Land market mechanisms for preservation of space for coastal ecosystems: An agent-based analysis

lable at ScienceDirect

Environmental Modelling & Software 26 (2011) 179e190

Contents lists avai

Environmental Modelling & Software

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

Land market mechanisms for preservation of space for coastal ecosystems:An agent-based analysisq

Tatiana Filatova a,*, Alexey Voinov b, Anne van der Veen b

aCentre for Studies in Technology and Sustainable Development, University of Twente, 7500 AE Enschede, The Netherlandsb International Institute for Geo-Information Science and Earth Observation, University of Twente, The Netherlands

a r t i c l e i n f o

Article history:Received 29 April 2010Received in revised form4 August 2010Accepted 6 August 2010Available online 17 September 2010

Keywords:Agent-based modellingEcosystem servicesCoastal zoneLand market mechanisms

q The earlier version of this paper was presented atwas awarded ‘Student paper commendation’.* Corresponding author. Tel.: þ31 53 489 3530; fax

E-mail addresses: [email protected] (T. [email protected] (A. van der Veen).

1364-8152/$ e see front matter � 2010 Elsevier Ltd.doi:10.1016/j.envsoft.2010.08.001

a b s t r a c t

This paper presents an agent-based model of a land market, which is used to explore the effects of landtaxes on the land use in a coastal zone. The model simulates the emergence of land prices and urban landpatterns from bottom-up via interactions of individual agents in a land market. A series of modelexperiments helps visualize and explore how economic incentives in a land market may influence thespatial distribution of land prices and urban developments, either leaving space for coastal ecosystems ornot. We demonstrate that economic incentives do affect urban form and pattern, land prices and welfaremeasures. However, they may not always be sufficient to reduce the pressure on coastal ecosystems. Ourresults show that preservation of ecosystems may involve difficult trade-offs between economic andecological priorities, as well as between healthy ecosystems and social equity. We also show howconventional economic modelling based on a representative agent, which is usually employed by policymakers, overestimates both environmental benefits and economic costs associated with the tax meant topreserve coastal ecosystems.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Coastal zones (CZ) are important from both ecological and socio-economic points of view (Martínez et al., 2007). These are some ofthe most productive areas on our planet that provide manyecosystem services such as erosion control and sediment retention,habitat for species, food production, recreation and others. Theseareas are especially rich in biodiversity and have one of the highestvalues for ecosystem services per hectare of area (Costanza et al.,1997). CZs require a delicate balance between human systemsand ecosystem functions provided by the interactions of land andsea. This makes protection of CZs in their pristine form an impor-tant component of environmental management.

At the same time, CZs are also a very lucrative place for devel-opment. They are one of the most densely populated areas wheretwo thirds of world population reside (Costanza et al., 1999). Inparticular, in the Netherlands 70% of the Gross National Producttoday is generated in the CZ (Veraart et al., 2007). CZs have been

the iEMSs congress 2008 and

: þ31 53 489 4850.a), [email protected] (A. Voinov),

All rights reserved.

historically developed due to access to marine and river trans-portation. Further developments occurred in the proximity tohistoric cities causing even more construction in areas where vitalcoastal ecosystem services needed space as well. Waterfront prop-erties are known to be several times more expensive than similarproperties in land. People are willing to pay high prices for waterview and water access (Pompe and Rinehart, 1999; Spalatro andProvencher, 2001). The economic forces come into play reacting tohigh land prices by further attracting urban developers and puttingpressure on spatial planners to allow urbanization in coastal zones.

In addition to the deterioration of coastal ecosystems, expansionof urban developments increases the potential damage fromflooding or erosion. According to IPCC damage from natural disas-ters has rapidly increased over the past decades mainly due to thegrowth of capital in flood-prone areas (Nicholls et al., 2007).Consequently, the replacement of natural coastal ecosystems byresidential developments further increases the risk of flooding(Costanza et al., 2008). The combined effects of human and physical(or climatic) pressures together lead to an effect known as ‘coastalsqueeze’. Coastal squeeze occurs in the coastal margin, which issqueezed between fixed landward boundary and the rising sealevel shrinking the areas available for natural coastal processes andecosystem dynamics to take place (Schleupner, 2008). Pressure onCZ induced by economic activities causes the disruption of coastal

Fig. 1. Conceptual scheme of the land market.

1 Filatova et al. (2009a) provide a detailed review of spatial ABMs with respect toland markets.

2 Fig. 1 is a simplified version of a more general figure representing a compre-hensive ABM market model described in Parker and Filatova (2008).

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190180

ecosystem functions including modification of shoreline, reductionof habitat carrying capacity and, as a result, diminishing recrea-tional value of coasts (Martínez et al., 2007). Thus, withstanding thethreats of coastal squeeze is largely a matter of preserving unde-veloped space along a coastline.

If we are to protect coastal ecosystems we need to find ways tocreate disincentives for people to locate near the coastline. Accordingto Costanza et al. (1999)marketmechanismsmight bemore efficientthan traditional command and control regulation in achieving envi-ronmental goals, if the incentives are put in place and enforced atrelatively low cost. As recent analysis shows (Bagstad et al., 2007),markets are very promising in reducing the risk of natural coastalhazards while preserving natural capital. Market mechanisms appli-cable in CZ may include insurances, taxes and subsidies for coastalland. In fact, market mechanisms are more efficient if they differen-tiate land in its proximity to the coast (Pompe and Rinehart, 1999).This paper aims to explore whether market mechanisms, such as taxon land in CZ, can be effective in preserving coastal ecosystems fromincreasing pressures of urban developments. We formulate thefollowing researchquestions: 1)Domarketmechanisms, suchas landtax, indeed preserve space for coastal ecosystems to function? 2)What can be the side-effects associated with these measures? Tovisualizeandquantify theeffectsof the landtaxmechanismonspatialdevelopments in CZ a spatially-explicit modelling tool that incorpo-rates a land market is needed.

Different tools that couple human and environmental systemsexist. Models applied to various aspects of water managementinclude system dynamics (Elshorbagy and Ormsbee, 2006), cellularautomata (de Kok et al., 2001; Marshall and Randhir, 2008), ora combination of both (Voinov et al., 1999; Fitz et al., 2003). Othersinclude spatial statistical models (Bockstael, 1996; Irwin andBockstael, 2004; Wu et al., 2004), equilibrium economic models(Wu and Irwin, 2008), just to name a few. They use either one ortwo way coupling between economic and natural systems (Parkeret al., 2008), and they differ in incorporation of economicprocesses, which are essential to explore the effects of marketmechanisms on ecosystems. The majority of integrated modelsfocus mainly on representing processes in natural systems.Economic part is usually modelled at the aggregated level, i.e. aseconomic indices and macro socio-economic variables (changes inGDP, population growth, etc.). This might be sufficient for certainpurposes. However, when we need to understand economicprocesses in markets, i.e. interactions among economic agents thatstand behind an aggregated index, we need a different approach.Modelling of economic processes in CZ involves incorporatinghuman behaviour in the spatial landscape. One of the most suitabletechniques for this is agent-based modelling (ABM) (Janssen, 2002;Bolte et al., 2007; Brown et al., 2008).

ABM allows to represent interacting economic agents overa spatial landscape (Parker et al., 2002), and iswidely used for spatialanalysis (Barreteau et al., 2001; Berger, 2001; Parker et al., 2003;

Bousquet and Le Page, 2004; Brown et al., 2004; Gross et al., 2006;Happe et al., 2006; Matthews et al., 2007; Monticino et al., 2007;Moreno et al., 2007; Crooks et al., 2008; Gotts and Polhill, 2009;Lagabrielle et al., 2010; Naivinit et al., 2010). These land usemodels are good inmodelling the spatial landscape, the biophysicalprocesses and a variety of agent behaviours. However, their repre-sentation of a land market is limited1. To study the effects ofmarketmechanisms on land use change and preservation of space forecosystem services, themarket has to be a part of the model.

In this paper we present a simple ABM that integratesa heterogeneous landscape and explicit individual location choicesin a land market to explore the influence of market instruments onthe preservation of open space in CZ. Since the undeveloped spaceprovides ecosystem functions in CZ, the extent of urban conversionindirectly indicates how much pressure/relief is put on coastalecosystems. With the help of ABM one can track how policymeasures enter individual land use decisions, what aggregatedresults (i.e. land patterns and prices, undeveloped space in CZ, andassociated economic costs) emerge from many individual interac-tions, what types of behaviour lead to them and how policies canaffect those behaviours (Parker and Filatova, 2008).

The paper is organised as follows. First, we present the spatially-explicit land market model applied to a coastal town. Then, wedescribe the experiment setup and discuss the results underdifferent assumption about economic agents. Finally, we concludewith a discussion about the use of market mechanisms forecosystems’ preservation and their utility in understanding theconsequences of policies.

2. The model

2.1. Conceptual scheme

We try to keep the model simple. At this stage we do not modelbiophysical processes in space: there are many integrated modelsthat do this in a quite advanced way (Voinov et al., 1999; Barreteauet al., 2001; Berger, 2001; Fitz et al., 2003; Elshorbagy and Ormsbee,2006; Marshall and Randhir, 2008). It is usually the fully-modelledspatially-explicit land market part of a model that is absent. Thus,we focus on individual market behaviour. The main processes wewant to capture is how different attributes of land and economiccharacteristics of agents are capitalised in land prices, and affect theformation of land patterns, and what effect taxes have on thepreservation of space for coastal ecosystems. The conceptualscheme of a market is presented in Fig. 12.

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190 181

The Agent-based Land MArket model (ALMA) simulates theemergence of urban land patterns and land prices as a result ofmicro-scale interactions betweenbuyers and sellers of land (Filatovaet al., 2009a). ALMA-C is the application of this model to a coastaltown. ALMA-C borrows much from the analytical monocentricurbanmodel (Alonso,1964) and its application to a coastal city (Wu,2001). Consequently, many of the assumptions of ALMA-C are inagreement with the analytical model, e.g. trade-offs between travelcosts to the central business district (CBD) and land price, or the factthat the highest bidder gets the land. However, ABM methodologyallows us to go beyond some of the restrictive assumptions of theconventional economic model as discussed below.

The model was programmed using NetLogo (Wilensky, 1999).The core of the ALMA model was previously described (Filatovaet al., 2009a,b) using UML diagrams and MR.POTATOHEAD frame-work standard for presenting agent-based land use models (Parker,Brown et al., 2008). This implementation of the model we describeusing the ODD protocol, which is common for ecological individual-based models (Grimm et al., 2006).

2.2. The ALMA-C model description

2.2.1. PurposeThe purpose of the model is to explore, quantify and visualize in

a 2D landscape the effects of market mechanisms, such as landtaxes, on land use change and preservation of space for ecosystemservices in CZ. ALMA-C allows direct modelling of interactions ofmany heterogeneous agents in a land market over a heterogeneous

Table 1State variables of the ALMA-C model.a

Variable name Brief description

I. Spatial landscape (grid of cells)Coordinates X and Y coordinates of the cellPatchID A unique ID of a cell (spatial good)nature? A Boolean variable defining whether a cell iproximityAmen Normalized value of proximity to natural arproximityCBD Normalized value of proximity to the CBD (cmarketPrice An attribute of a cell where first seller’s ask

then market transaction price are recordedProperty tax A value of the environmental land tax for atr Tax ratek1 Coefficient in the hyperbolic tax function, Eqk2 Coefficient in the hyperbolic tax function, Eq

II. Economic agents (traders) e parent classBuyer?/seller?/traded? Three states in which a trader can be each tAlpha Preferences for coastal amenitiesfixedAlpha Boolean, determines whether agents have h

preferences for environmental amenities an

II.1 Buyer e subclass of traders classBudget Disposable budgetHousing budget Budget net of travel costs from the specifictcu Transport costs per unit of distance in EquafixedBudget Boolean, determines whether agents have hUtility Agent’s utility for a specific spatial good (caWTP Buyer’s willingness to pay (WTP) for a specib A constant in Equation (4) that serves as a p

good (i.e. prices for all other goods except hBid Buyer’s bid price for a specific spatial goodPricing strategy Buyer’s strategy to determine her bid price

II.2 Seller e subclass of traders classReservation price Seller’s reservation price for undeveloped la

willingness to accept e WTA). Assumed to bAsk An asking price the seller assigns to the cellPricing strategy Seller’s strategy to determine his asking pric

a) for variable ‘Coordinates’ ***means one of many possible values of coordinates x and ydenotes one of the three states.

spatial landscape simplifying further coupling of economic modelswith ecological ones. As other ABMs of markets it helps to under-stand how aggregated patterns and economic indices result frommany individual interactions of economic agents (Tesfatsion, 2006).

2.2.2. State variables and scalesThe agents and the spatial environment that comprise ALMA-C

are presented in Table 1.Spatial scales: the model runs at the scale of a town. The whole

landscape is divided into a grid of equal cells, each occupied by oneeconomic agent. In the theoretical model, which is ALMA now,there is no specific resolution defined. Temporal scales: during onetime step many buyers and sellers interact in a land market andseveral market transactions happen. Their number depends on theamount of vacant space, the number of interested buyers and theirwillingness to pay (WTP), the competition among traders and, inthis case, the land tax.

2.2.3. Process overview and schedulingThe main process in ALMA is land trading and allocation of

households in a town. Each time step the trade process consists ofseveral phases: listing of vacant spatial goods in a market by sellers,search for the best location under budget constraint by buyers,formation and submission of bids by buyers to sellers, evaluation ofreceived bids by sellers, transaction and registration of trade(Fig. 2). These steps are repeated until there are no economicallyfeasible transactions possible, i.e. until market equilibrium isreached.

Value

******

s an open space or coastline 0/1ea Equation (1)entral business district) Equation (1)ing price and Endogenously

determinedspecific cell Equation (5)

˛½0%;2%�uation (5) 50uation (5) 31

ime step ***0.6

omogeneous or heterogeneousd proximity to the CBD

1

˛½800; 850�cell to the CBD Equation (3)tion (3) 1omogeneous or heterogeneous incomes 0/1n be logarithmic or Cobb-Douglass) Equation (2)fic spatial good Equation (4)roxy for the price of a compositeousing)

70

Set of strategies‘Pure WTP’, i.e. bidprice is equal to WTP

nd (i.e. hise equal to the price of agricultural land.

200

he owns Set of strategiese Asking price is equal

to reservation price

; b) for variable ‘PatchID’ *** denotes an unique ID c) for ‘Buyer?/seller?/traded?’ ***

Fig. 2. Sequence of events.

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190182

2.2.4. Design concepts2.2.4.1. Emergence. Land patterns and land prices emerge asa result of bilateral trades between buyers and sellers with diverseincomes and specific preferences in a heterogeneous landscape.These aggregated outcomes are important for the evaluation of theeffects of environmental land tax policy on preservation of space forecosystem services as well as on the welfare of agents. A set ofmacro-scale measures presented in Table 2 (Section 3) is used tomonitor these emergent outcomes.

2.2.4.2. Adaptation. In the simulation presented in this paperadaptation is not modelled explicitly. In other simulations,however, the pricing strategy of agents adapts to the situation ina land market and previous trades (Filatova et al., 2009a). Otheradaptive price expectation formation mechanisms are presentedelsewhere (Parker and Filatova, 2008).

2.2.4.3. Fitness/objectives. At the individual level buyers seeka location that maximises their utility under budget constraints.Utility is a formalized way to present individual preferences forlocations. These preferences can be homogeneous or drawn fromsome distribution. Sellers aim to maximize their surpluses byselling their properties to the highest bidder. Bids should be abovesellers’ reservation prices. At the level of the whole town there isanother objective in addition to maximizing economic welfare:preservation of space along the coastline which is most importantfor ecosystems.

2.2.4.4. Prediction. Agents aremyopic andmake decisions based onthe current state of the landscape and the market.

2.2.4.5. Sensing. Information about spatial characteristics of thelandscape and tax rates is public. However, buyers are assumed to

be boundedly rational. Bounded rationality captures the fact that anindividual is not able to foresee all the consequences of his choice,take into account all the factors, and has limited computing abilities(Simon, 1997). Searching for a house in reality is very costly (time-wise and money-wise). Not all properties are listed in onlinedatabases, choice and viewing of listed properties is time-consuming, and, moreover, often it is the realtor who subsampleshouses for a buyer according to their own goal functions maxi-mizing their own profit. Since a global optimum is not likely to beidentified in real world housing markets, buyers do not search forthe maximum throughout the whole landscape, but rather finda local maximum among a set of randomly chosen cells. Sellersasking prices are public information but a buyer is not aware of thebids other buyers make.

2.2.4.6. Interaction. Agents are involved in market interactionswith each other: a) land prices and land patterns are determinedvia direct interactions of buyers and sellers; b) buyers competewitheach other submitting bids to the same seller. Local governmentagent is not modelled explicitly but the tax rate is assumed to bedetermined by local government. So, there are obvious feedbacksbetween the tax rate, amount of space preserved and economicwelfare.

2.2.4.7. Stochasticity. The random seed affects the sequence ofactivation of the agents during the simulation (i.e who tradeswith whom) and the distribution of income endowment. We runeach experiment 30 times to check the robustness of the resultsagainst random effects. Parametric t-tests were used to confirmthe statistical significance of differences in the outcomesbetween model simulations. In addition, a user can fix therandom seed number and precisely repeat a certain simulation ifnecessary.

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190 183

2.2.4.8. Observation. Spatial distribution of land uses and landprices is recorded in the format of.png image files (procedure‘export-rent’). In addition, all the transaction data is recorded as.csvcomma-delimited files for further statistical analysis. This dataincludes all spatial characteristics of the traded cell, its transaction,bid and asking price, as well as all agents’ characteristics (budgetand preferences). All unsuccessful trade attempts can also berecorded in a separate.cvs file for further study, if needed. Inaddition, various macro-scale measures such as size of developedarea in a city, average utility level and amount of preserved landalong the coast due to tax introduction, are also recorded inanother.cvs file. Our major metric of success with respect toreducing pressure on ecosystems is the number of undevelopedcells within the prime CZ, which is defined as the strip of land 8cells wide along the coast (all the area that lies to the left of theCBD). We will call this zone the CZ eco-buffer. One of the mainaggregated economic measures is total value of developed area(total property value), which is the sum of the value of developedland in a town. It is important for municipalities since propertytaxes are based on this value.

2.2.5. Initialization2.2.5.1. Spatial environment. All model experiments presented inthis paper were performed on a 35� 63 grid of cells with part of thelandscape (a strip 5 cells wide on the left) representing sea andcoastline. A single CBD3 was exogeneously set at the cell withcoordinates (�5; 0). Each spatial good, i.e. cell, gets a unique ID, andis assigned a proximity to the CBD,

PxCBD ¼ DCBDmaxþ 1� DCBD (1)

where DCBDmax is the maximum distance from the CBD in thecurrent landscape (size of the landscape can be changed in themodel), DCBD is the distance from the CBD of a specific land lot.Similarly a level of environmental amenities estimated as prox-imity to the coast PxA is defined. Both attributes are normalizedbetween 0 and 100. At initialization all land is assumed to beundeveloped.

2.2.5.2. Economic agents. For this study there are 3780 buyers andsellers participating in the land market. They are assigned pref-erences for environmental amenities (a) and for the CBD (b), aswell as budget endowments. Both preferences and budgets canbe heterogeneous or homogeneous with an average defined bythe user. The user also defines the form of utility function thatagents will use and the pricing strategies that buyers and sellerswill have, i.e. whether they will form bid/asking price equal totheir WTP/WTA or exhibit strategic pricing (Filatova et al.,2009a).

Several global variables and settings are also defined at initial-ization stage: tax rate, average agricultural land price, type of thelandscape (monocentric or coastal town with/without flood riskprobability) and position of the CBD.

2.2.6. InputNo further input is required in addition to the parameters set

during the initialization. In other implementations of the ALMAmodel we used survey data distributions to set up preferences orhazard risk perceptions of agents (Filatova et al., 2009c).

3 There are applications where multiple CBDs are considered. However for ourapplications in the Netherlands multiple service centres are of lesser interest. Whatis really important here is the location of a train station since many people live inone town and go to work by train to another.

2.2.7. Submodels2.2.7.1. Listing of vacant spatial goods in a market by sellers. In thecode of the ALMA-C model a seller is programmed to be either anurban resident willing to relocate or an owner of undevelopedland, however only the latter case is employed here. Each timestep those sellers who have not traded yet announce their askingprices. To quantify pure effects of environmental tax policy onpricing and land conversion we employ the simplest pricingstrategies for both sellers and buyers. Specifically, seller’s askingprice is equal to his4 reservation price, i.e. the price of undevel-oped land, which is assumed to be at least no less than the priceof average agricultural land same everywhere in the landscape(Pag). Note that it is undeveloped land which is converted intourban use, not agricultural. Average agricultural land price servesas opportunity costs for conversion into urban land. In the realworld agricultural productivity, cost of production and prices foragricultural goods, affecting price for agricultural land areheterogeneous. Although technically in the model it is possible toinstantiate different levels of agricultural productivity and askingprices for agricultural land, it is inessential for the purposes of ourstudy of the environmental tax policy effects on outcomes ofa land market.

2.2.7.2. Search for the best location under budget constraints bybuyers. Buyer households search for the location that maximisestheir utility U and is affordable for their budget, net of transportcosts (Y). Agent utility5 follows a logarithmic formula:

U ¼ a$lnðPxAÞ þ b$lnðPxCBDÞ (2)

Thus, agents’ utility for location depends on two spatial attributes:distance to the CBD and level of environmental amenities. Thisassumption is based on extensive empirical research and is used inother ABMs, specifically in the SLUCE model (Brown et al., 2004).Although functional forms of utility are different, both ALMA andSLUCE models assume that agents have potentially heterogeneouspreferences for CBD (service centres) and environmental amenities(aesthetic quality). SLUCE allows more than one CBD, and newservice centres appear endogenously during the simulation. ALMAhas an endogenously modelled market, which implies that alloca-tion of land is not only demand driven (i.e. agents occupy anylocation which gives them maximum utility) but happens viamarket interactions: through budget-constrained utility max-imisation and competition among buyers of land.

As in the original model of Alonso (1964), in ALMA-C distance tothe CBD enters the utility function directly, not just in the budgetconstraint. In this way we represent the disutility of commutingtime separately from travel costs. Transport costs are assumed to bea linear function of distance. Consequently, the budget of an agentwho is willing to buy land at a distance DCBD is estimated accordingto:

Y ¼ Ytotal � tcu$DCBD (3)

where tcu is transport cost per unit of distance, and Ytotal e is thetotal amount of money a household owns (to be spent on purchaseof land and other goods). ALMA-C is based on an open city model(Anas et al., 1998) assuming that individuals who did not finda location in the modelled city move to another town.

4 Following the terminology used in previous papers, we use ‘he’ for a seller and‘she’ for a buyer to make it sex-neutral.

5 A composite good is not explicitly included in the utility function; rather, tosimplify the model, the choice of housing goods is treated as weakly separable fromother consumption (Parker and Filatova, 2008).

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190184

2.2.7.3. Formation and submission of bids by buyers to sellers. Whena buyer finds her highest-utility cell she defines a bid price6 for it,based on utility she gets from a location, her budget and prices ofother goods:

Pbid ¼ Y$U2

b2 þ U2 : (4)

Following a standard spatial economics assumption that a buyershares her budget between a housing and a composite good, i.e. allother goods such as food, education, entertainment and etc. (Fujitaand Thisse, 2002), we employ the constant b as a proxy for the priceof a composite good. The properties of this demand function arediscussed in details in Filatova et al. (2009a). This bid is thensubmitted to the seller of the highest-utility cell.

2.2.7.4. Evaluation of bids received by sellers. During one time stepa seller might receive several bids, one or none. At the end of thetime step he chooses the highest possible bid among submitted,and a transaction happens.

2.2.7.5. Transaction and registration of trade. The final transactionprice is an arithmetic average of the asking price and the highestbid price.7 Transactions for matched buyers and sellers are regis-tered (i.e. a set of procedures which change ownership and assignsa new transaction price to a cell are performed) and data abouttraders and property characteristics is recorded in the output file.Land allocation and price formation happens through competitionin a land market. The main difference from the conventionalspatial economic models is that there is no unique equilibriumimposed in the model (Arthur, 2006). Prices are not determined ina centralized fashion through a global equilibrium but are insteaddefined via spatially-distributed bilateral trading8 (Figs. 1 and 2).Unsuccessful buyers and sellers participate in a land market in thenext time step. The model stops running when there are no bidsthat are higher than asking prices, i.e. all gains from trade areexhausted.

2.3. Market mechanisms for preservation of coastal ecosystems

Ecosystems and economic developments compete for space inCZs, which are squeezed under the pressure from both(Schleupner, 2008). The land closest to the coast is most valuable interms of providing ecosystem services such as habitat for biodi-versity, erosion control, and sediment retention. It is also mostattractive for people who enjoy coastal amenities. Preservation ofspace along the coastline makes it possible for natural process totake place and for coastal ecosystems to function in a moresustainable way. Ecosystem services are public goods and usuallyfollow the ‘tragedy of commons’ path (Hardin, 1968). Naturally,there are no incentives at the individual level for people to

6 In the model code we differentiate between willingness to pay e WTP e

(private information) and bid price (public information). This is interesting for theexperiments with different pricing strategies (Filatova et al., 2009a). The same isvalid for sellers’ willingness to accept (WTA) and asking price. However, for thepurpose of this paper we assume that agent states her actual WTP as her bid price(his WTA as his ask price) and in the text directly appeal to the latter.

7 Price negotiation is not modelled directly in ALMA-C, so following otherexamples (Berger, 2001; Happe, 2004) we use average of bid and ask prices. Anexample of a more advanced land price negotiation mechanisms with directmodelling of auctions is given by Polhill et al. (2008).

8 The replacement of centralized price clearance by a set of bilateral trades hasboth conceptual (i.e. that equilibrium is not an essence but is just one of possibleoutcomes) and practical advantages (e.g. a possibility to incorporate agents’ andlandscape heterogeneity, and richer represntation of interactions) (Tesfatsion,2006).

preserve them. Thus, often the only way to guarantee their healthystate is by government intervening and creating these incentives atthe individual level by making more sustainable locations morelucrative. Theoretically, such market mechanism as land tax makescoastal properties more expensive and should decrease demand forthem. With a spatially-explicit model that captures processes ina land market we can quantify and visualize the aggregated effectsof this policy that creates incentives to change individual micro-economic decisions.

In this paper the land tax is differentiated with respect toland proximity to the coast following Pompe and Rinehart,(1999). Thus, a buyer of land closer to the coast needs to payhigher taxes. The way the tax function is defined determineshow effective the measure will be in terms of actually preservingspace for ecosystems to function. We performed sensitivityanalysis with linear, logarithmic and exponential tax functionswith respect to the proximity to coast. It showed that the impactof tax on preservation of space in CZ is negligible. Apparently theamenities provided by the coast are valued still higher than thetax addition to the land price that individuals face. The taxfunction needs to differentiate almost in a stepwise way betweenthe land, which is directly along the coastline, and that, which ismore landwards. For the experiments presented below we useda hyperbolic function:

T ¼ tr$Pag$k1

ðk2 � PxAÞ; (5)

where tr is a tax rate, k1 and k2 are coefficients of a hyperbola.Coefficient k2 is set equal to the maximum proximity to the coastplus one. Coefficient k1 defines how big the impact of tax is onagent’s income and, consequently, her purchasing power. Thehigher the k1, the higher is the impact of tax on the outcomes of theland market dynamics. The rate of land tax used in the real world,for example in the USA, ranges from 0.55% to 2.25%. This is not theland tax meant to preserve ecosystems but it gives an idea of therange of property taxes used in reality.

The amount of tax, which is to be paid by a buyer of land, entersher budget constraint on the right hand side in Equation (3) (i.e.amount of tax to be paid is subtracted from buyer’s income).Consequently, it affects the bid price she is willing to pay (Equation(4)). Thus, agents with different income levels, which compete ina land market for the most attractive land in CZ, will have differentbid prices and will be differently affected by tax.

3. Results

To understand the influence of tax on the aggregated outcomeswe ran a series of 16 types of experiments. We changed only twoparameters in the experiments: tax rate and individual income,which both affect households’ willingness to pay. Each experimentis associated with the tax rate and income, specifically experiment“0.015e850”means that we ran the model with a land tax equal to1.5% and individual income equal to 850 units. We show theoutcomes of 7 representative experiments in Table 2 and discussthe results of the others below.

We begin with the experiment that replicates the monocentricurban model with amenities such as coastline (Wu, 2001).Research in agent-based computational economics advises to startwith a theoretical economic model as a benchmark case and thengradually introduce changes (LeBaron, 2006). This also serves asa structural validation of an ABM, i.e. provides its validationagainst an analytical model with the same assumptions (Manson,2002).

Fig. 3. Land price gradients (households income ¼ 800, no tax for land introduced).White circle in the centre shows the CBD. Dashed line represents CZ eco-buffer.

Table 2Economic and spatial metric outcomes of the ALMA-C experiments.

Experiments 0e800* 0.015e800 0.015-ND 800/20 0e825* 0.015e825 0e850* 0.015e850

Outcome measures

Individual utility: mean 43 43.23 42.8 42.6 42.8 42.2 42.4St. dev. 1.2 1.1 1.3 1.41 1.3 1.6 1.5Aggregate utility 26 756.6 19 798.1 26 614.26 31 758.1 25 581 36 702.3 31 572.4Urban transaction price: Mean 208 205.5 205.7 209.9 207.2 211.6 208.8St. dev. 4.8 3.4 4.61 5.9 4.3 6.9 5.4Total property value 129 399.5 94 095.6 127 988.9 156 337.5 123 903.4 183 859 155 364.5% from the base case (denoted with *) 100% 72.7% 98.9% 100% 79.3% 100% 84.5%City size (urban population) 622 458 622.1 745 598 869 744Undeveloped land in CZ eco-buffer 116 239 167.9 74 189 42 130Distance at which city border stops 26.9 20.2 25.07 28.7 23.4 30.8 26.3

The significance of bold, italic and underlined is discussed in the text in Section 3.

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190 185

3.1. Base case

The main difference between the analytical urban model andour first simulation experiment is that the centralized equilibriumland price determination is replaced by a series of bilateral tradesdistributed in space and time. Individual households endowedwithincome equal to 800 try to buy a property that maximises theirhousehold utility in the coastal city. In this experiment, so far thereis no tax for land in this city. The quantitative macro indices can befound in column “0e800” of Table 2 and the resulting map of urbanland price distribution is presented in Fig. 3. The intensity of greycolour symbolises the value of land: the darker the colour, thehigher the land price. The urban land prices are the highest alongthe coast and around the CBD. As in the analytical benchmark caseof a monocentric urbanmodel, the land price gradient is decreasingwith distance from the CBD. In addition, the presence of thecoastline serves as an attractor for households and increases thevalue of land to the left of the CBD. The city expansion stops atthe location where the bid price of a buyer falls below the price ofthe agricultural land. The light grey area in Fig. 3 shows where theundeveloped area begins and symbolises the city border.

3.2. Income increase

With the increase of individual income the city significantlyexpands and the CZ eco-buffer representing the amount of openspace along the coastline decreases (compare results of theexperiments “0e800”, “0e825” and “0e850” in bold in Table 2). Sodoes the average individual utility from location, showing thatpeople enjoy a less densely populated city, which provides moreopen space along the coast and less commuting in terms of timeand money. In spite of decrease in average utility, average urbanland prices still grow because the purchasing capacity, i.e. income,has increased.

3.3. Tax introduction

Let us now introduce an environmental tax on land and see howthe location behaviour of households may change. Now agents willhave to pay extra for a location closer to the coast, assuming allother factors are the same. Economic theory suggests that if thisoccurs, the market mechanisms start to work, and demand forthese locations goes down, since it becomes harder to find buyersthat would afford to live there. Our model is not only demanddriven: land prices are determined via bilateral trades with sellers.Therefore the bid price that buyers offer may go below the askingprice. In this case the transactions do not occur, the land is notdeveloped, and there is more space for coastal ecosystems toexercise their function and provide their services.

Let us set the tax rate equal to 1.5%. Fig. 4 presents the map ofland price distribution resulting frommarket interactions of agentswith homogeneous budgets equal to 800. In comparison to the firstexperiment (Fig. 3) the spatial form of the city has changed. Thearea closer to the coast does not get households’ bids that are highenough to cover the threshold of the Pag of sellers. The householdsin this experiment basically “voted with their feet” in favour ofliving in another town because living along the coast now bearsadditional costs. As expected, a lot of land closest to the coastlinewas left unsold and remained in natural conditions. A positiveoutcome is that the CZ eco-buffer has significantly expanded: itbecame almost double the size in the city without the land tax(compare columns “0e800” and “0.015e800” in Table 2, especiallynumbers in italic).

Introduction of tax drove developments away from this town. Asa result, in the “0.015e800” experiment the total value of devel-oped area dropped by 27% (Table 2, in italic) as compared to the

Fig. 5. Land price gradients (households income ¼ 850, and land tax ¼ 1.5%). Whitecircle in the centre shows the CBD. Dashed line represents CZ eco-buffer.

Fig. 4. Land price gradients (households income ¼ 800, and land tax ¼ 1.5%). Whitecircle in the centre shows the CBD. Dashed line represents CZ eco-buffer.

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190186

base case. This may be a significant loss for this municipality, sinceit will no longer be able to gather property taxes from these ‘lost’developments. Governments may need to make difficult trade-offsbetween preservation of ecosystems and bearing economic losses.However, this loss in property taxes can be partly compensatedfrom the funds generated by the environmental land tax. In addi-tion, as mentioned above ALMA-C assumes an open city model(Anas et al., 1998), i.e. individuals who did not find a location in themodelled city move to another town. This implies that although atthe municipality level there is an economic loss, at a higher level(province or country) there is just a redistribution of developments.In fact, this outcome could be very much in line with the stategovernment conservation goals: instead of occupying ecologically-vital areas along the coast, urban developments shift to other areas,where the competition between coastal ecosystems and urbandevelopments for space is lower.

3.4. Affluent households and tax

The results of experiment “0.015e800” indicate that in theabsence of real markets for the environmental services of healthyecosystems, economic incentives, such as taxes imposed by thegovernment, help to influence the land market to achieve moreenvironmentally friendly land use patterns. However, the outcomecan change quite dramatically if the market is opened up for indi-viduals with higher budgets. Effectively this means that we areinviting more affluent buyers to come from elsewhere and partic-ipate in the market.

We repeat the second experiment, assuming that richerhouseholds enter this urban land market (the tax remains at 1.5%

but income is increased to 850 units). The results are presented inTable 2 (see “0.015e850” column) and in Fig. 5.

In the presence of more affluent households the city expandsover the borders of the city from the previous “0.015e800”experiment (compare Figs. 4 and 5). High-income agents can affordthe additional cost of the land tax and settle along the coastline. Asa result the land right along the coastline, which was successfullypreserved with low-income population now again becomes con-verted into urban use. As a result the CZ eco-buffer actually shrinksto 89% of what it was when the agents’ income was 800 (compareunderlined numbers in columns “0.015e800” and “0.015e850” inTable 2). As expected, the higher the income the smaller the effectsof the tax on economic decisions. However, this relationship is non-linear: the total property value decrease due to 1.5% tax introduc-tion is 27.3% in the case of a representative agent with income equalto 800, 20.7% in the case of income equal to 825, and only 15.5% inthe case of income equal to 850 (underlined values in Table 2).

3.5. Income distribution and tax

In reality there is a certain distribution of income, it is nothomogeneous. Let us suppose that individuals with incomesfollowing normal distribution with a mean of 800 and a standarddeviation of 20 trade in a landmarket with imposed land tax of 1.5%(see experiment “0.015-ND 800/20” in Table 2). Since there aresome households with income higher than the average of 800 (as inexperiment “0.015e800” and “0e800”), those few individuals canafford to bid higher for the houses in the CZ eco-buffer and paya land tax. This fact can often be observed in reality: the wealthiestindividuals buy houses at the waterfront. This implies that ifincome levels are heterogeneous then a market mechanism for

Fig. 6. Land price gradients (households incomes follow normal distribution and landtax ¼ 1.5%). White circle in the centre shows the CBD. Dashed line represents CZeco-buffer.

Table 3Number of undeveloped cells in the CZ eco-buffer under different tax rates andincome values.

Individual income 800 825 850

Tax rate

0% 116 74 420.5% 148 100 601.0% 196 138 901.5% 239 189 1302.0% 285 215 169

10%

20%

30%

40%

50%

60%

CZ

ec

o-b

uffe

r

800

825

850

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190 187

preservation of space for ecosystems can exacerbate this negativeside effect. Namely, it further cuts-off low-income people fromenjoying coastal amenities leaving the latter as a privilege for high-income group only. Although even without any tax high-incomeindividuals have stronger purchase power and are likely to getproperties along the coast, the presence of the land tax, whichmakes land less affordable in general, can intensify income segre-gation. We did not anticipate this result at the beginning; it is anemergent outcome of sorting via the land market. Social equity (i.e.fair access to resources) becomes an issue here: while pursuing thegoal of preserving coastal ecosystems the policy makes the coastalland even more unaffordable for low-income people and sets itaside for the benefits of the affluent ones only.

In terms of ecological benefits market incentives still work andpart of the CZ eco-buffer becomes vacant (168 instead of 116 in thecase with no land tax). However, this is nowhere near the resultpredicted by amodel with homogeneous income levels (239 vacantlots). In addition, while trade-offs between ecosystem preservationand social equity become more intense, the trade-offs betweenecological benefits and economic costs vanish. The loss in the totalproperty value due to the introduction of the environmental tax onland in the casewith heterogeneous incomes is only 1% (Table 2).9 Ithappens mainly because some individuals with higher income arealways ready to pay more for the land along the coast, even

9 Economic losses are smaller in heterogeneous income case compared tohomogeneous aslo because there is less undeveloped land in CZ eco-buffer.Nevertheless, the difference in CZ eco-buffer of 29.7% is not the main reason for the96.3% decrease in economic losses.

considering extra costs for paying environmental tax on land.Consequently, they effectively reimburse the loss from thedecreased land prices paid by individuals with lower incomes. Theaverage urban transaction price is almost the same as inthe experiment with homogeneous agents compared to heteroge-neous case. However, land patterns are very different (comparecolumns “0.015e800” and “0.015-ND 800/20” and Figs. 4 and 6).This underlines the importance of spatially-explicit modelling ofland markets. This 1% decrease in total property value in hetero-geneous case is significantly lower than the 27% decrease in thehomogeneous case. In spite of the fact that the population of agentsin experiments “0.015e800” and “0.015-ND 800/20” is on averagethe same, the effects of market mechanisms designed for coastalland preservation are different. Since there is always a certaindistribution of incomes in the reality, a conventional representativeagent model is likely to overestimate economic losses due to taxintroduction. In practice, however, many models for policy decisionsupport, especially integrated models, use a simple representativeagent model (Kirman, 1992). As we see, policy decision supporttools based on the average representative agent model are likely tooverestimate the effects of market measures on the preservation ofspace for ecosystems, as well as overestimate subsequenteconomics losses. In addition, the impact of the environmental taxon social equity is likely to be omitted.

In fact, the last three experiments (“0.015e800”, “0.015e850”and “0.015-ND 800/20”) showed that economic incentives work asplanned only if agents have fixed and homogeneous incomes. Iftaxes are introduced but higher-income households enter the landmarket then the effect of the tax is almost eliminated. The

0%0% 0.5% 1.0% 1.5% 2.0%

Tax rate

Fig. 7. The percentage of the CZ eco-buffer not occupied by urban land use.

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190188

introduction of tax and increase in household incomes drive spatialand land price dynamics in opposite directions. To further investi-gate the relationship between these two factors, we performeda series of additional experimentswith tax rates changing from0% to2% and income changing from800 to 850 (running each for 30 timesas before). The results in terms of the undeveloped CZ eco-bufferarea are presented in Table 3. Visually the relationship is presentedin Fig. 7. One can see that for a population with high income of 850(about 6% increase compared to the base case of 800) the preservedCZ buffer decreases almost twice under the tax rate of 2%. Thus, themodel allows us to analyze and visualize the size and location of thepreserved space for ecosystems as a function of income and tax rate.

4. Conclusions

This paper presents an ABM designed to explore the effects ofeconomic incentives on preservation of space for CZ ecosystems.The introduction of a land market that models emergence ofmacro economic indices and patterns from the bottom-up in a 2Dlandscape is a vital part of the model. The ALMA-C model isgrounded in spatial economics and is adjusted for coupling withspatially-explicit environmental models. Consequently, it departsfrom some of the restrictive assumptions of conventionaleconomic models. In particular, the centralized global equilibriumprice determination mechanism is replaced by a set of distributedbilateral trades, the rationality of economic agents is bounded, andboth heterogeneity of agents and the landscape can be combinedand modelled in a spatially-explicit way while maintaining fully-modelled market. We performed 16 types of simulations with themodel to understand what aggregated results (spatial patternsand land prices) emerge from individual interactions in a landmarket under different assumptions of income distribution andtax rate.

4.1. Effects of market mechanisms such as land tax on preservationof space for coastal ecosystems

Our analysis has shown that economic incentives may affecturban developments in CZ and ease coastal squeeze, but there mayalso be unforeseen outcomes. The land tax aimed at preservingspace for ecosystems, works best if a population has fixed andhomogeneous incomes. If incomes grows or even if they stay thesame on average but are non-uniformly distributed, then taxbecomes less effective. Moreover, agents with homogeneous andheterogeneous (with the same average) incomes produce qualita-tively different results: although average land price is almost thesame, the individual land prices and 2D spatial patterns are verydifferent. A representative agent model overestimates both the sizeof the undeveloped CZ eco-buffer (ecological benefits) and the lossin the total value of developed area (economic costs). Besides, ifpopulation is heterogeneous in incomes, there are negative effectsof the tax with respect to social equity.

4.2. Side-effects of the tax introduction

I Trade-offs between environmental benefits and ecological losses.It is a common expectation that measures aimed at moresustainable and ecologically friendly outcomes inevitablyinvolve economic losses. Our results show that first of all, ina coastal town where individuals have heterogeneousincomes the loss in total property value due to tax intro-duction can be very small (just about 1% compared to the basecase with no tax introduction). Secondly, an open city modelimplies that unsettled households will search for a locationelsewhere. Thus, at the state or provincial scale the

government is likely to achieve the result it needs: redistri-bution of developments into the areas where urbanizationdoes not have to compete for space with coastal ecosystems.This implies that, while total property value along the coastmay go down, it will go up in some other municipality else-where. Thirdly, the loss in total value of developed land ina city where the environmental tax is introduced can bepartly reimbursed from the funds generated by this tax.

II Trade-offs between ecological benefits and social equity. Theenvironmental tax may also result in making coastal landaffordable exclusively tomore affluent residents. Often the CZbecomes the domain of luxury McMansions. Even withoutany land tax, high-income people are willing to pay more forland with coastal amenities and are more likely to get it. Theenvironmental land tax exacerbates this process. It playsa role in driving low-income people away from the coastalland, which then gets developed and shaped for the moreaffluent residents. This still may have a positive effect on theenvironment assuring less residential density, however gainsin environmental quality come with more social inequity.Whether this should be accepted as an inevitable attribute ofmarket-based environmental management, remains an openquestion, but at least should be recognized andmade open forpublic scrutiny.

All this indicates that market mechanisms should be verycarefully designed. Form of a tax function and distribution ofincomes largely affect success of taxes in managing ecosystemservices availability. It may seem easier to simply prohibit devel-opment in the CZ. However, local governments optimize overmultiple criteria striking a balance between ecosystem preserva-tion and increasing taxable property value. Market mechanismsallow more flexible use of protected areas and sharing costs orsometimes risks between government and individuals. Anyway, inthis paper we have limited ourselves to exploring the effectivenessof market instruments, rather than choosing between regulatoryand market instruments.

The ALMA-C model is a big simplification of a real world coastaltown.We used artificial data for both the agents’ behaviour and thespatial environment. Validation of ABMs is an important issue(Manson, 2002; Richiardi et al., 2006). Our model successfullypassed structural validation (Filatova et al., 2009a,b). Micro andmacro validation is restricted due to the difficulty and costs ofobtaining empirical data about individual behaviour in a landmarket (Robinson et al., 2007). However, the simplicity of a modelis also an advantage. It has been noted in numerous studies thatsmall, relatively simple models are easier to relate to mentalmodels prevailing in the society (Forrester, 1994). Simple modelsare easier to change as new data become available and newprocesses are incorporated. This makes it also easier to communi-cate the results to the public, and to use themodels in participatory,stakeholder driven studies (Voinov and Bousquet, 2010). The ALMAmodel can be envisioned as part of a decision support system,producing results that help to visualize the emergent propertiesand to understand the driving forces in the system. In spite of itssimplicity, the model still captures the complexity and emergenceof macro phenomena from human interactions in a market andindirectly with governments who determine macro conditions forthose economic actions and who are interested in the macro-outcomes of those actions.

Obviously, this simple model can be improved in many ways.Keeping in mind the ‘keep-it-simple’ principle for models ofcomplex adaptive systems, largely advocated in the ABM commu-nity (Gilbert and Troitzsch, 2005), the following challenges seem tobe the most interesting to address in the future:

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190 189

- More advanced modelling of the spatial environment, partic-ularly dependence of ecosystem functions and services on theproportion, pattern and location of development (Wu andIrwin, 2008). Preservation of large contagious clusters of landis usually considered more ecologically valuable, than a frag-mented landscape. Also, the more people live in CZ, the lowerlevel of coastal amenities it starts to provide, which, in turn,makes it less attractive for newcomers.

- Explicit modelling of a governmental agent who might want tooptimize different social welfare functions (dependent on theaccumulation of economic activities only or maintenance ofecosystem services or social equity). Most interesting is tomodel several neighbouring municipalities with differentecosystem management policies to see which of them attractmore wealth or do a better job preserving ecosystems ormanage to do both. The idea goes in line with the concept ofTiebout (1956): how households sort themselves in spacebetween jurisdiction on the basis of the level of provision ofpublic goods, healthy environment, coastal in this case, asmoderated by different tax rates.

- Introducing neighbourhood effects (e.g. agents choosing locationdepending on who are the other people in the neighbourhood)mayadd an interestingdynamics to themodel. The ‘BeverlyHills’segregation and centripetal effect of high prices makes it moredesirable to buy in places where prices are already high.

Policy makers have a difficult task of balancing ecological,economic and social aspects. Effects of policies aimed at managingcomplex human-environmental systems may not always be fore-seen. Theoretical analytical models can be very useful in under-standing the main trend of system development. However,spatially-explicit coupled ecological-economic systems are neededto quantify and visualize potential outcomes. Our simulationsindicate that while aggregated economic measures are the same(e.g. average land price), spatial distribution of developed land canbe very different and can even have opposite effects for the CZ eco-buffer. These results also illustrate how measures meant for pres-ervation of space for ecosystems may add to social inequity.Nevertheless, analytical and statistical economic models used forpolicy support in managing coupled ecological-economic systemsare usually based on the concept of a representative agent (i.e.homogeneous agents) and would not be able to capture this effect.In addition, representative agent models overestimate ecologicalbenefits and economic costs due to the introduction of the tax.Spatial ABMs, in contrast, are able to combine a spatially-explicitset up with environmental concerns and the socio-economicbehaviour. Thus, they are able to track, quantify and visualize howpolicies in complex environments may affect individual behavioursand what aggregated patterns and economic features emerge fromindividual interactions in land markets, for example.

Acknowledgements

The authors are grateful to the staff of the Centre for SocialComplexity at George Mason University and to Dr. D.C. Parker inparticular for valuable input and advice.We are also very grateful toanonymous reviewers whose comments helped us to improve themanuscript.

References

Alonso, W., 1964. Location and Land Use. Harvard University Press, Cambridge, MA.Anas, A., Arnott, R., Small, K.A., 1998. Urban spatial structure. Journal of Economic

Literature 36 (3), 1426e1464.

Arthur, W.B., 2006. Out-of-equilibrium economics and agent-based modeling. In:Tesfatsion, L., Judd, K.L. (Eds.), Handbook of Computational Economics. Agent-basedComputational Economics, vol. 2. Elsevier B.V., Amsterdam, pp.1551e1564.

Bagstad, K.J., Stapleton, K., D’Agostino, J.R., 2007. Taxes, subsidies, and insurance asdrivers of United States coastal development. Ecological Economics 63 (2e3),285e298.

Barreteau, O., Bousquet, F., Attonaty, J.-M., 2001. Role playing game for opening theblack box of multi-agent systems: method and lessons of its application toSenegal River Valley irrigated systems. Journal of Artificial Societies and SocialSimulation 4 (2), 12.

Berger, T., 2001. Agent-based spatial models applied to agriculture: a simulationtool for technology diffusion, resource use changes, and policy analysis. Agri-cultural Economics 25 (2e3), 245e260.

Bockstael, N.E., December 1996. Economics and ecological modeling: the impor-tance of a spatial perspective. American Journal of Agricultural Economics,1168e1180.

Bolte, J.P., Hulse, D.W., Gregory, S.V., Smith, C., 2007. Modeling biocomplexity e

actors, landscapes and alternative futures. Environmental Modelling & Software22 (5), 570e579.

Bousquet, F., Le Page, C., 2004. Multi-agent simulations and ecosystem manage-ment: a review. Ecological Modelling 76 (3e4), 313e332.

Brown, D.G., Page, S.E., Riolo, R., Rand, W., 2004. Agent-based and analyticalmodeling to evaluate the effectiveness of greenbelts. Environmental Modelling& Software 19 (12), 1097e1109.

Brown, D.G., Robinson, D.T., An, L., Nassauer, J.I., Zellner, M., Rand, W., Riolo, R.,Page, S.E., Low, B., Wang, Z.F., 2008. Exurbia from the bottom-up: confrontingempirical challenges to characterizing a complex system. Geoforum 39 (2),805e818.

Costanza, R., d’Arge, R., de Groot, R., Farberk, S., Grasso, M., Hannon, B., Limburg, K.,Naeem,S.,O’Neill, R.V.,Paruelo, J., Raskin, R.G., Suttonkk,P.,vandenBelt,M.,1997.Thevalue of the world’s ecosystem services and natural capital. Nature 387, 253e260.

Costanza, R., Andrade, F., Antunes, P., van den Belt, M., Boesch, D., Boersma, D.,Catarino, F., Hanna, S., Limburg, K., Low, B., Molitor, M., Pereira, J.G., Rayner, S.,Santos, R., Wilson, J., Young, M., 1999. Ecological economics and sustainablegovernance of the oceans. Ecological Economics 31, 171e187.

Costanza,R., Perez-Maqueo,O.,Martinez,M.L., Sutton,P.,Anderson,S.J.,Mulder,K., 2008.The value of coastal wetlands for hurricane protection. Ambio 37 (4), 241e248.

Crooks, A., Castle, C., Batty,M., 2008. Keychallenges in agent-basedmodelling for geo-spatial simulation. Computers Environment and Urban Systems 32 (6), 417e430.

Elshorbagy, A., Ormsbee, L., 2006. Object-orientedmodeling approach to surfacewaterquality management. Environmental Modelling & Software 21 (5), 689e698.

Filatova, T., van der Veen, A., Parker, D.C., 2009c. Why Does Individual RiskPerception Matter in Land Use Modeling? Combining Survey Data and Agent-Based Land Market Model. 6th European Social Simulation AssociationConference. University of Surrey, Guildford, UK, September 14e18. ElectronicPublication, 14 pp.

Filatova, T., Parker, D., van der Veen, A., 2009a. Agent-based urban land markets:agent’s pricing behavior, land prices and urban land use change. Journal ofArtificial Societies and Social Simulation 12 (1), 3.

Filatova, T., van der Veen, A., Parker, D., 2009b. Land market interactions betweenheterogeneous agents in a heterogeneous landscapedtracing the macro-scaleeffects of individual trade-offs between environmental amenities and dis-amenities. Canadian Journal of Agricultural Economics 57 (4), 431e457.

Fitz, H.C., Sklar, F.H., Waring, T., Voinov, A., Costanza, R., Maxwell, T., 2003. Devel-opment and application of the everglades landscape model. In: Costanza, R.,Voinov, A. (Eds.), Spatially Explicit Landscape Modeling. Springer.

Forrester, J.W., 1994. Policies, decisions, and information sources for modeling. In:Morecroft, J.D.W., Sterman, J.D. (Eds.), Modeling for Learning Organizations.Productivity Press, Portland, OR, pp. 51e84.

Fujita, M., Thisse, J.-F., 2002. Economics of Agglomeration. Cities, Industrial Locationand Regional Growth. Cambridge University Press.

Gilbert, N., Troitzsch, K.G., 2005. Simulation for the Social Scientist, second ed. OpenUniversity Press, Glasgow.

Gotts, N.M., Polhill, J.G., 2009.When andhow to imitate your neighbours: lessons fromand for FEARLUS. Journal of Artificial Societies and Social Simulation 12 (3), 2.

Grimm, V., Berger, U., Bastiansen, F., Eliassen, S., Ginot, V., Giske, J., Goss-Custard, J.,Grand, T., Heinz, S.K., Huse, K., Huth, A., Jepsen, J.U., Jørgensen, C., Mooij, W.M.,Muller, B., Peer, G., Piou, K., Railsback, S.F., Robbins, A.M., Robbins, M.M.,Rossmanith, E., Ruger, N., Strand, E., Souissi, S., Stillman, R.A., Vabø, R., Visser, R.,DeAngelis, D.A., 2006. A standard protocol for describing individual-based andagent-based models. Ecological Modelling 198, 115e126.

Gross, J.E., McAllister, R.R.J., Abel, N., Stafford Smith, D.M., Maru, Y., 2006. Australianrangelands as complex adaptive systems: a conceptual model and preliminaryresults. Environmental Modelling & Software 21 (9), 1264e1272.

Happe, K., 2004. Agricultural Policies and Farm Structures e Agent-Based Modellingand Application to EU-Policy Reform. IAMO Studies on the Agricultural andFood Sector in Central and Eastern Europe 30.

Happe, K., Kellermann, K., Balmann, A., 2006. Agent-based analysis of agriculturalpolicies: an illustration of the agricultural policy simulator AgriPoliS, its adap-tation and behavior. Ecology and Society 11 (1), 49.

Hardin, G., 1968. The tragedy of the commons. Science 162, 1243e1248.Irwin, E.G., Bockstael, N.E., 2004. Land use externalities, open space preservation,

and urban sprawl. Regional Science and Urban Economics (34), 705e725.Janssen, M.A., 2002. Complexity and Ecosystem Management: The Theory and

Practice of Multi-Agent Systems. Edward Elgar Publishing Ltd.

T. Filatova et al. / Environmental Modelling & Software 26 (2011) 179e190190

Kirman, A.P., 1992. Whom or what does the representative individual represent?Journal of Economic Perspectives 6 (2), 117e136.

de Kok, J.L., Engelen, G., White, R., Wind, H.G., 2001. Modeling land-use change ina decision-support system for coastal-zone management. EnvironmentalModeling & Assessment 6 (2), 123e132.

Lagabrielle, E., et al., 2010. Modelling with stakeholders to integrate biodiversityinto land-use planning lessons learned in Reunion Island (Western IndianOcean). Environmental Modelling & Software 25 (11), 1413e1427.

LeBaron, B., 2006. Agent-based computationalfinance. In: Tesfatsion, L., Judd, K.L. (Eds.),Handbook of Computational Economics. Agent-Based Computational Economics,vol. 2. Elsevier B.V., pp. 1187e1233.

Manson, S.M., 2002. Validation and verification of multi-agent systems. In:Janssen, M.A. (Ed.), Complexity and Ecosystem Management: The Theory andPractice of Multi-Agent Systems. Edward Elgar Publishing Ltd, pp. 63e75.

Marshall, E., Randhir, T.O., 2008. Spatial modeling of land cover change andwatershed response using Markovian cellular automata and simulation. WaterResources Research 44 (4).

Martínez, M.L., Intralawan, A., Vázquez, G., Pérez-Maqueo, O., Sutton, P.,Landgrave, R., 2007. The coasts of our world: ecological, economic and socialimportance. Ecological Economics 63, 254e272.

Matthews, R.B., Gilbert, N.G., Roach, A., Polhill, J.G., Gotts, N.M., 2007. Agent-basedland-use models: a review of applications. Landscape Ecology 22, 1447e1459.

Monticino, M., Acevedo, M., Callicott, B., Cogdill, T., Lindquist, C., 2007. Coupledhuman and natural systems: a multi-agent-based approach. EnvironmentalModelling & Software 22 (5), 656e663.

Moreno, N., Quintero, R., Ablan, M., Barros, R., Davila, J., Ramirez, H., Tonella, G.,Acevedo, M.F., 2007. Biocomplexity of deforestation in the Caparo tropical forestreserve in Venezuela: an integrated multi-agent and cellular automata model.Environmental Modelling & Software 22 (5), 664e673.

Naivinit, W., et al., 2010. Participatory agent-based modeling and simulation of riceproduction and labor migrations in Northeast Thailand. EnvironmentalModelling & Software 25 (11), 1345e1358.

Nicholls, R.J., Wong, P.P., Burkett, V.R., Codignotto, J.O., Hay, J.E., McLean, R.F.,Ragoonaden, S., Woodroffe, C.D., 2007. Coastal systems and low-lying areas. In:Parry, M.L., Canziani, O.F., Palutikof, Van der Linden, P.J., Hanson, C.E. (Eds.),Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution ofWorking Group II to the Fourth Assessment Report of the IntergovernmentalPanel on Climate Change. Cambridge University Press, Cambridge, UK, pp.315e356.

Parker, D.C., Berger, T., Manson, S.M. (Eds.), 2002. Agent-based Models of Land-Useand Land-Cover Change: Report and Review of an International Workshop,October 4e7, 2001. LUCC Report Series Bloomington. LUCC.

Parker, D.C., Manson, S.M., Janssen, M.A., Hoffmann, M., Deadman, P., 2003. Multi-agent systems for the simulation of land-use and land-cover change: a review.Annals of the Association of American Geographers 93 (2), 314e337.

Parker, D.C., Hessl, A., Davis, S.C., 2008. Complexity, land-use modeling, and thehuman dimension: fundamental challenges for mapping unknown outcomespaces. Geoforum 39 (2), 789e804.

Parker, D.C., Filatova, T., 2008. A conceptual design for a bilateral agent-based landmarket with heterogeneous economic agents. Computers, Environment andUrban Systems 32, 454e463.

Polhill, J.G., Parker, D., Gotts, N.M., 2008. Effects of land markets on competitionbetween innovators and imitators in land use: results from FEARLUS-ELMM.In: Hernandez, C., Troitzsch, K., Edmonds, B. (Eds.), Social Simulation Tech-nologies: Advances and New Discoveries. Information Science Reference,Hershey, PA, pp. 81e97.

Pompe, J.J., Rinehart, J.R., 1999. Establishing fees for beach protection: paying fora public good. Coastal Management 27 (1), 57e67.

Richiardi, M., Leombruni, R., Saam, N., Sonnessa, M., 2006. A common protocol foragent-based social simulation. Jasss-the Journal of Artificial Societies and SocialSimulation 9 (1).

Robinson, D.T., Brown, D., Parker, D.C., Schreinemachers, P., Janssen, M.A.,Huigen, M., Wittmer, H., Gotts, N., Promburom, P., Irwin, E., Berger, T.,Gatzweiler, F., Barnaud, C., 2007. Comparison of empirical methods for buildingagent-based models in land use science. Journal of Land Use Science 2 (1),31e55.

Schleupner, C., 2008. Evaluation of coastal squeeze and its consequences for theCaribbean island Martinique. Ocean & Coastal Management 51 (5), 383e390.

Simon, H., 1997. Models of Bounded Rationality. In: Emperically GroundedEconomic Reason, vol. 3. MIT Press.

Spalatro, F., Provencher, B., 2001. An analysis of minimum frontage zoning topreserve lakefront amenities. Land Economics 77 (4), 469e481.

Tesfatsion, L., 2006. Agent-based computational economics: a constructiveapproach to economic theory. In: Tesfatsion, L., Judd, K.L. (Eds.), Handbook ofComputational Economics. Agent-Based Computational Economics, vol. 2.Elsevier B.V., Amsterdam, pp. 831e880.

Tiebout, C.M., 1956. A pure theory of local expenditures. Journal of PoliticalEconomy 64 (5), 416e424.

Veraart, J.A., Brinkman, S., Klostermann, J.E.M., Kabat, P. (Eds.), 2007. ClimateChanges Spatial Planning. Introduction to the Dutch National Research Pro-gramme. CcSP-Brochure, Wageningen, The Netherlands.

Voinov, A., Bousquet, F., 2010. Modeling with stakeholders. Environmental Model-ling & Software 25 (11), 1268e1281.

Voinov, A., Costanza, R., Wainger, L., Boumans, R., Villa, F., Maxwell, T., Voinov, H.,1999. Patuxent landscape model: integrated ecological economic modeling ofa watershed. Environmental Modelling and Software 14, 473e491.

Wilensky, U., 1999. NetLogo. Center for Connected Learning and Computer-BasedModeling, Northwestern University, Evanston, IL. http://ccl.northwestern.edu/netlogo/.

Wu, J., 2001. Environmental amenities and the spatial pattern of urban sprawl.American Journal of Agricultural Economics 83 (3), 691e697.

Wu, J., Adams, R.M., Plantinga, A.J., 2004. Amenities in an urban equilibriummodel: residential development in Portland, Oregon. Land Economics 80 (1),19e32.

Wu, J.J., Irwin, E.G., 2008. Optimal land development with endogenous environ-mental amenities. American Journal of Agricultural Economics 90 (1), 232e248.