valat,Le Sambuc, F-13200 Arles, France

b IRRI-Cirad, IRRI-Thailand Office Rice Research Institute Building, Kasetsart Campus, Bangkhen,Bangkok P.O. Box 9-159, Chatuchak, Bangkok 10900, Thailand

c CIRAD-GREEN, Campus International de Baillarguet, BP 5035, F-34032 Montpellier Cedex, France

Received 4 March 2002; received in revised form 29 November 2002; accepted 7 January 2003

Abstract

Understanding and predicting how bird populations and land cover respond to natural and anthropogenic changes is amajor challenge for environmental planning. Multi-agent modelling enables horizontal relationships (spatial configurations)and vertical relationships (socio-economic organisation) to be integrated. GEMACE is a multi-agent model for simulatingfarming-hunting-duck interactions in the Camargue (Southern France). For an archetypal region, we simulated land-use conver-sion and ecological change in space and time resulting from the interaction between environment and human drivers. A winteringduck population is simulated and distributed heterogeneously in its habitats. The duck population is affected by various environ-mental factors such as land-use changes, wetland management, hunting harvest, and disturbance. Land-use decisions are madeat farmland level by farmers and hunting managers. Important biophysical drivers are water and salt through land relief, land-usehistory, infrastructure, spatial neighbourhood, and current land use. Important human drivers are the economic conditions of theworld agricultural market and land-use strategy. Through three scenario runs, we discuss the implications that can be drawn fromthis modelling application with regard to the viability of a spatial conservation management alternative and multi-field researchon sustainable development.© 2003 Elsevier Science B.V. All rights reserved.

Keywords:Agent-based simulation; Land-use changes; Hunting; Rice-farming; Waterbirds; Camargue; Wetland

1. Introduction

As a major component of the African flyway, theCamargue (i.e. the Rhone River delta, lying on theMediterranean coast of France) is a wetland of inter-

∗ Corresponding author.E-mail address:[email protected] (R. Mathevet).

national importance for waterbirds (Heath and Evans,2000) as well as one of France’s major waterfowlhunting areas. In terms of total cash income, revenuefrom leasing of hunting-rights rivals that of someagricultural activities, such as leasing of grazing orreed harvesting rights (Mathevet and Mesléard, 2002).The development of private hunting clubs has becomeone of the major events in the economic development

0304-3800/03/$ – see front matter © 2003 Elsevier Science B.V. All rights reserved.doi:10.1016/S0304-3800(03)00098-X

Page 2: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

108 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

of the Camargue in recent decades. Making a size-able income from hunting involves intensifying themanagement of hunting marshes. The logic of thismanagement is in conflict with the conservation ofMediterranean wetlands (Tamisier and Grillas, 1994).Thus, understanding and predicting how game pop-ulations and land cover respond to natural and an-thropogenic changes represents a major challengefor regional planning and Mediterranean wetlandconservation.

Many landscapes all over the world have beenformed in a co-evolution between human systemsand the biophysical environment. Consequently, mod-els of these landscapes must integrate ecology andsocio-economics (Odum, 1982; Costanza et al., 1991;Turner, 1991; Grossmann, 1994). Numerous scientistsnow believe that the study of ecosystems requiresa multi-field approach in order not to neglect thebehaviour of the social groups involved in land man-agement (McDonnell and Pickett, 1993; Anderies,2000; Bossel, 2000). To take into account the com-plexity of the ecosystems involved, modelling hasbeen carried out for several years at an individuallevel—individual-based modelling—(Grimm, 1999;Turner et al., 1993). To model such complex pro-cesses, multi-agent systems (MAS) are increasinglyemployed (Ginot et al., 2002). MAS enable modelsto be built which integrate human beings as an ele-ment of the ecosystem. These models relate to naturalresources management, fishing in the Central deltaof the Niger River (Bousquet et al., 1993), the dy-namics of deer populations in Florida (Abott et al.,1995), wildlife management in Cameroon (Bousquetet al., 2001), irrigated systems in Senegal (Barreteauand Bousquet, 2000), the management of conflicts inregional planning (Ferrand and Deffuant, 1999), thestudy of the evolution of settlements, the dynamicsof systems of cities (Bura et al., 1993; Kohler et al.,2000) and cultural changes (Dean et al., 2000).

MAS can integrate socio-economic, ecological andspatial dynamics into one single model. They al-low a better understanding of how the properties ofhuman-constructed landscapes at a macroscopic levelcan arise from the interactions of system componentsat a microscopic level (Ferber, 1995). For this pur-pose, the effects of the interactions between a gameresource, hunting, and agriculture for an archetypalregion, whose systems of production are strongly in-

fluenced by the total economic context, were studiedusing a multi-agent model. Firstly, we present the fieldstudy and the objectives of the multi-agent modelling.We then describe the main stages of the modelling,and the results obtained for the three scenario runs.Finally, we discuss the implications of this exper-iment for both prospective, local development andmulti-field research into natural habitat conservation.

2. Background

The Camargue (43◦30′N, 4◦30′E) is a vast area(ca. 145,000 ha) of rice fields, marshes, halophyticscrublands and lagoons. Rice is the most widespreadcrop in the delta, either in rotation with wheat or inmonoculture, alternative crops being limited by soilsalinity (Barbier and Mouret, 1992). Agriculture inthe Camargue has been transformed over the last 50years through a number of rapid and notable pro-cesses of intensification and crop specialisation. Ricefarming has been a determining factor in the develop-ment of the Camargue, promoting the desalinisationof uncultivated lands and contributing considerably tothe development of the hydraulic network. Previousstudies showed that the response of farmers to Com-mon Agriculture Policy (CAP) reforms was stronglydependent on the farm business context, in particularfarm debt and the availability of land to develop othersources of income (Mathevet et al., 2002). The mar-ket also appears to be an important vector of changein agriculture on a local scale. In recent decades,the expansion of rice farming in the Camargue hasbeen carried out to the detriment of natural wetlands.Nowadays, ricefields are a component of the mosaicof ecosystems. Earlier studies have outlined the strongcomplementarity of ricefields with natural habitats forwaterbirds (Pirot et al., 1984). The recent develop-ment of hunting marshes at the expense of agriculturallands (restoration of former rice fields) is due to theneed of some landowners to secure a regular incomebecause of changing agricultural markets (Mathevetand Tamisier, 2002). Tamisier and Dehorter (1999)have highlighted the existence of functional units inthe Camargue, and defined a conceptual model ofspatial and social organisation of ducks. Within eachfunctional unit, winter teals (Anas crecca) in particu-lar have a diurnal roost and widely dispersed feeding

Page 3: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126 109

grounds. The ducks gather during the day in protectedareas (ca. 22,000 ha), whereas most of their night feed-ing grounds are located inside hunting areas. Most ofthe hunting areas are private hunting clubs, which areestates where the landowner leases seasonal huntingrights either directly to several hunters or, usually, toa third party who sub-leases to hunters, who create aninformal hunting club (Mathevet and Mesléard, 2002).This lease-broker may only be interested in makinga profit and thus may not have a long-term interestin the management of the land. Depending on hisinterests, the hunting manager determines the amountof the individual fees payable, specifies the numberof hunters, and establishes the internal regulations.

A study of the origins of the wildfowling leasingsystem on farms in the Camargue highlighted theimportance of individual decisions as well as spatialopportunities (Mathevet and Tamisier, 2002). Becauseof the hydraulic characteristics of the delta, the devel-opment of hunting activity is highly dependent uponagriculture. Although the preservation of huntingmarshes on farmlands is sometimes due to historicalreasons, it is very often due to spatial configurationsand natural constraints. The Camargue is thus a hi-erarchically arranged system of production where thedevelopment issues are the profitability of land useor the creation of other land uses. The ecological andsocio-economic consequences of individual manage-ment decisions go beyond the estates, and relate to thewhole delta on a different time scale. The underlyingassumption behind the interest of the multi-agent ap-proach is based on the role of individual decisions inspatial organisations at a meso or macroscopic level(Ferber, 1995; Ulanowicz, 1997).

3. Modelling of this ecocomplex: GEMACE1

The aim of this model is to determine (1) how thehunting activity development process will adjust to aspatial logic and (2) how this development will ad-just to the agricultural and ecological constraints ofthe estates. Our model aims to study the importance

1 GEMACE: GEstion des Milieux Agricoles et Cynegetiquescamarguais et ses Effets. French acronym for a multi-agent modelto simulate agricultural and hunting management of the Camargueand its effects.

of the hunting activity development process in the Ca-margue area according to three economic scenarios inorder to highlight the effects of the competition be-tween, and complementarity of, hunting and agricul-ture in this ecocomplex defined as geographical spacecomposed of interacting ecosystems whose mode ofdisturbance, spatial organisation, and common historycause a specific functioning to emerge (Blandin andLamotte, 1988).

3.1. General structure of GEMACE

We simulated a socio-economic dynamic betweenhunting manager agents (HMAs) and farmers agents(FAs), through the market of the wildfowling leas-ing system, in interaction with ecological and spatialdynamics (duck population, variation of land use ac-cording to the objectives and spatial constraints offarming, Fig. 1). The lowest level of organisation ofour model was that of the farm unit, an individualdecisional unit and functional hydraulic unit. Eachfarm was assumed to correspond to an estate. Ona regional scale, we simulated, corresponding to afunctional duck unit, the structuring of the waterfowlhunting area resulting from the individual functioningof farms in interaction with a nature reserve and otherhunting units. The developed model generalises thefindings of previous field studies (Mathevet, 2000)and integrates empirical information about heteroge-neous distributions of the duck resource spatially. Themodel is based on a spatial representation of the maintypes of estates, distributed around a nature reserve.

3.2. Modelling method

Our model was constructed using CORMAS2

(http://cormas.cirad.fr), a simulation tool based on theVisualWorks® programming environment softwareused for programming in Smalltalk® object-orientedlanguage (Bousquet et al., 1998). The CORMAS tooloffers some predefined generic entities which agentsand objects specific to each model can inherit, andfacilitates the design of a virtual landscape. The firststage in the construction of the model was to de-termine and specify the entities of which it would

2 Common-Pool Resources and Multi-Agents Systems.

Page 4: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

110 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

Fig. 1. The conceptual diagram of GEMACE multi-agent model.

be comprised. The three entities represented in ourmodel were: (1) the elementary spatial unit, the patch,which inherits the predefined generic entity “Cell”;(2) and (3) the two social entities “farmer/landowner”and “hunting manager”, which inherit the class“Communicant/Agent”, allowing them to exchangemessages. Once the characteristics (attributes) andbehaviour (methods) of each entity were specified, thesecond stage of modelling was to define the dynamicsof the interactions between entities. This consistedin organising the order of the interactions betweenspatial entities and agents.

4. GEMACE components

4.1. The virtual landscape

The region was represented by a space-grid of 2700cells (45 lines× 60 columns). The cells were square,with a connexity of eight, and closed spatial bound-aries. The cell was the smallest unit of analysis, andsupposed to be internally uniform. Each cell corre-sponded to an elementary spatial unit of 2 ha, which

corresponded to the average surface area of agricul-tural fields compatible with the duck population sim-ulation (Chauvelon, 1998; Pirot et al., 1987). Eachcell had various attributes including land use, relief,the presence of a public path and main network ofirrigation and finally, its estate’s identifier. These at-tributes were downloaded from several maps basedon our field study. Within the model five differentland-use types representative of the Rhone River deltawere available. The land-use systems could be di-vided into three groups, commercial land use (ricecrop, dry crop), bull grazing-related land use (foddercrop) and natural land use (permanent-flooded marsh,temporary-flooded marsh: halophytic scrubland). Foreach land-use type, surface cover data is available, asfor all the other parameters, on a yearly basis for thewhole region and for each estate. Land cover is deter-mined by land suitability (relief, past culture) and thefarming strategy (rotation).

The initial land cover corresponds to the currentland use observed in the fluvio-lacustral part of theCamargue.Fig. 2shows the various viewpoints of thespatial grid depending on the attributes of the cells.The virtual region consisted of 18 estates of variable

Page 5: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

R.

Ma

theve

te

ta

l./Eco

logica

lM

od

ellin

g1

65

(20

03

)1

07

–1

26

111

Fig. 2. Various viewpoints of the virtual landscape and initial land cover common to simulation runs. (a) Hydraulic (dark) and public path (grey) networks; (b) relief(low-lying land in light grey); (c) estate delineation; (d) nature reserve (in black); (e) initial land cover (white: wheat crop, light grey: rice field, grey: fodder crop, dark greyhalophytic scrubland, dark: permanent-flooded marsh).

Page 6: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

112 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

surface areas, one of which was a nature reserve. Thedistribution of natural land uses was carried out ac-cording to the relief; we allocated marshes to low-lyingland and halophytic scrublands to higher land. Agri-cultural land use was preferentially allocated to higherland, starting from the irrigation network. The pro-portions of the various land uses and surface areasof utilised agricultural land were taken from the fieldsurvey statistical analysis (Mathevet, 2000).

4.2. The natural flooding process

Many authors highlight the importance of hydro-logical conditions in the carrying capacity of habitatsfor ducks (Tamisier and Dehorter, 1999). We used aflooding index provided by Alain Tamisier (CNRS,Montpellier). This index, established by plane over thelast 30 years, enables the flooded areas of halophyticscrublands and other temporary-flooded habitats to bemeasured. In the model, this index enabled the flood-ing of cells whose land use was halophytic scrubland.All marshes were considered flooded throughout theentire hunting season. The flooding of rice fields forhunting purposes depended on the management strate-gies of the agents as well as spatial neighbourhoodconditions.

4.3. The model for duck resource dynamics

We retained the granivorous model for winter tealsA. creccaof Tamisier and Dehorter (1999). Carryingcapacities of habitats are strongly influenced by win-

Table 1Distribution of the duck resource according to the month, the land use and the disturbance in number of ducks per cell (adapted fromTamisier and Dehorter, 1999; Pirot, 1982)

Land use Disturbancea Month 1 Month 2 Month 3 Month 4 Month 5 Month 6 Month 7

Flooded marsh ≤10 12 6 6 10 20 28 26]10, 15] 6 3 3 5 10 14 13>15 4 2 2 3 6 9 8

Flooded halophytic scrubland ≤10 0 30 30 8 12 16 24]10, 15] 0 15 15 4 6 8 12>15 0 10 10 2 4 5 8

Flooded rice field ≤10 0 0 18 34 48 28 0]10, 15] 0 0 9 17 24 14 0>15 0 0 6 11 16 9 0

a Disturbance (in hunting day per month) is assumed to be independent of the size of the hunting team.

ter climatic variability on seasonal and multi-annualtime scales. Because hunting disturbance causesunder-exploitation of potential feeding grounds, suchdisturbance has an impact at population level (Madsenand Fox, 1995). Cells were thus assumed to be loca-tions containing, or having the potential to contain,an aggregation of ducks. In order to simulate the dis-tribution of a duck population, we used the densitiesof ducks per hectare established by land-use typeaccording toPirot (1982). These densities integrate:

• The phenology of the presence of birds during thehunting season.

• The feeding resources of the land uses.• The energy demand of the wintering birds.

Consequently, the number of ducks in each cell canbe assumed to depend upon:

• Land use.• The month of the year.• The natural flooding index, and water management

by agents.• The hunting disturbance of the cell.

In order to simulate the hunting disturbance, it wasconsidered as a function of the number of hunting daysper month on the estate, regardless of the number ofhunters. The disturbance influences the distribution ofthe duck resource through the densities inTable 1.

Our simulated population was a population thatpartially depended upon our virtual region during thewintering period. Thus, we assumed that only a partof the total population winters within our spatial grid

Page 7: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126 113

during any given month. This transit process allowsfor the possible temporary and local extinction of thewintering subpopulation without driving the wholepopulation to extinction.

In order to define the size of the population depen-dent upon on our virtual region as a place of transitand wintering, we carried out a first series of simula-tions starting from the initial state of land use, withouthunting pressure or modification of the land coverduring simulation runs. This series of simulations de-fined the monthly carrying capacity of our functionalunit by determining the number of ducks. We obtaineda virtual population of 79,000 birds correspondingto 50% of the estimated transient and wintering pop-ulation of winter teals in the region (Tamisier andDehorter, 1999). The size of this population was usedat the initialisation of simulation runs.

In order to ensure the biological coherence of thepopulation’s dynamics, we assumed a birth rate ofb = 2 and a natural mortality rate ofmn = 0.65.With the equationr = b − mn for the population’snatural growth rate, we obtained the following modelof population dynamics:

xt+1 = r(xt − ht)

wherext denotes the size of the population at timet and h represents the rate of harvesting. We thusconsidered hunting to be an additional factor in mor-tality. We assumed that the age and size structures ofthe population were not affected by the harvest. Theresource therefore has its own dynamic. The latterdepends on land-cover installation decisions and thehunting effort of agents.

4.4. The simulated society

The principal actors of regional planning in theCamargue are the landowners who are bull breedersand/or rice farmers. However, the landowners are in-fluenced by the hunters because of the income theyreceive from the lease of hunting rights. Huntersare determining factors in the management and in-stallation decisions of the estate. We thus made thechoice to represent these two agents only, knowingthat depending on his strategy the landowner agentcan adopt the behaviour of a bull breeder. The socialstructuring which results from it is a function of thehunting-rights leasing market.

The agents are localised virtually insofar as theyexploit a defined space (one estate). The agents cancommunicate with one another. For the sake of clarity,we describe here only the main attributes and meth-ods necessary for understanding the functioning of themodel. The agents were defined by attributes (Table 2).The choice of strategies was established, using the pro-portions observed in real life. After consultation withagricultural experts, we made the choice to renew FAstrategy every 3 years. The HMA strategy is renewedannually.

We specified three strategies for FAs:

• A strategy for rice-crop monoculture.• A strategy for mixed-farming and leasing of natural

lands for pasture.• A strategy for mixed-farming and bull breeding.

For each strategy, there is a corresponding systemof crop rotation and an allocation of land use (Table 3).

Several authors highlight the importance of repre-sentations and values in the decision-making process(Jodelet, 1989; Beedell and Rehman, 2000). We thusdefined several values for each type of agent. The FAswere given three values: “Hunting, Peasant, and BullPassion”. These values were determined by the eco-nomic results of the farm, and the profitability of theland use. However, the aim of the “Bull Passion” valueis to limit the changes of strategy and thus the changeof allocation of land use of the estate. Thus, we simu-lated a bull breeder who would not upset the structureof his farm because of a change of agricultural eco-nomic situation. Unlike the other attributes, this oneis not put at zero at the initialisation of the simulationruns, but allotted 10 units depending on the strategyand size of the estate. The “Hunting” and “Peasant”values are determinant in the process of farmland man-agement towards hunting or agricultural management.

HMAs have two values: “Shooting Passion” and“Money”. The “Shooting Passion” value is a indirectfunction of the annual hunting harvest, i.e. of the sat-isfaction of the hunting teams. The “Money” valuecorresponds to the profitability of the activity for theHMAs. These two values, associated with the financialbalance of the past year, determine the choice of thestrategy. We distinguished three possible strategies:

• The “Elite” strategy, strongly dependent onthe land-use characteristics of the estates and

Page 8: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

114 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

Table 2The main attributes of the agents

Attributes Modalities FA HMA

Estate Estate’s identifier√ √

Bank Bank account√ √

The others Address of the other agents√ √

Farmer table Address of the FAs√

My hunting unit Collection of the leased cells for hunting purpose√

Agricultural area Cells that land-use attribute is agriculture√

Natural area Cells that land-use attribute is natural land√

Hunting area Cells that land-use attribute is natural and rice-field contiguousto marshes

√ √

Strategy The agricultural or hunting strategy√ √

Memory strategy Collection of the past strategies√ √

Selected Agreement or not with a HMA or FA√ √

Agricultural results Annual balance sheet√

Optimum results Optimum balance sheet√

Hunting results Equal to the hunting lease for FAs or hunting profit for HMAs√ √

My hunting manager Identifying number of the HMA√

My farmer Identifying number of the FA√

Hunting lease Amount of hunting lease√ √

Memory hunting lease Collection of the past amounts of hunting leases√ √

Agricultural satisfaction (AS) Satisfaction about the agricultural season√

Hunting satisfaction (HS) Satisfaction about the hunting season√ √

Memory AS Collection of the past AS√

Memory HS Collection of the past HS√ √

Peasant value Individual value about agriculture√

Hunting value Individual value about waterfowl hunting√

Bull value Individual value about bull breeding√

Passion value Individual value about hunting activity√

Money value Individual value about hunting profit√

Memory hunting bag Collection of the past hunting bags√

Reputation table Collection of the reputation of the HMAs or FAs√ √

simulating the functioning of a high-standard hunt-ing club (i.e. low team size and high individual fees).

• The “Passion” strategy which consists in devel-oping hunting activity and limiting the number ofhunters whilst balancing its costs and profits.

• The “Profitable” strategy is a mercantile strategy.This results in developing the activity by increasing

Table 3Farmer agent strategies and crop rotations

Agriculture strategy Size of the estate (number of cell) % of the land use within the agricultural area

Rice farming / 100% rice crop

Mixed-farming >137 70% rice crop, 30% dry crop≤137 50% rice crop, 50% dry crop

Mixed-farming and bull breeding >137 40% rice crop, 40% dry crop, 20% fodder crop≤137 35% rice crop, 30% dry crop, 25% fodder crop

the number of hunters. However, the cost of thefees is limited by the characteristics of the estates.

The hunting harvest is evaluated each month byHMAs using the following formula:

ht+1 = v × α × et × nDt × pDt

Page 9: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126 115

whereh is the hunting harvest;v corresponds to themonthly vulnerability coefficient of winter teals (i.e.the catchability) according toDehorter and Tamisier(1996); α is the multiplying coefficient of the huntingeffort which depends on the month and the strategy;eis the effort (i.e. the number of hunters of the huntingteam reported to the number of flooded cells for thepurpose of hunting during the month);nD is the num-ber of hunted days during the month;pD is the numberof ducks present in each cell of the hunting unit.

To simulate the impact of the disturbance generatedby public paths on small hunting units, the huntingharvest of the season was assumed to decrease by 50%for hunting units that cover fewer than 35 cells andare contiguous to a public path.

Hunting teams were defined as an attribute ofHMAs. They represent a social group whose numberdepends on the strategy of their hunting manager.The hunting teams have a method of evaluating theirsatisfaction with regard to the annual hunting harvest.For this we used the equation of the regression lineof the cost of the fees as a function of the huntingharvest obtained during our field survey (Mathevet,2000). The method consists in calculating a theoret-ical harvest according to the individual fee paid tothe hunting manager, then to locate the value of theeffective annual harvest in a range of 50% of this the-oretical harvest. If the effective harvest is not includedin this range, then the team will not be satisfied. Atthe initialisation, whatever the strategy, each huntingteam has 10 hunters, and the number of first-seasonhunting days for each hunting units is 15 per month.In addition, no rice field is flooded.

5. Schematic overview of the organisation of thesimulation

First, we describe the initialisation and the first com-munication between agents. We then go on to describethe seasonal dynamics of the system as a whole.

5.1. Initialisation and communications

A land cover was created and an estate given to eachFA. Before the first time step of the simulation run(Fig. 3), the control activated the FAs, who consultedthe cells of their estates to find out their characteristics(relief, land use, and location). From this information,

the FAs defined their agricultural, natural and huntingareas to determine their agricultural strategies. Oncetheir strategy was established, they proceeded to theallocation of the land use of their agricultural areasaccording to the proportions relating to the strategies.The land-use attribute of the cells was then modified.When all the FAs had carried out this series of oper-ations and when the cells had completed their changeof state, the HMAs asked all FAs to give details of thephysical characteristics of their estates (surface areaof the various land uses, adjacency to a reserve and/ora public path). From this information, the HMAs de-termined and classified in their attribute #Table theleasing price corresponding to each FA, and there-fore each estate. When all the HMAs had constructedthese tables, the control activated his #Selectmethod.This reset the #Selectedattributes of the agents, i.e.from now on each agent was regarded as non-selectedby the others. For the moment, there was no wild-fowling leasing agreement between FAs and HMAs.The HMAs then activated their method #makeYour-Proposal. In their #Table, they looked for an FA towhom they had not sent a proposal and sent a proposalto him. When an FA had three messages in his mail-box, he assessed the best proposal according to twocriteria: the amount of the proposal and the reputationof HMAs. If these were not consistent, only the repu-tation was taken into consideration. At the time of theinitial communication, all the HMAs had a reputationof zero, the FAs therefore made their choice based onthe best proposal. Under other circumstances, the FAswould have selected the HMA with the best proposaland the best reputation. Once the choice was made, theFA informed the HMA selected with an #ok message.When the HMA consulted his mailbox and found an#ok message, he modified his #Selectedattribute andleft the loop of control. From now on, he had been se-lected, so did not have to make another proposal. Inreturn he validated the link with his FA, who consid-ered that he had definitively rented his hunting rights,so before also leaving the loop, he informed the un-successful HMAs using #sorry messages. The HMAsthat received a #sorry message then made a proposalto a FA to whom they had not yet sent a proposal. Thisprocedure continued as long as FAs remained who hadnot yet validated a link with an HMA. Once all thelinks had been validated, the HMAs determined theirstrategy and the management of their hunting unit, and

Page 10: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

116 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

Fig. 3. Sequential diagram of the communication initialisation following by the #Selectmethod.

Page 11: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126 117

updated their attribute by storing the reputations of theFAs. When all the HMAs had carried out these ac-tions, the FAs also created a table containing the iden-tifiers of the HMAs, to which they would associate areputation at the time step of the seasonal assessment.Once the allocation of land use and communicationswas completed, the #EvolveResourcemethod of thecontrol updated the size of the duck population. Thecontrol then activated his #Evolvemethod (Fig. 4).

5.2. The whole evolution: a representation of thetotal dynamics

We began with a method that carries out the nat-ural flooding of the cells with the land-use attributeof “halophytic scrubland” (Fig. 4). Depending onthe month, the control distributed ducks according toflooding, land use, and disturbance of cells. It thenasked HMAs to evaluate, for each month, the numberof ducks present in their hunting unit, to determinetheir hunting effort and monthly hunting harvest.When all HMAs had carried out these instructions,we carried out the same procedure over the follow-ing months. After 7 months, the control activatedthe methods of the cells, enabling them to store intheir memory their land use, as well as any changesnecessary for later determination of the past-culture.The HMAs were activated in their turn. They evalu-ated their annual balance sheet (harvest, profits, andlosses) and evolved their memory in relation to vari-ous attributes (satisfaction, values, and harvests). Thecontrol determined the agricultural economic scenarioand authorised the FAs to calculate their general as-sessment, to modify their memories and carry outcrop rotations of their agricultural areas. Dependingon their profits, the FAs made some installations ontheir farmlands. The cells of any estate with newinstallations changed their land-use attribute. TheFAs then redefined their various areas (agricultural,hunting, and natural). The HMAs then sorted theirreputation table attributes, in order to make their firstproposal to an FA with whom they were satisfied.If the first hunting season was not satisfactory, theywould randomly make a first proposal to the FAs witha reputation of zero. The control activated the #Selectmethod as described before. When the communica-tions were finished, the HMAs modified their strate-gies and management methods, then decided upon the

flooding of some rice fields. Finally, before complet-ing the loop of duck resource allocation once again,the control carried out his evaluation starting from thetotal harvest which took place during the season.

6. Simulations of the farming-wildfowling-ducksinteractions

6.1. Scenarios

The first series of simulations integrated a sce-nario in which the development of agriculture waseconomically favourable: scenario A “high rice-cropprofitability”. It tested the capacity of hunting activ-ity to contribute to the maintenance of natural landson privately owned agricultural estates. Scenario Bcorresponded to an unfavourable agricultural context:rice farming scarcely profitable for large farms, andsmall farms in financial difficulties overdrawn forthe others. With CAP subsidies, wheat is thereforethe substitution crop. The aim of the scenario wasto test the capacity of hunting activity to contributeto the maintenance of farms and to seriously com-pete with agriculture during critical periods for theagricultural market. Finally, we simulated the alterna-tion of economic scenarios, the question being: couldhunting specialisation by the estates, as observed inthe Camargue, be obtained by simulating an eco-nomic alternation chronologically comparable withthe agricultural history of the Camargue?

The profits of the two economic scenarios are shownin Table 4. The most relevant indicators to characterisethe scenarios are:

• The changes in the duck resource representing thecarrying capacity trends of the whole region.

• The proportion of agricultural area in the regionwhich highlights the spatial development of huntingor agriculture.

• The principal parameters characterising hunting de-velopment, namely: the hunting harvest, the averagefee, the total number of hunters, the total turnover,and the total amount paid for hunting leases by FAs.

Each scenario was run 20 times to test the consis-tency of the results. Kruskal–Wallis one-way ANOVAtests were carried out for each series of simulationsand the significant difference in the scenario results

Page 12: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

118 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

Fig. 4. Sequential diagram of the #Evolvemethod.

Page 13: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126 119

Table 4Agricultural profits of the two economic scenarios

Agricultural area

Scenario A: “high rice-crop profitability” and rice-farming strategyPrecedent crop of the cell ≤60 =]60, 175] >175‘1111’ +640 if R1 if not +457 +732 if R1 if not +534 +899 if R1 if not +701‘x111’ +640 +732 +899‘xx11’ +1021 +1113 +1281‘xxx1’ +1021 +1113 +1281

Scenario A: “high rice-crop profitability” and mixed-farming or bull breeding strategiesPrecedent crop of the cell ≤60 =]60, 175] >175‘1111’ +305 if R1 if not +107 +381 if R1 if not +198 +549 if R1 if not +366‘x111’ +488 +579 +747‘xx11’ +869 +960 +1128‘xxx1’ +869 +960 +1128‘2222’ +274 if R1 if not +137 +274 if R1 if not +137 +412 if R1 if not +137‘x222’ +412 +412 +717‘xx22’ +488 +488 +793‘xxx2’ +488 +488 +793Land use = 3 +244 if AS = #C if not+107 +244 if AS = #C if not+107 +244 if AS = #C if not+107

Scenario B: “critical period for agricultural market” and rice-farming strategyPast crop of the cell ≤60 =]60, 175] >175

‘1111’ −244 if R1 if not −351 +2 if R1 if not −46 +107 if R1 if not +0‘x111’ −137 +46 +107‘xx11’ +61 +152 +320‘xxx1’ +61 +152 +320

Scenario B: “critical period for agricultural market” and mixed-farming or bull breeding strategiesPast crop of the cell ≤60 =]60, 175] >175‘1111’ −244 if R1 if not −351 +2 if R1 if not −168 +0 if R1 if not −107‘x111’ −137 −46 +107‘xx11’ +61 +152 +320‘xxx1’ +61 +152 +320‘2222’ +427 if R1 if not +290 +427 if R1 if not +290 +564 if R1 if not +290‘x222’ +564 +564 +869‘xx22’ +640 +640 +945‘xxx2’ +640 +640 +945Land use = 3 (+244 if AS = #C if

not +107) + 122(+244 if AS = #C ifnot +107) + 122

(+244 if AS = #C ifnot +107) + 122

AS: agricultural strategy of the farmer agent; #C: mixed-farming and bull breeding; R1: relief attribute of the cell (1: high land); 1: ricecrop; 2: dry crop (wheat); 3: fodder crop; x: other land use. Agricultural area in number of cell and profits are inper cell.

(P < 0.01) noted. The tests were carried out on themeans of the last 30 seasons of each scenario.Table 5gives the mean values and S.D. for each indicator re-tained for the comparison of the scenarios.

6.2. Results of scenario A: high rice-crop profitability

An homogenisation of the land use of the regionalland cover was observed (Fig. 5). Agricultural landsquickly increased to cover nearly 80% of the re-gion. The majority of the FAs adopted the “rice-crop

monoculture” strategy, whatever the size of the estate.Some hunting marshes were created on the bound-aries of the reserve and a large hunting unit, whoseHMA chose the “Elite” strategy. Hunting activity wasalso maintained on some flooded rice fields on theperiphery of the reserve. The duck resource increasedto reach more than 120,000 individuals, correspond-ing to a January wintering number of 30,000 birds.In the end, the system reached an equilibrium. Thedominant strategy of the HMAs was the “Passion”strategy. Hunting pressure was relatively low (the

Page 14: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

120 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

Table 5The main results of the simulations

Scenario A Scenario B Economic variability

Scenario ABABA Scenario BABAB

Duck resource 122121± 9364 54126± 3173 137923± 20296 105528± 33877Hunting harvest 5889± 2638 12385± 2075 23542± 3732 21501± 7439Harvest/ducks (%) 4.8± 2.2 22.9± 3.1 20.4± 4.8 17.1± 1.8Agricultural land 80± 0.2 46± 1.2 68± 2 52 ± 6Number of hunters 97± 5 463± 10 193± 25 334± 32Hunting leasing 207.18± 0.01 365.12± 0.03 342.55± 0.02 441.95± 0.03Individual fee 1710± 487 1019± 149 3259± 537 2285± 420Hunting turnover 133± 40 441± 44 316± 43 507± 70

Data are the mean± S.D., duck resource is in number of ducks; agricultural lands are in % of unprotected area; fee is inyear−1;hunting leasing and turnover are in 1000year−1.

harvest represented less than 10% of the resource).The cost of individual fees, being a function of the har-vest and the strategy, fluctuated between 1220and2744 , the average being around 1700. The num-

Fig. 5. Landscape configuration after 420 time steps (60 years) according to the scenario simulation runs. White: wheat crop, light grey:rice field, grey: fodder crop, dark grey: halophytic scrubland, dark: permanent-flooded marsh.

ber of hunters was low and relatively stable at around100. The hunting turnover was not very high, withan average of about 132,630for the whole of thehunting units. The total number of hunting leases was

Page 15: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126 121

stable and equivalent to an average leasing of 6860per year by FAs (not including the large hunting unit).

6.3. Results of scenario B: critical period for theagricultural market

An obvious development of natural lands makespossible for them to cover 55% of the region be-cause of the increase of hunting marshes (Fig. 5).After reaching a total of 140,000 individuals theduck resource fell sharply after 10 years to stabiliseat around 55,000 individuals (Fig. 5). We observedan homogenisation of the HMA’s strategies towardsthe “profitable” strategy, resulting in a substantialincrease in the number of hunters, which despite adecrease in individual fees to below 1525led to anincrease in turnover (Table 5). The average incomedrawn from the lease of hunting rights by the FAswas about 16,769 .

6.4. Results of scenario alternation

We were interested here primarily in the spatial as-pects (Fig. 5). We observed a hunting specialisation ofsome estates. Alternation of the economic scenarios,whatever the order (ABABA or BABAB), results in ahunting specialisation on the estates which had naturallands at the beginning of the simulation and/or werelocated near the reserve or the large hunting unit. Thespatial development of the marshes managed for hunt-ing observed is scenario B (i.e. the anarchistic spatialdiffusion starting from the natural land uses within theestates) is channelled because of agricultural devel-opment limits the expansion of hunting-related landuse. Thus, only estates close to the large hunting unitor reserve, or estates with spatial opportunities thatwere interesting at the start of the simulation, spe-cialised in hunting activity, while the agricultural es-tates, through their own spatial system, maintainedhunting on flooded rice fields. A functional structuringof the region was accomplished.

7. General discussion

At the end of each simulation run, a landscape wasdeveloped within the region as a result of the com-bined effects of local suitability, the irrigation net-

work, agents’ strategies, and distribution of the duckresource. The qualitative results were inline with theempirical and field study findings. Naturally, the anal-ysis was conditional on our functional rules, parame-ter assumptions, and initial conditions. Although themodel used here reflected the Camargue environment,it is clear that it raised new empirical and theoreticalquestions for further research.

7.1. Discussion of the results and their implicationsfrom a prospective point of view

The pattern of resource exploitation by the HMAswas a process in which areas of high carrying ca-pacity were exploited intensively at the beginningof each simulation. As a result of the first high ef-fort periods, with corresponding over-harvesting, thenumber of ducks in high-profitable areas, where theHMAs have a different strategy from the “Elite”strategy, dropped more rapidly than in the other cellswithin less-profitable hunting units. Thus, in the caseof scenario A, the difficulty in competing with theagricultural income of FAs did not permit spatial de-velopment of the hunting areas and the maintenanceof natural land uses within the majority of the estates.The maintenance of natural lands in the large huntingunit and the reserve generated positive externalitiesfrom a duck resource perspective. However, theseexternalities did not allow an increase in the leasingof the hunting rights of the estates because the leaseswere partially linked to other land uses than agricul-tural ones. Thus, the maintenance of the large huntingunit contributed to the maintenance of smaller hunt-ing units but not to their spatial development. Thesupply of hunting units, as sought by hunters, doesnot equal the demand. When the equilibrium wasreached, HMAs changed their strategy (“Passion”)and the duck resource was under-exploited.

In the case of scenario B, the amount of incomefrom hunting leasing was higher than the agriculturalincome of the FAs. It thus enabled spatial expansionof the hunting area, and increased the carrying capac-ity of the region as a whole. However, with the initialover-harvesting described before and the drop in ducknumbers in the hunting areas, it became more prof-itable for HMAs to move to other estates. This increasein the movement of HMAs generated a high level ofhunting leasing but the decrease in the harvest did not

Page 16: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

122 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

allow an increase in the fee. The HMAs thus adoptedthe “Profitable” strategy. This process contributed tomaintain the movement of HMAs and the intensiveexploitation of the resource, which generated a highlevel of disturbance and a region-exit of the winteringsubpopulation or its over-harvesting.

Thus, in the case of a high profitability of the ricecrop (scenario A), agricultural land quickly becamepredominant on the majority of the estates. The exis-tence of hunting marshes, with their high profitabilitycompared with other activities, is not enough to limitthe change of natural land uses into rice fields exceptin the best locations.

However, hunting activity whose profit is based onlimited access to the space and a high hunting harvestbecame the victim of its own profitability when thesystem adopted a dynamics of imitation of large hunt-ing units without the “Elite” strategy, and spread to allthe estates (scenario B). The critical period of agricul-tural market led the farmers with spatial opportunitiesto develop other profitable activities (i.e. rural lodg-ing, reed harvesting, bull grazing) in particular huntingleasing. From a spatial point of view, this phenomenonshowed new allocation of land use and an increase inthe number of activities on the same piece of land. Thiscaused an attenuation of the functional structuring ofthe region and an intensification of inter-relationships,i.e. through a spatial rearrangement of the region.Thus, if economic growth is favourable and the prof-itability of rice crop is weak, this conjunction wouldinvolve a deep modification of the region, which inreality would also result in the parcelling out of someestates and their sale. The situation is then favourableto tourism-related and hunting-related investments.

The process can be generalised by comparison withthe well known tragedy of free access, the tragedy ofthe Commons byHardin (1968). Hardin showed that infree access, i.e. in the absence of limitation and accesscontrol to a resource, a dynamics of dilapidation of theresource, over-exploitation, and over-investment, isset up when the resource concerned has a market. Thescarcity of official hunting leases in the Camargue canbe the origin—even in the case of a private estate—ofa possible over-exploitation of the resource becauseof the mobility of the hunting teams. The need forefficiency (price/harvest) led some hunting managersto move their investments according to the huntingharvest, exploiting firstly the high-profitable hunting

units, thereby supporting a process that could con-tribute to the deterioration of the duck resource. Thus,the economic incentives for the conservation of thehabitats of wild fauna will remain significant factorsin the results of environmental policies and economicdevelopment (Williams and Lathbury, 1996).

In this paper, we have examined the effects of twoactions on population dynamics (i.e. land-use changesand harvest). AlthoughPhong Chau (2000)has shownthat simple constant harvesting may render the dy-namic predictable, the complexity of the interactionssystem orients the managers towards an adaptive man-agement process.

7.2. Discussion of MAS integration of levels ofinteraction and conservation strategy implications

From a whole-system perspective, the MAS enabledus to integrate three levels of interactions:

• Level of interaction 1: between use⇔ spatial sys-tem of the estate.

• Level of interaction 2: between ecological dynamic(here the duck resource)⇔ hunting dynamic⇔agricultural dynamic.

• Level of interaction 3: between the spatial systemof the estate⇔ neighbourhood spatial system⇔natural constraints.

The maintenance or development of natural landswithin the estates depends on economic and naturalconstraints. From the resource distribution perspec-tive, our model highlights that the large hunting unit,associated with the reserve, appears to be a “source”of ducks depending on the carrying capacity of thearea on the one hand and on the hunting strategy onthe other. However, duck distribution being linkedto disturbance and flooding, a change in either watermanagement and/or hunting strategy (from “Elite”to “Profitable”) can change the hunting unit froma “source” to a “sink”. As a result of this change,over-harvesting or over-disturbance can occur and theduck resource could decrease throughout the entiresystem. This illustration has useful implications forenvironmental planning, and choosing a location fora nature reserve.

In the case of scenario A, we noted for examplethat the removal of a large hunting unit which heav-ily influences level of interaction 2 can destabilise

Page 17: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126 123

not only nearby geographical agents, but also all ofthe ecological and hunting dynamics. The concreteimplications are, in a configuration of this type, thatthe purchase of an estate for conservation purposesshould preferably be made in large hunting units inorder to limit the degradation of the system if these areput up for sale and suitable for agriculture. However,the spatial configuration results generated some ques-tions about the surface area of protected areas, andthe existence of threshold effects which, once a givenlevel was exceeded, would authorise a developmentof the duck resource in the Camargue. This could bewhy the development of some nature reserves overthe last decade never resulted in an increase in ducknumbers on a regional scale (Tamisier and Dehorter,1999; Mathevet and Tamisier, 2002).

In the case of scenario B, all the levels of interac-tions have an importance, although level 3 seems to bepredominant. Development is thus more diffuse, de-pending on the relief and the heterogeneity of naturalland as well as the nearby spatial system. However,when the interactions of level 2 weaken, the level ofinteraction 1 tends to increase its influence on the sys-tem. In such a configuration, the creation of a naturereserve on an estate will obviously have some impacton the area nearby, but the interactions of level 2 willbe reinforced and limit the positive impact of the cre-ation of the new protected area on the duck resource.In the case of a variability in agricultural economics,the alternation of the importance of each level of in-teractions will be favourable to the development ofthe resource and partly favourable to the maintenanceof natural lands. The effects of spatial neighbourhoodand natural constraints seem to dominate. The creationof a natural reserve should have a positive impact onthe resource and thus release positive externalities interms of hunting for all the estates. The reserve wouldthen contribute to the development of hunting at a timeof agricultural decline as well as to the specialisationof the weaker estates.

7.3. Possibilities for further modifications ofGEMACE and research implications

Many models have been developed for assessmentof wildlife populations but none of these models in-tegrated environmental fluctuations and economicsparameters. Environmental variation causes both

short- and long-term fluctuations in harvest.Jensen(2002)has shown that the sources of the year-to-yearfluctuations of the Mississippi flyway mallard duckwas not a result of harvesting but rather was a resultof environmental variations. Our model used avail-able harvest and hunter data, and our outcomes areconsistent with these results. However,Kokko andLindström (1998)have studied the consequences ofharvest timing to equilibrium population sizes underdensity-dependent mortality. They showed that timingof harvesting has a strong impact on the sustainabil-ity of a harvesting quota. Thus, we might developother simulations to explore the consequences ofharvest timing to equilibrium population sizes undermortality-dependent mortality to better estimate thereduction in population size caused by hunting.

This paper made a preliminary study of qualitativeeffects of different agricultural and hunting policiesin a realistic environment. The results obtained hereshould be useful for studying the effects of environ-mental planning and harvest policies (Costa et al.,2000), but it is advantageous for the model to integratemuch more fully:

• Strong natural constraints and qualitative indicatorson natural lands.

• Landscape representation through the use of spatialobjects aggregation (which exists as predefined en-tities in CORMAS) for a better integration of topol-ogy and the development of several useful viewpoints of spatial aggregates over layers of scale.

• The social representations of the environment by thevarious agents and values relating to bull breeding.The latter has proven to be one of the essential ele-ments of the Camargue system for the maintenanceof certain ecological units.

• Recent biological findings, to develop an age- andsex-structured population model of the two targetedgroups of game species (herbivorous and granivo-rous).

Future research efforts planned for this model in-clude: (1) examining the effect of the creation of anew reserve from a land use and economic perspectivein order to have a better understanding of their spa-tial and socio-economic impacts on the whole system;(2) exploring consequences of the dismemberment ofestates (AAU/natural lands), on the conservation ofhabitats; and (3) analysing the impact of a reduction

Page 18: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

124 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

in the hunting period on disturbances and the manage-ment of habitats, and therefore on the duck resource.

The development of our model went by a multi-fieldapproach in which the question of knowledge inte-gration was a fundamental one. The most difficultquestions to answer were those relating to the lev-els of complexity of the interactions. Our approachcombined spatial and non-spatial dynamic processesand required addressing of the scale-transfer question(Franc and Sanders, 1998). We represented objects andagents at the micro level, which corresponded to theindividuals observed in reality. The knowledge whichenabled us to implement them in the model, however,was at a more intermediate level between the microand macro level, a level which corresponded to thelevel of the questioning of the individual (Bousquetet al., 1994). The MAS enabled a virtual laboratoryto be created in which we reproduced the real worldaccording to the points of view of various scientificdisciplines and where these interact. However, due tothis confrontation of points of view within the model,the validation phase was a delicate stage. The resultsof our scenario runs showed that plausible patternsemerge. The qualitative and quantitative parameters ofeach agent or object were based on documented pat-terns or explicit assumptions. We calibrated the pa-rameters, the values of the attributes, in order to obtainindividual characteristics and behaviours close to re-ality. The comparison of the results of simulation withthe data collected in the real world made it possibleto determine “structural similarities”, and it was thisco-variation that ensured the adequacy of the modelas a reliable representation of the modelled system(Gilbert, 1998). Validation is the relevance of the sim-ulation results compared with the initial questions. Theresults were also discussed with several experts (ecol-ogist, economist, manager,. . . ) to evaluate the pro-cesses, the assumptions and, possibly, to re-examinesome methods or parameter values.

The various stages of multi-agent modelling con-tributed to enriching the theoretical field and clarifyinginteractions between disciplinary knowledge on theenvironmental components. There is a growing inter-est in developing dynamic models within GIS (Ball,1994), and recent progress in the coupling of MAS andGIS tools allows us to foresee a use for modelling in re-gional planning and as a decision-making aid (Lardonet al., 1998; Zunga et al., 1998). Coupling MAS and

GIS allows the representation of scale effects and fa-cilitates the transferral of knowledge between levels(Zunga et al., 1998). In the same way, the formalisationof the interactions between the processes makes it pos-sible to define rules which, through role-playing game,can be used in a patrimonial approach or for the studyof protagonists during real-life negotiation (Barreteauet al., 2001; Bousquet et al., 1999; Lynam et al., 2002;Benoit et al., 1998). It is then a question of developingthe model to a simulator from which various actors canchoose parameters or behaviours, and thus test howindividual or social choices can be reflected on a largerscale (Gilbert, 1998; Barreteau and Bousquet, 2000).

8. Conclusion

Several implications can be drawn from this mod-elling application regarding the viability of the spatialconservation management alternative. First of all,in the development of the Camargue, and from aprospective point of view, two possibilities emerge.The first case is that of very high profitability for ricefarming, in which we would observe a developmentof former rice fields only partly attenuated by huntingactivity. Locally, some residual natural lands shouldbe maintained due to the high income from hunting.The second case corresponds to what we currentlyobserve in the Camargue. The low profitability of therice crop involves the development of dry crops andhunting marshes on former rice fields, while hunt-ing installations are also appearing on natural lands.In the long run, the spatial generalisation of hunt-ing activity is likely to involve a disaffection of thehunters for the Camargue and a fall, first in the priceof hunting fees, then in the lease income, associatedwith a considerable decrease in the duck resource ifthis does not adapt to heavy hunting pressure (theincreasingly late departure of ducks towards the feed-ing grounds mentioned by many hunters could limitthe range of this fall but not of the harvest). In thiscase, tourism-related investment is likely to becomethe region’s only speculative value. This developmentcould then perturb the region, depending on land-useplanning regulations and the capacity of the naturallands used for hunting and the reserves to guaranteethe image of the Camargue as a wild and picturesqueplace.

Page 19: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126 125

From a theoretical point of view, our model suggeststhat understanding more about spatial configurationis critical to any regional planning and policies analy-sis, in particular in a heterogeneous spatial system ofnatural resource exploitation. Moreover, our resultsindicated that though MAS allowed the integration ofseveral levels of interactions, we were not able to fullyunderstand the ecological dynamics operating in ahunting harvested system without a larger understand-ing of the spatial implications of the managementpractices, attitudes and values of the protagonistsinvolved. By integrating this knowledge, the modelmight prove helpful in attempts to answer questionssuch as how agro-environmental policy or reservecreation choices based on socio-economic-spatial het-erogeneity might be made, and the extent to whichecological socio-economic relationships will affectnew environmental policies.

Finally, the multi-agent approach enables the inte-gration of multiple items of scientific and empiricalknowledge; it also underlines the lack of knowl-edge, and the relevance of the use of top-down andbottom-up approaches. Thus the multi-agent approachenables the development of multi-field studies whosetheoretical and applied implications feed the renewalof disciplinary points of view on the question ofrelationships between humans and nature.

Acknowledgements

We thank the landowners of the estates in which thefieldwork was carried out. We thank the hunting man-agers for their collaboration during this study. Thanksto Alain Tamisier, André Mauchamp, Jean-LaurentLucchesi, François Mesléard, Robert Lifran for theirconstructive discussions.

References

Abott, C.A., Berry, M.W., 1995. Computational models of whitetailed dear in the Florida everglades. UTK-CS.http://www.es.utk.edu/∼berry/deer/deer.html.

Anderies, J.M., 2000. On modeling human behavior andinstitutions in simple ecological economic systems. Ecol. Econ.35, 393–412.

Ball, G.L., 1994. Ecosystem modeling with GIS. Environ. Manage.18 (3), 345–349.

Barbier, J.M., Mouret, J.C., 1992. Le riz et la Camargue. Histoireet Recherche. INRA Mensuel 64/65, 39–51.

Barreteau, O., Bousquet, F., 2000. SHADOC: a multi-agent modelto tackle viability of irrigated systems. Ann. Operations Res.94, 139–162.

Barreteau, O., Bousquet, F., Attonaty, J.-M., 2001. Role-playinggames for opening the black box of multi-agent systems: methodand lessons of its application to Senegal River Valley irrigatedsystems. J. Artif. Societies Soc. Simulation 4 (2)http://www.soc.surrey.ac.uk/JASSS/4/2/5.html.

Beedell, J., Rehman, T., 2000. Using social-psychology models tounderstand farmers’ conservation behaviour. J. Rural Stud. 16,117–127.

Benoit, M., Chicoisne, G., Deffontaines, J.-P., Herve, D., Lardon,S., Le Ber, F., Mullon, C., Papy, F., Souchere, V., Thinon,P., Tichit, M., Treuil, J.-P., 1998. Coordonner des choix decultures sous contraintes environnementales: des jeux de rolesaux modèles multi-agents. In: Ferrand, N. (Ed.), Actes duColloque SMAGET. Clermont-Ferrand, pp. 279–292.

Blandin, P., Lamotte, M., 1988. Recherche d’une entité écologiquecorrespondant à l’étude des paysages: la notion d’écocomplexe.Bulletin d’Ecologie 19 (4), 547–555.

Bossel, H., 2000. Policy assessment and simulation of actororientation for sustainable development. Ecol. Econ. 35, 337–355.

Bousquet, F., Cambier, C., Mullon, C., Morand, P., Quensiere, J.,Pavé, A., 1993. Simulating the interaction between a societyand a renewable resource. J. Biol. Syst. 1, 199–213.

Bousquet, F., Cambier, C., Morand, P., 1994. Distributed artificialintelligence and object-oriented modelling of a fishery. Math.Comput. Model. 20 (8), 97–107.

Bousquet, F., Bakam, I., Proton, H., Lepage, C., 1998. Cormas:common-pool resources and multi-agent systems. Lect. NotesArtif. Intelligence 1416, 826–837.

Bousquet, F., Barreteau, O., Le Page, C., Mullon, C., Weber,J., 1999. An environmental modelling approach. The useof multi-agents simulations. In: Blasco, F., Weill, A. (Eds.),Advances in Environmental and Ecological Modelling. Elsevier,Paris, pp. 113–122.

Bousquet, F., LePage, C., Bakam, I., Takforyan, A., 2001.Multi-agent simulations of hunting wild meat in a village ineastern Cameroon. Ecol. Model. 138, 331–346.

Bura, S., Guérin-Pace, F., Mathian, H., Pumain, D., Sanders, L.,1993. Multi-agent systems and the dynamics of a settlementsystem. In: Artificial Societies. UCL Press, Siena.

Chauvelon, P., 1998. A wetland managed for agriculture asan interface between the Rhone river and the Vaccarèslagoon (Camargue, France): transfers of water and nutrients.Hydrobiologia 373/374, 181–191.

Costa, M.I.S., Kaszkurewicz, E., Bhaya, A., Hsu, L., 2000.Achieving global convergence to an equilibrium population inpredator–prey systems by the use of a discontinuous harvestingpolicy. Ecol. Model. 128, 89–99.

Costanza, R., Daly, H.E., Bartholomey, J.A., 1991. Goals, agenda,and policy recommendations for ecological economics. In:Costanza, R. (Ed.), Ecological Economics: The Science andManagement of Sustainability. Columbia University Press, NewYork, pp. 1–20.

Page 20: Agent-based simulations of interactions between duck population, farming decisions and leasing of hunting rights in the Camargue (Southern France)

126 R. Mathevet et al. / Ecological Modelling 165 (2003) 107–126

Dean, J., Gumerman, G., Epstein, J., Axtell, R., Swedlund,A., Parker, M., 2000. Understanding Anasazi culture changethrough agent-based modeling. In: Kohler, T., Gumerman, G.(Eds.), Dynamics in Human and Primate Societies. OxfordUniversity Press, New York, pp. 179–206.

Dehorter, O., Tamisier, A., 1996. Wetland habitat characteristicsfor waterfowl wintering in Camargue, Southern France:implications for conservation. Rev. Ecol. (Terre Vie) 5, 161–172.

Ferber, J., 1995. Les systèmes multi-agents. Vers une intelligencescollectives, Paris, Inter Editions.

Ferrand, N., Deffuant, G., 1999. Aide à la gestion des conflits enaménagement du territoire. Approche multi-agents et modèlescognitifs des acteurs, Cemagref Lisc, Clermont-Ferrand.

Franc, A., Sanders, L., 1998. Modèles et systèmes multi-agents enécologie et en géographie; état de l’art et comparaison avec lesapproches classiques. In: Ferrand, N. (Ed.), Actes du ColloqueSMAGET. Clermont-Ferrand, pp. 17–34.

Gilbert, N., 1998. The simulation of social processes. In: Ferrand,N. (Ed.), Actes du Colloque SMAGET. Clermont-Ferrand,pp. 121–135.

Ginot, V., Le Page, C., Souissi, S., 2002. A multi-agentsarchitecture to enhance end-user individual-based modelling.Ecol. Model. 157, 23–41.

Grimm, V., 1999. Ten years of individual-based modelling inecology: what we have learned and what could we learn in thefuture? Ecol. Model. 115, 129–148.

Grossmann, W.D., 1994. Socio-economic ecological models:criteria for evaluation of state-of-the-art models shown on fourcase studies. Ecol. Model. 75/76, 21–36.

Hardin, G., 1968. The tragedy of the commons. Science 162,1243–1248.

Heath, M.F., Evans, M.I., 2000. Important bird areas in Europe:priority sites for conservation. In: Southern Europe, vol. 2.Birdlife International Conservation Series No. 8, Cambridge,UK.

Jensen, A.L., 2002. Analysis of harvest and effort data for wildpopulations in fluctuating environments. Ecol. Model. 157, 43–49.

Jodelet, D., 1989. Les Représentations Sociales. PUF, Paris.Kohler, T., Kresl, J., Van West, C., Carr, E., Wilhusen, R.,

2000. Be there and then: a modeling approach to settlementdeterminants and spatial efficiency among late ancestral pueblopopulations of the Mesa Verde region, US Southwest. In:Kohler, T., Gumerman, G. (Eds.), Dynamics in Human andPrimate Societies. Oxford University Press, New York, pp.145–178.

Kokko, H., Lindström, J., 1998. Seasonal density dependence,timing of mortality, and sustainable harvesting. Ecol. Model.110, 293–304.

Lardon, S., Baron, C., Bommel, P., Bousquet, F., Le Page,C., Lifran, R., Monestiez, P., Reitz, P., 1998. Modéliser lesconfigurations et les stratégies spatiales dans un système multi-agents pour la maıtrise de dynamiques d’embroussaillement.In: Ferrand, N. (Ed.), Actes du Colloque SMAGET. Clermont-Ferrand, pp. 169–185.

Lynam, T., Bousquet, F., D’Aquino, P., Barreteau, O., LePage, C., Chinembiri, F., Mombeshora, B., 2002. Adaptingscience to adaptive managers. Conservation Ecol. 5 (2), 24http://www.consecol.org/vol5/iss2/art24.

Madsen, J., Fox, A.D., 1995. Impacts of hunting disturbance onwaterbirds—a review. Wildl. Biol. 1–4, 193–207.

Mathevet, R., 2000. Les usages des zones humides camarguaises:enjeux et dynamique des interactions environnement/usagers/territoire. Ph.D. thesis, Université J. Moulin Lyon 3.

Mathevet, R., Mesléard, F., 2002. The origins and functioning ofthe private wildfowling lease system in a major Mediterraneanwetland: the Camargue (Rhone River delta, Southern France).In: Land Use Policy 19 (4) 277–286.

Mathevet, R., Tamisier, A., 2002. Creation of a nature reserve, itseffects on hunting management and waterfowl distribution inthe Camargue (Southern France). Biodivers. Conservation 11,209–519.

Mathevet, R., Mouret, J.-C., Mesléard, F., 2002. Du mondial aulocal: originalité et dépendances de la riziculture camarguaise.Faire-Savoirs 2, 49–56.

McDonnel, M., Pickett, S.T.A., 1993. Humans as Componentsof Ecosystems. The Ecology of Subtle Human Effects andPopulated Areas. Springer-Verlag, New York.

Odum, H., 1982. Systems Ecology. Wiley, New York.Pirot, J.Y., 1982. Exploitation alimentaire des zones humides de

Camargue par cinq espèces de canards de surface en hivernageet en transit. CNRS Centre d’Ecologie de Camargue, ONC.Rapport d’activité no. 4, Arles.

Pirot, J.Y., Chessel, D., Tamisier, A., 1984. Exploitation alimentairedes zones humides de Camargue par cinq espèces decanards de surface en hivernage et en transit. Modélisationspatio-temporelle. Rev. Ecol. (Terre et Vie) 39, 167–192.

Phong Chau, N., 2000. Destabilising effect of periodic harvest onpopulation dynamics. Ecol. Model. 127, 1–9.

Tamisier, A., Dehorter, O., 1999. Camargue, canards et foulques.Fonctionnement et devenir d’un prestigieux quartier d’hiver.CoGard, Nımes.

Tamisier, A., Grillas, P., 1994. A review of habitat changes inthe Camargue: an assessment of the effect of the loss ofbiological diversity on the wintering waterfowl community.Biol. Conservation 70, 39–47.

Turner, R.K., 1991. Economics and wetland management. Ambio20 (2), 59–63.

Turner, M.G., Wu, Y., Romme, W.H., Wallace, L.L., 1993.A landscape simulation model of winter foraging by largeungulates. Ecol. Model. 69, 163–184.

Ulanowicz, R.E., 1997. Ecology: The Ascendant Perspective.Columbia University Press, New York.

Williams, C.E., Lathbury, M.E., 1996. Economic incentives forhabitat conservation to the Inland Pacific Northwest. Wildl.Soc. Bull. 24 (2), 187–191.

Zunga, Q., Vagnini, A., Le Page, C., Toure, I., Lieurain, E.,Bousquet, F., 1998. Coupler systèmes d’information géogra-phique et systèmes multi-agents pour modéliser les dynamiquesde transformation des paysages. In: Ferrand, N. (Ed.), Actesdu Colloque SMAGET. Clermont-Ferrand, pp. 193–206.