Cities 32 (2013) 33–42

Contents lists available at SciVerse ScienceDirect

Cities

journal homepage: www.elsevier .com/locate /c i t ies

Spatiotemporal simulation of urban growth patterns using agent-basedmodeling: The case of Tehran

Jamal Jokar Arsanjani a,⇑, Marco Helbich a, Eric de Noronha Vaz b

a Heidelberg University, Institute for Geography, GIScience Group, Heidelberg, Germanyb Ryerson University, Department of Geography, Toronto, Canada

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

Article history:Received 3 August 2012Received in revised form 17 December 2012Accepted 26 January 2013

Keywords:Urban growthAnalytic hierarchy processMulti criteria analysisAgent-based modelingTehran

0264-2751/$ - see front matter Crown Copyright � 2http://dx.doi.org/10.1016/j.cities.2013.01.005

⇑ Corresponding author. Address: BerlinerstrasseHeidelberg, Germany. Tel.: +49 6221 544520.

E-mail addresses: jamaljokar@gmail.com (J. Jokargeog.uni-heidelberg.de (M. Helbich), evaz@ryerson.ca

Rapid urban growth is becoming a serious problem in most developing countries. Tehran, the capital ofIran, stands out as a vibrant metropolitan area, facing uncontrolled urban expansion. Public authoritiesand decision makers require planning criteria regarding possible spatial developments. To monitor pastdevelopmental trends and to simulate emerging spatiotemporal patterns of urban growth, this researchapplies a geosimulation approach that couples agent-based modeling with multicriteria analysis (MCA)for the period between 1986 and 2006. To model the major determinants controlling urban develop-ment, three agent groups are defined, namely developer agents, government agents, and residentagents. The behaviors of each agent group are identified by qualitative surveys and are considered sep-arately using multi-criteria analysis. The interactions of the agents are then combined through overlayfunctions within a Geographic Information System (GIS). This analysis results in the creation of a pro-pensity surface of growth that is able to identify the most probable sites for urban development. Sub-sequently, a Markov Chain Model (MCM) and a concise statistical extrapolation are used to determinethe amount of probable future expansion in Tehran. For validation purposes, the model is estimatedusing 2011 data and then validated based on actual urban expansion. Given the accurate predictionsof the Markov Chain Model, further predictions were carried out for 2016 and 2026. This simulationprovides strong evidence that during the next decade planning authorities will have to cope with con-tinuous as well as heterogeneously distributed urban growth. Both the monitoring of growth and sim-ulation revealed significant developments in the northwestern part of Tehran, continuing toward thesouth along the interchange networks.

Crown Copyright � 2013 Published by Elsevier Ltd. All rights reserved.

Introduction

Changes in land use in the last decades have resulted in a wid-ening debate on the future impact of urban growth (e.g., Bhatta,Saraswati, & Bandyopadhyay, 2010; Halleux, Marcinczak, & vander Krabben, 2012; Tv, Aithal, & Sanna, 2012; Weber & Puissant,2003). Land use change is intrinsically caused by the spatial changein population and socio-economic dynamics, which has dire conse-quences for land use and exerts a continuous strain on the land-scape. With economic growth and the population increase, manygreen spaces have been transformed into residential, industrial,and commercial use. To foster improved decision-making, thesedynamic processes require permanent monitoring with respect topast developments and to forecast future growth.

013 Published by Elsevier Ltd. All r

48, Institute for Geography,

Arsanjani), marco.helbich@-(E. de Noronha Vaz).

Tools such as city evolution trees (Wang et al., 2012) offer anew way to investigate urban growth on a global scale and pro-vide an integrative vision of land use change and the conse-quences of such change for the urban environment. At the locallevel, the combination of agent-based models (ABM) (Crooks,Castle, & Batty, 2008; Manson, 2005; Perez & Dragicevic, 2012;Robinson, Murray-Rust, Rieser, Milicic, & Rounsevell, 2012) repre-sents the latest technique to simulate the effect of the behavior ofindividuals on land change. ABMs are elegant tools for addressingthe complex, non-linear behavior of urban systems. Alternativeurban simulation approaches rely on a set of fixed methods andassumptions (Koomen, Vasco, Dekkers, & Rietveld, 2012), includ-ing cellular automata (Barredo, Demicheli, Lavalle, Kasanko, &McCor mick, 2004; Clarke, Hoppen, & Gaydos, 1997; He, Okada,Zhang, Shi, & Zhang, 2006), Markov chain models (Haibo, Longji-ang, Hengliang, & Jie, 2011; Wu et al., 2006), spatial logisticregression (Hu & Lo, 2007), rule-based models (Shi, Sun, Zhu, Li,& Mei, 2012; Tayyebi, Pijanowski, & Pekin, 2011), evolution trees(Wang et al., 2012), and multi-agent models (Fang, Gertner, Sun,

ights reserved.

34 J. Jokar Arsanjani et al. / Cities 32 (2013) 33–42

& Anderson, 2005; Lagarias, 2012). These approaches seek solu-tions to various problems, such as (i) the velocity of simulationprocessing times, (ii) the spatial resolution of urban growth simu-lations, (iii) the incorporation of policy dimensions and qualitativeinformation, and (iv) the reliability and accuracy of urban growthsimulations.

The advantage of using ABM is strongly linked to the possibilityof representing movement independently of scale and in a non-constricted relation, such as with cellular automata (Longley &Batty, 2003). This advantage enables using ABM as a micro-simula-tion technology for a wide range of spatial applications to modelurban change. Similar to other models, ABM facilitates the incorpo-ration of subsets of different ancillary data and discrete variables.In addition, the ubiquity of the agents results in a larger vision ofwhat the agents represent (Bakker & van Doorn, 2009). Thus, theagents can represent human behavior more accurately and offerbetter assessments based on empirical findings, not only concep-tual economic models (Fujita & Kashiwadani, 1989). This integra-tion of complex systems science in the ABM modeling frameworkresults in more elaborate representations of scalable urban dynam-ics in space and time (Matthews, Gilbert, Roach, Polhill, & Gotts,2007). The complexity of urban systems is well documented inBatty (2008). Such systems include many changeable parametersand large numbers of discrete actors (e.g., inhabitants and govern-mental agencies). A bottom-up approach to urban phenomena en-ables understanding dynamics on a local level and results in a moreelaborate explanatory framework for urban growth (Crooks, 2006).These improvements have only become possible in recent yearsbecause of technological advances and facilitate better spatial visu-alization than traditional urban growth models (Torrens, 2006). Jo-kar, Kainz, & Mousivand, 2011 used a CA-Markov model to monitorland use changes and to predict future states. In addition, A. Jokar,Helbich, Kainz, & Darvishi (2013) employed a hybrid logisticregression-CA-Markov model to illustrate the most probable sitesfor development based on a number of environmental and socio-economic driving forces. Tayyebi et al. (2011) developed an urbangrowth boundaries model based on artificial neural networks andcorresponding factors and predicted the dynamics of the Tehranurban boundary for 2012. Nonetheless, most of the cited studiesmodel urban expansion using a number of physical and socio-eco-nomic predicators and neglect the effect of behavior of individualson urban growth dynamics. However, the inhabitants and theirdecisions with respect to where to live are the main actors of landconstruction (e.g., Koomen, Stillwell, Bakema, & Scholten, 2007;Torrens, 2006). Therefore, the primary objective of this study isto consider the behavior and the preferences of individuals andto simulate urban growth accordingly. Thus, this paper is intendedto monitor the spatiotemporal patterns of Tehran and to distin-guish the driving forces of the recent expansions as the major pre-dictors of future growth.

This paper develops an approach to modeling urban growthbased on the simulation capabilities of ABM. From an empiricalperspective, Tehran, Iran, is used as a laboratory to assess the com-bined effects of urban transition based on biophysical, social, andeconomic driving forces and to predict future land use changes.Tehran has been the primary goal for migrants from other Iraniancities because the city offers better incomes and infrastructureand accessibility to authorities, high-ranking universities, andinstitutions, among other advantages (Sabet Sarvestani, Ibrahim,& Kanaroglou, 2011; Sintusingha, & Zamani, 2013; Zebardast,2006).

The paper is organized as follows. The next section provides anoverview of the materials that were used. The section that followspresents the modeling process and discusses the results. In theconcluding section, the paper presents the major implications ofthe research findings.

Materials

Study site

Tehran is the capital of Iran and the administrative center ofTehran Province. The center of the study area is located at a lati-tude and longitude of 35.6962�N and 51.4230�E, respectively.Fig. 1 represents an overview of Tehran and the city’s geographicextent. The study area covers an approximate area of 1865 km2,which includes Tehran and a few nearby cities (Islam Shahr, NasimShahr, Shahr-e Ghods) as part of the interior metropolitan area.However, the greater metropolitan area of Tehran encompassesadditional nearby cities, including Karaj, which was officially des-ignated as Alborz Province to shift administrative entities and busi-ness functions toward Karaj. From 1975 to 2010, Tehran’sresidential population nearly doubled as a result of political adjust-ments, which made the city the world’s 17th-largest and one of thecities with the largest annual growth in Asia (World Gazetteer,2012). This population growth in recent decades has resulted indrastic changes in the urban landscape (Jokar A. et al., 2011Mulli-gan & Crampton, 2005; Rafiee, Mahiny, Khorasani, Darvishsefat, &Danekar, 2009). Today, this expansion is mainly driven by subur-ban growth. For instance, Shahriar, one of Tehran’s suburbs, has re-ported a population increase of approximately 9.6%. In comparison,the population growth of inner Tehran is 2.2% (World Gazetteer,2012). For European cities, it is well documented (e.g., Helbich,2012; Helbich & Leitner, 2010; Vaz, Cabral, Caetano, Painho, & Nijk-amp, 2012) that such growth processes have a substantial impacton the metropolitan landscape. In recent years, Tehran’s dynamicurban fabric and the city’s future development have been studied(e.g., Mousivand, Alimohammadi Sarab, & Shayan, 2007; Tayyebiet al., 2011).

Pre-processing of the input data

For this research, a set of environmental attributes (e.g., topog-raphy, recreation areas, and the transport network), socioeconomicvariables (e.g., population and land and housing prices), and tem-poral multi-spectral satellite images were collected (see Table 1).All data were converted into the raster file format and classifiedusing the nearest neighbor algorithm at a 30 m resolution and aUTM-WGS 1984 Zone 39N projection system.

A temporal set of Landsat images (i.e., MSS, TM, and ETM+images) with the highest signal-to-noise ratio index under non-cloudy circumstances were collected for the last three decadesusing the Earth Science Data Interface of the Global Land CoverFacility and the Earth Resources Observation and Science Centerof the US Geological Survey. The Landsat imagery was chosen tomonitor Tehran’s urban growth because of the fine temporal reso-lution and the richness of the archive. This satellite imagery en-sured the accuracy of the land use maps generated by thenational authorities and enabled the correction of misclassifica-tions and wrong attributes. The years 1986, 1996, 2006, and2011 were chosen because of the availability of socioeconomicdata for these dates. Furthermore, these time frames comprise dif-ferent eras of land use policy in Iran. For example, 1986 witnessedthe Iran-Iraq war, which had consequences for urban policies andpopulation movements. The year 1996 was the beginning of finan-cial growth and the climax of post-war reconstruction and of gov-ernment-sponsored developments. The year 2006 marks themiddle of the era of substantial private and national constructioninvestment by land developers and the beginning of governmentconcerns regarding Tehran’s enormous urban growth. During thelast period (1996–2006), strict land policies were issued toconserve the environment and farmland. The establishment of

Fig. 1. The interior metropolitan area of Tehran.

Table 1Description of the collected geospatial and socio-economic data.

Data type Dataset Source

(1) Environmentalfeatures

Building blocks Tehran GIS Center,Iranian NationalCartographic Center

Suburb citiesTransport networkPublic parksProtected parksRiver streamsTehran districtsLand use maps (1986, 1996,2006)Digital elevation model

(2) Socio-economicdata

Population Iranian NationalCartographicCenter, Own survey

Land priceHouse price

(3) Satellite images Landsat images (Path/Row:164/35; June 1986, July1996, July 2006, June 2011)

U.S. Geological Survey

J. Jokar Arsanjani et al. / Cities 32 (2013) 33–42 35

artificial parks and green spaces has been the primary focus of Teh-ran’s current mayor. The year 2011 was selected for the model val-idation process and used as a reference dataset for Tehran’s actualurban growth.

Methods

Monitoring and visualization of past urban growth

The land use maps of the three time frames were collected fromthe national mapping agencies and evaluated against the Landsat

satellite images and aerial photographs. The 2011 land use mapwas created by classifying a June 2011 Landsat image using a max-imum likelihood algorithm. Several misclassifications were de-tected on the maps and subsequently updated. The accuracy ofthe land use maps for 1986, 1996, 2006, and 2011 was assessedusing the Kappa index calculation (Pontius, Huffaker, & Denman,2004; Pontius & Schneider, 2001) and validated at 91%, 88%, 90%,and 87%, respectively. The land use maps, which consist of six landtypes, such as open land, farmland, water body, residential, non-residential, and public parks, were reclassified into binary mapsthat represented only built-up and non-built cells. The urbangrowth over time is shown in Fig. 2.

Between 1986 and 2006, built-up areas were spreading primar-ily in three directions: (a) southwest of Tehran along the Tehran–Islam Shahr road, where Islam Shahr, Shariar, and Nasim Shahrare located; (b) west of Tehran along the three Tehran–Karaj high-ways (i.e., the Tehran–Karaj highway, the Karaj–Special highwayand the Karaj–Old highway), where Karaj (the second core of theTehran metropolitan area) is located; and (c) northwest, wheremuch open land with affordable prices existed in the vicinity ofmountains areas. Additionally, a number of built-up areas couldbe noted along the exit highways to nearby cities, specificallyTehran–Varamin, Tehran–Qom, and Tehran–Firoz koh (Fig. 2).

Agent-based modeling

In this paper, an ABM is developed that combines threedifferent types of agents that govern urban growth. The overallworkflow is shown in Fig. 3.

In their study on Guangzhou, China, Li and Liu (2007) recog-nized three main agents: the resident agents, the government agents,

Fig. 2. Spatial pattern of urban growth within 1986–1996 and 1996–2006.

36 J. Jokar Arsanjani et al. / Cities 32 (2013) 33–42

and the developer agents. In developing countries, such as Iran, be-cause of rapid population growth and migration to large cities,massive housing projects are being constructed by land developers.The developments are controlled by the national governmentaccording to approved land policies and development plans (Sar-kheyli, Sharifi, Rafieian, Bemanian, & Murayama, 2012; Zamani &Arefi, 2013). Initially, to be consistent with the available datasets,the study site was gridded into 30 m cells. Moreover, it was as-sumed that in each cell, the three agents co-exist as the main ac-tors of land use change. Thus, a cell is only prone to develop ifthe probability of development by the resident agent and thedeveloper agent is high, provided that the government agent al-lows construction in this cell.

Corresponding variables for resident agents

Two different types of resident are distinguished: individualswho relocated from other cities into the study area to live and cur-rent residents who relocate for personal and economic reasons. Inboth categories, the behavior of the residents influences the type ofchange and the behavior of the developer agents. Furthermore, thisresident fluidity can affect the investment plans of the developeragents. These resident agents and their interactions with the devel-oper agents are the primary factors in the creation of new urbangrowth (Hu & Lo, 2007). Based on significant growth-driving forcesdetermined by Jokar A. et al. (2013), including infrastructure acces-sibility and high density residential areas, the impact of each var-iable was computed using an analytical hierarchy process (AHP),as described by Saaty (1977). The AHP provides weights bycross-comparing all factors against one another based on a 9-pointrating scale. A verification of the weighting schema was performedusing the consistency ratio (Saaty, 1999) to evaluate the logicalconsistency of the weighting results. To ensure accurate weighting,

Malczewski (1999) recommends values below a threshold value of0.1. The results demonstrate that steepness and accessibility toroad networks and railways influence the preferences of residentsin terms of livability. These variables were adopted for the ABMbased on the resulting weights, where higher weight values reflectmore important factors. Subsequently, a multi-criteria analysis(MCA) was performed to determine the cells suitable for develop-ment based on the probability of change.

A probability surface (Fig. 4) was formed, which identifies thepreferable residence locations of the resident agents.

Governing factors on developer agents

Property developers are the second major component and haveconsiderable influence on urban growth in Tehran. To facilitate theconstruction of a massive housing project, housing companies areprovided loans by the government through domestic banks to in-vest in housing, which eases and accelerates the provision of on-demand accommodation for citizens. The primary role of thedevelopers is to maximize profit by considering the preferencesof existing and potential residents with respect to purchasing orrenting property and government policies on land resource super-vision. Thus, the criterion of maximizing commercial gain is usedto assess the behavioral decisions of the developer agents (Li &Liu, 2007).

Eq. (1) was employed to compute the investment profit:

DtProfit ¼ Ht

price � Ltprice � Dt

cost ð1Þ

where DtProfit represents the investment profit, Ht

price the housingprice, Lt

price the land price, and Dtcost the development cost. Thus,

the probability of development by the developer agents can be rep-resented as follows:

Fig. 3. Conceptual model to simulate urban growth patterns.

Fig. 4. Probability surface of growth generated by the resident agents. Values around 1 indicate more suitable places for settlements, while values around 0 refer to lesssuitable ones.

J. Jokar Arsanjani et al. / Cities 32 (2013) 33–42 37

Ptdeveloperðk; ijÞ ¼

Dtprofit � Dtprofit

Dmprofit � Dtprofitð2Þ

where Ptdeveloperðk; ijÞ defines the development probability related to

the developer agents, Dtprofit is a threshold value that identifies the

minimum anticipated benefit from the investment that developerscan expect to achieve, and Dmprofit is the maximum value of theinvestment profit. Developer agents will invest in a site if the esti-mation is in favor of the development according to Eq. (2) (Li &Liu, 2007). Therefore, the sub-model of the developers inputs the

38 J. Jokar Arsanjani et al. / Cities 32 (2013) 33–42

formulas described above and generates the probability of urbangrowth by this agent. The investment profit is computed for eachpixel of the case study.

Fig. 5 depicts the results when developer agents concentratetheir interests on property construction. However, any develop-ment is subject to approval by a government agent, who has thefinal decision in allowing such developments. Recently, govern-mental restrictions have inhibited new construction in stipulatedareas, and the agents must consider this fact.

Governmental supervision via government agents

The local government i.e., municipal government can prohibitany change from a certain land use type to another use (ZanganehShahraki et al., 2011). For instance, the unapproved change offarmland to a built-up area carries severe penalties. In thisresearch, this mechanism is identified as a government agent. Agovernment agent has the authority to prohibit construction inany area to maintain sufficient control over urban expansion inaccordance with national land policies. Furthermore, the authoritycan reserve certain public areas for governmental use. Moreover,government agents must consider the suitability of any residenceor construction based on the current land use, for example, the sur-rounding environment, transportation supplies, affordable generalfacilities, and educational benefits. Moreover, no construction ispermitted in steep areas or near stream networks (Jokar A. et al.,2013; Jokar A, Helbich, & Mousivand, in press). Thus, an applicationwill only be approved when no conflict exists between present andplanned land use. Additionally, government decisions can be sub-ject to unforeseen factors, such as a high migration rate, unex-pected population growth, highway interchange expansion, andincreased demand for new development plans. However, a

Fig. 5. Probability surface of potential development produced by developer agents. Valueones.

government agent must also consider the demands of residentswith respect to where they wish to live. Although local authorityplanning could be entered into the government agent sub-model,this data type is largely inaccessible. Therefore, expert inferenceswere used considering the following factors: river streams riskzone, roads network buffer, highways buffer, airports risk buffer,military facilities risk zone, power facilities risk zone, parks buffer,and non-suitable slope. Notably, the behavior of governmentagents is a function restricted by the previously mentioned compo-nents, even though government agents are responsible for approv-ing restrictions on development and risk zone identification. Thesevariables were converted to a Boolean image that representsrestricted areas and non-restricted areas. Challenges betweenresident agents and developer agents are not the final step in thedecision-making process in land development. Because the govern-ment agent is the final decision-maker in planning application andconstruction approval, the government agent also has the right topreserve any area for governmental construction and in certaincases can approve applications or reject them and prohibit con-struction. Fig. 6 presents a binary surface that indicates the latestpolicies issued by the government and taking into account possibleprohibited locations that may be considered for applications foracceptable and legal building developments.

Computing urban growth demand

According to Torrens and Alberti (2000) and Torrens (2006), todetermine how many cells will be converted to built-up cells,urban growth modeling must consider, on the one hand, anallocation process of urban expansion and, on the other hand, thequantification of change. In the ABM, a module is designed to ac-count for these requirements by estimating the quantity of future

s around 1 indicate more attractive sites, while values around 0 refer to less suitable

Table 2Estimation of urban growth in hectare extracted through the Markovian transitionarea matrix for 2011, 2016 and 2026.

Year Conversion Built-up area (ha) Non-built area (ha)

2011 Built-up 58,721 380Non-built 3267 124,034

2016 Built-up 58,355 746Non-built 6413 120,888

2026 Built-up 58,420 681Non-built 13,342 113,959

J. Jokar Arsanjani et al. / Cities 32 (2013) 33–42 39

change (Fig. 3). The developed ABM starts by identifying the mostprobable cells for development. Then, the model estimates thequantity of change and consequently allocates this quantity inthe entire image (from the most probable cell downward). Oncethe estimated quantity is reached, the ABM stops the allocationprocess. The quantity of change can be computed by means of aMarkov chain model and statistical extrapolation, as discussed inthe next section.

Approach 1: Markov chain model

The Markov chain model is a stochastic progression thatdescribes the probability of conversion of one state to another(Cabral & Zamyatin, 2009). The model generates two tables: aso-called transition probability matrix, which determines theprobability of change from one category (e.g., built-up) to anothercategory (e.g., non-built), and a so-called transition area matrix,which presents the number of cells of one land type to be alteredto another land type (Eastman, 2009). Because the mathematicalmodel does not entail a geospatial output, the model must beintegrated into a spatial-explicit model for land use planning.As a first approach, change demand is calculated through theMarkov chain to obtain predicted land developments for 2011.Table 2 shows the expected change demand through the Markovmodel.

The table shows that by 2011, nearly 3300 ha of non-built cellswill be converted to built-up cells. Although 380 ha of built-upcells are expected to be changed to non-built cells, it is assumedthat this expectation is an artifact of land use maps accuracy be-cause to the best of the authors’ knowledge, Tehran’s urbanizationis a one-way process, and a building can rarely be found that issubject to conversion to an artificial public park.

Fig. 6. Constraints on development imposed by g

Approach 2: statistical extrapolation

Statistical extrapolation (Munroe & Mueller, 2007) is an alterna-tive method to estimate land change demand. This method calcu-lates the prospect index of construction per capita. The averagevalue of this index can be calculated to input the equation, forexample, to incorporate the predicted population for 2011, 2016,and 2026 (as provided by the Iranian Statistic Center). Table 3 liststhe index results for the previous time frames (i.e., 1986, 1996, and2006). Thus, it is feasible to predict the anticipated quantity ofchange, by inputting the average construction per capita indexfor the next steps (e.g., 2016). Therefore, the future population isstatistically predicted to manipulate the expected quantity ofchange based on socio-economic data. In Table 3, the estimatedquantity of the resident population in the study area and the con-struction per capita index are shown.

Spatial allocation of urban growth

The generated probability surface only demonstrates wheregrowth is more probable. Therefore, the growth must be allocated

overnment regulations (crosshatched areas).

Table 3Statistical extrapolation of construction per capita index for 2016 and 2026.

Year Residentspopulation(person)

Occupied area(i.e., built-up) (ha)

Constructionper capita(person/ha)

1986 6,257,713 44,772 139.771996 7,024,295 53,056 132.392006 8,154,691 59,095 137.992011 9,047,827 66,178 136.722016 9,940,964 72,711 136.722026 11,553,771 84,508 136.72

Table 4Kappa indices of simulated map of urbandevelopments for 2011 for each approach.

Approach Kappaindex

Markov chain model 0.8463Statistical

extrapolation0.8241

40 J. Jokar Arsanjani et al. / Cities 32 (2013) 33–42

within a space. The higher probability values represent a higherchance of urban growth. To this end, a cellular automata functionallocates the land developments according to the change demandvalue and stops when the predefined quantity of change is reached.The function initiates the allocation process from the nearest cells(White & Engelen, 2000) using a 3 � 3 neighborhood size and thevon Neumann (Wolfram, 1994) transition rule, as recommendedby Jokar A. et al., 2011).

Fig. 7. Simulated urban grow

Model calibration, evaluation, and growth simulation

Before the model is used for predictions, validation and statisti-cal evaluation are necessary.

The model is run to predict urban growth for 2011, for whichthe actual growth has already been computed. A 2011 urban devel-opment map is extracted from the Landsat data and statisticallycross-compared with the simulation results for 2011 in the previ-ously mentioned approaches (i.e., the Markov chain model and sta-tistical extrapolation). An overall Kappa index (Pontius &Schneider, 2001; Pontius et al., 2004) of built-up and non-built pix-els is calculated (Table 4).

According to Landis and Koch (1977), a Kappa index larger than0.8 refers to nearly perfect accordance. The results imply that theMarkov chain model performs slightly better than the statisticalextrapolation of demographic data to quantify growth. Therefore,the Markov model is used to predict urban growth for the comingyears. Thus, using previously generated transition area growthmatrices for 1996–2006 and 1986–2006, the urban growth mapsof 2016 and 2026 are simulated (Fig. 7).

Discussion

Three major agents were classified to model urban growththrough ABM: resident agents, property developer agents, and gov-ernment agents. These agents generate major effects: from an indi-vidual decision to a collective outcome. Developer agents follow aprofit-driven agenda to construct new housing, whereas residentagents make decisions based on preferential choices according tomultiple variables. The developer agents can guide prospective res-idents to choose properties based on economic factors, locationand infrastructure that are favorable for both parties. However, if

th for 2016 and 2026.

J. Jokar Arsanjani et al. / Cities 32 (2013) 33–42 41

demand for and interest in selecting a neighborhood in which tolive increase, property prices increase sharply. The price increaseforces developer agents to construct new and concentratedhousing, with the inevitable consequence of neighborhoodoverdevelopment. Thus, if the purchase and rent prices exceedthe affordable threshold of the residents, the residents are com-pelled to seek residence elsewhere. Thus, property developersmust change their investment strategies and planning accordingto the behavior of residents without compromising profit. How-ever, in this three-tier interactive system, the government agentremains the most powerful and must approve all constructionplans. The government is the final decision-maker and makes deci-sions based on environmental circumstances and internal policies,whereby an economic rationale can also be considered.

A visual analysis of urban growth from 1986 to 2006 demon-strated that three main transport corridors were extending in thefollowing manner: (i) toward the west zone of the greater metro-politan area, (ii) toward the southeast zone of the greater metro-politan area, and (iii) toward the northwest part of Tehran. Thistransport corridor extension is the result of environmental condi-tions and the accessibility of Tehran’s north-south and west-eastbelt highways. Moreover, the new developments are occurringaround the city’s main core. However, Tehran is bounded by the Al-borz Mountains in the north, the Sorkhe-Hesar national park in theeast, and farmland in the southeast. Therefore, outward-orientedexpansion to Tehran’s south and west is expected. Conversely,the building height is increasing despite the city’s size.

The range of the longitudinal slope varies from 900 to 3700 m.Therefore, altitude and steepness are key factors in guiding growth,although in the northern part of the study area, where the high Al-borz Mountains are located, urban expansion is impossible (exceptfor recreation areas). As the capital of a developing country thathas experienced a diverse range of events during the last three dec-ades, such as political revolution, an 8-year, highly destructive war,and rapid physical and economic development, Tehran has alwaysbeen an attractive goal for migrants. Therefore, adjusting an opti-mal urban growth modeling technique is difficult because admin-istrative resolutions and land policies are not stable. The growth isongoing and accompanied by the problematic issues of air pollu-tion and chaotic traffic jams. Obviously, the accessibility of andproximity to infrastructure are significant determinants of Tehran’surban growth pattern, as in other cities, as studied by, e.g., Chengand Masser (2003), Hu and Lo (2007), and Poelmans and vanRompaey (2009). The examination of urban growth until 2026demonstrates that growth will occur along highways towardTehran’s south and southwest. The simulated future growth until2026 reveals that the expected developments will occur in the sur-rounding cells of the built-up areas in eastern, southwestern andnorthwestern Tehran, where land remains available for develop-ment. These areas are predicted to undergo development becausethey are located along Tehran’s exit corridors. Specifically, a newhighway to the Caspian Sea is under construction and has attractedsubstantial attention from developers and potential residents.

Conclusions

Recently, the developing world has experienced urban expan-sion on an unprecedented scale, which has had a substantial im-pact on intensive land use (Habibi & Asadi, 2011; Sintusingha &Mirgholami, 2013). Therefore, the simulation and forecast of urbangrowth patterns is an essential task for urban planners and landconservationists to formulate sustainable development strategies.The simulation of built-up development can afford helpful andvaluable information regarding future land demand and landscapechanges. There is a large and increasing quantity of research that

uses bottom-up techniques, such as cellular automata and ABM,to simulate urban expansion. The primary problem in using cellu-lar automata models is the inability to integrate human, social andeconomic factors and to thus produce accurate simulation.However, ABM is better suited to accommodate these obstacles(Crooks, 2006). In this research, the corresponding socio-economicvariables and biophysical parameters were incorporated using anindividual behavior modeling approach to predict the most proba-ble development sites. The added value of local analysis among thevarious methods available in geographic information systems ex-tends to the possibility of retrieving information on urban futures.Whereas local analysis permits an understanding of spatial interac-tions on larger scales, smaller scales may benefit significantly fromthese models. From a decision-making perspective, complex sys-tems and ABMs share an unparalleled quantity of tacit knowledge,which otherwise would be unpredictable in linear modeling. Urbangrowth models may also benefit substantially from the interac-tions of the behavior of individuals on different scales, such as re-gions, whereas megacities tend to be the cornerstone of a betterspatial understanding of urban growth dynamics. In summary,such a technique can be applied to similar dynamic regions wheregovernment constraints exist and the behavior of individuals mustbe considered.

Given the importance of this analytical approach, we recom-mend further improvement in data acquisition and analysis. Forexample, in this study, land and housing price data played a majorrole in investigating the behavior of developer agents. However,the certainty of this data is questionable because their sourcewas not a reliable domestic administrative organization. Instead,the data were collected in interviews and from non-formal datasources, which affects the results. Additionally, as critical feedbackon the model performance, the allocation process allocated theurbanized cells globally from the most probable cell to the least,which could neglect the model’s localization. Because low-priceddistricts, which are located outside of Tehran and benefit the devel-oper agents, are omitted in the allocation process, these districtsare developed by local developers, who are confident of earninga profit. An overall recommendation for further studies on Tehran’surban growth is to consider urban growth in the greater metropol-itan area and to study all of the areas affected by urban growth.

Acknowledgments

The authors acknowledge the constructive comments of theanonymous reviewers and the associate editor, which helped im-prove the paper. Furthermore, the authors thank the Geoinformaticresearch group, University of Heidelberg, for providing resourcesfor this research. Jamal Jokar Arsanjani and Marco Helbich werefunded by the Alexander von Humboldt Foundation.

References

Bakker, M. M., & van Doorn, A. M. (2009). Farmer-specific relationships betweenland use change and landscape factors: Introducing agents in empirical land usemodelling. Land Use Policy, 26(3), 809–817.

Barredo, J. I., Demicheli, L., Lavalle, C., Kasanko, M., & McCor mick, N. (2004).Modelling future urban scenarios in developing countries: An application casestudy in Lagos, Nigeria. Environment and Planning B: Planning and Design, 31(1),65–84.

Batty, M. (2008). The size, scale, and shape of cities. Science, 319(5864), 769–771.Bhatta, B., Saraswati, S., & Bandyopadhyay, D. (2010). Urban sprawl measurement

from remote sensing data. Applied Geography, 30(4), 731–740.Cabral, P., & Zamyatin, A. (2009). Markov processes in modeling land use and land

cover changes in Sintra-Cascais, Portugal. Dyna rev.fac.nac.minas, 76(158),191–198.

Cheng, J., & Masser, I. (2003). Urban growth pattern modeling: A case study ofWuhan city, PR China. Landscape and Urban Planning, 62(4), 199–217.

Clarke, K. C., Hoppen, S., & Gaydos, L. (1997). A self-modifying cellular automatonmodel of historical urbanization in the San Francisco Bay area. Environment andPlanning B: Planning and Design, 24, 247–261.

42 J. Jokar Arsanjani et al. / Cities 32 (2013) 33–42

Crooks, A. (2006). Exploring cities using agent-based models and GIS. Working/discussion paper. <http://eprints.ucl.ac.uk/3341/> (Retrieved 06.11.12).

Crooks, A., Castle, C., & Batty, M. (2008). Key challenges in agent-based modellingfor geospatial simulation. Computers, Environment and Urban Systems, 32(6),417–430.

Eastman, J.R. (2009). IDRISI Taiga. Worcester, MA: Clark University.Fang, S., Gertner, G. Z., Sun, Z., & Anderson, A. (2005). The impact of interactions in

spatial simulation of the dynamics of urban sprawl. Landscape and UrbanPlanning, 73(4), 294–306.

Fujita, M., & Kashiwadani, M. (1989). Testing the efficiency of urban spatial growth:A case study of Tokyo. Journal of Urban Economics, 25(2), 156–192.

Habibi, S., & Asadi, N. (2011). Causes, results and methods of controlling urbansprawl. Procedia Engineering, 21, 133–141.

Haibo, Y., Longjiang, D., Hengliang, G., & Jie, Z. (2011). Tai’an land use analysis andprediction based on RS and Markov model. Procedia Environmental Sciences (PartC(0)), 2625–2630.

Halleux, J.-M., Marcinczak, S., & van der Krabben, E. (2012). The adaptive efficiencyof land use planning measured by the control of urban sprawl. The cases of theNetherlands, Belgium and Poland. Land Use Policy, 29(4), 887–898.

He, C., Okada, N., Zhang, Q., Shi, P., & Zhang, J. (2006). Modeling urban expansionscenarios by coupling cellular automata model and system dynamic model inBeijing, China. Applied Geography, 26(3–4), 323–345.

Helbich, M. (2012). Beyond postsuburbia? Multifunctional service agglomeration inVienna’s urban fringe. Journal of Economic & Social Geography, 103(1), 39–52.

Helbich, M., & Leitner, M. (2010). Postsuburban spatial evolution of Vienna’s urbanfringe: Evidence from point process modeling. Urban Geography, 31(8),1100–1117.

Hu, Z., & Lo, C. (2007). Modeling urban growth in Atlanta using logistic regression.Computers, Environment and Urban Systems, 31(6), 667–688.

Jokar A, J., Helbich, M., Kainz, W., & Darvishi B, A. (2013). Integration of logisticregression, Markov chain and cellular automata models to simulate urbanexpansion. International Journal of Applied Earth Observations andGeoinformation, 21, 265–275. http://dx.doi.org/10.1016/j.jag.2011.12.014.

Jokar A. J., Helbich M. and Mousivand A. (in press). A Morphological approach topredicting urban expansion , Transactions in GIS.

Jokar, A. J., Kainz, W., & Mousivand, A. (2011). Tracking dynamic land use changeusing spatially explicit Markov chain based on cellular automata – The case ofTehran. International Journal of Image and Data Fusion, 2(4). http://dx.doi.org/10.1080/19479832.2011.605397.

Koomen, E., Vasco, D., Dekkers, J. E. C., & Rietveld, P. (2012). A utility-based, multi-sector framework for local scale land-use modeling. In Paper presented at theworkshop on complexity modeling for urban structure and dynamics, 15th AGILEinternational conference on geographic information science, Avignon, France (April24–27).

Koomen, E., Stillwell, J., Bakema, A., & Scholten, H. J. (2007). Modelling land-usechange: Progress and applications. Springer.

Lagarias, A. (2012). Urban sprawl simulation linking macro-scale processes tomicro-dynamics through cellular automata, an application in Thessaloniki,Greece. Applied Geography, 34, 146–160.

Landis, L. R., & Koch, G. G. (1977). The measurement of observer agreement forcategorical data. Biometrics, 33, 159–174.

Li, X., & Liu, X. (2007). Defining agents’ behaviors to simulate complex residentialdevelopment using multicriteria evaluation. Journal of EnvironmentalManagement, 85(4), 1063–1075.

Longley, P. A., & Batty, M. (2003). Advanced spatial analysis: Extending GIS. In P. A.Longley & M. Batty (Eds.), Advanced spatial analysis: The CASA book of GIS(pp. 1–17). Redlands, CA: ESRI Press.

Malczewski, J. (1999). GIS and multicriteria decision analysis. New York: John Wiley &Sons.

Manson, S. M. (2005). Agent-based modeling and genetic programming formodeling land change in the Southern Yucatán Peninsular Region of Mexico.Agriculture, Ecosystems & Environment, 111(1–4), 47–62.

Matthews, R. B., Gilbert, N. G., Roach, A., Polhill, J. G., & Gotts, N. M. (2007). Agent-based land-use models: A review of applications. Landscape Ecology, 22(10),1447–1459.

Mousivand, A. J., Alimohammadi Sarab, A., Shayan, S. (2007). A new approach ofpredicting land use and land cover changes by satellite imagery and Markov chainmodel (Case study: Tehran). MSc Thesis. Tehran: Tarbiat Modares University.

Mulligan, G. F., & Crampton, J. P. (2005). Population growth in the world’s largestcities. Cities, 22(5), 365–380. http://dx.doi.org/10.1016/j.cities.2005.07.001.

Munroe, D. K., & Mueller, D. (2007). Issues in spatially explicit statistical land use/cover change (LUCC) models: Examples from western Honduras and the CentralHighlands of Vietnam. Land Use Policy, 24, 521–530.

Perez, L., & Dragicevic, S. (2012). Landscape-level simulation of forest insectdisturbance: Coupling swarm intelligent agents with GIS-based cellularautomata model. Ecological Modelling, 231, 53–64. http://dx.doi.org/10.1016/j.ecolmodel.2012.01.020.

Poelmans, L., & Van Rompaey, A. (2009). Detecting and modelling spatial patterns ofurban sprawl in highly fragmented areas: A case study in the Flanders–Brusselsregion. Landscape and Urban Planning, 93(1), 10–19.

Pontius, R. G., Jr., Huffaker, D., & Denman, K. (2004). Useful techniques of validationfor spatially explicit land-change models. Ecological Modelling, 179(4), 445–461.

Pontius, R. G., Jr., & Schneider, L. C. (2001). Land-cover change model validation byan ROC method for the Ipswich watershed, Massachusetts, USA. Agriculture,Ecosystems & Environment, 85(1–3), 239–248.

Rafiee, R., Mahiny, A. S., Khorasani, N., Darvishsefat, A. A., & Danekar, A. (2009).Simulating urban growth in Mashad City, Iran through the SLEUTH model(UGM). Cities, 26(1), 19–26. http://dx.doi.org/10.1016/j.cities.2008.11.005(Elsevier Ltd.).

Robinson, D. T., Murray-Rust, D., Rieser, V., Milicic, V., & Rounsevell, M. (2012).Modelling the impacts of land system dynamics on human well-being: Using anagent-based approach to cope with data limitations in Koper, Slovenia.Computers, Environment and Urban Systems, 36(2), 164–176.

Saaty, T. L. (1977). A scaling method for priorities in hierarchical structures. Journalof Mathematical Psychology, 15(3), 234–281.

Saaty, T. L. (1999). Decision making for leaders: The analytic hierarchy process fordecisions in a complex world, New Edition 2001 (3rd ed.). RWS Publications.

Sabet Sarvestani, M., Ibrahim, A. L., & Kanaroglou, P. (2011). Three decades of urbangrowth in the city of Shiraz, Iran: A remote sensing and geographic informationsystems application. Cities, 28(4), 320–329. http://dx.doi.org/10.1016/j.cities.2011.03.002 (Elsevier Ltd.).

Sarkheyli, E., Sharifi, A., Rafieian, M., Bemanian & Murayama, A. (2012). Aninvestigation of the reasons for non-compliance with FAR regulations inTehran. Cities, 29(4), 223–233.

Shi, Y., Sun, X., Zhu, X., Li, Y., & Mei, L. (2012). Characterizing growth types andanalyzing growth density distribution in response to urban growth patterns inperi-urban areas of Lianyungang City. Landscape and Urban Planning, 105(4),425–433.

Sintusingha, S., & Mirgholami, M. (2013). Parallel modernization and selfcolonization: Urban evolution and practices in Bangkok and Tehran. Cities.

Tayyebi, A., Pijanowski, B. C., & Pekin, B. (2011). Two rule-based urban growthboundary models applied to the Tehran metropolitan area, Iran. AppliedGeography, 31(3), 908–918.

Torrens, P. M. (2006). Geosimulation and its application to urban growth modeling (pp.119–134). London: Springer-Verlag.

Torrens, P. M., & Alberti, M. (2000). Measuring sprawl. CASA working paper 27.University College London, Centre for Advanced Spatial Analysis.

Tv, R., Aithal, B. H., & Sanna, D. D. (2012). Insights to urban dynamics throughlandscape spatial pattern analysis. International Journal of Applied EarthObservation and Geoinformation, 18, 329–343.

Vaz, E., Cabral, P., Caetano, M., Painho, M., & Nijkamp, P. (2012). Urban heritageendangerment at the interface of future cities and past heritage: A spatialvulnerability assessment. Habitat International, 36(2), 287–294.

Wang, J.-F., Liu, X.-H., Peng, L., Chen, H.-Y., Driskell, L., & Zheng, X.-Y. (2012). Citiesevolution tree and applications to predicting urban growth. Population &Environment, 33(2), 186–201.

Weber, C., & Puissant, A. (2003). Urbanization pressure and modeling of urbangrowth: Example of the Tunis Metropolitan Area. Remote Sensing ofEnvironment, 86(3), 341–352.

White, R., Engelen, G. (2000). High-resolution integrated modelling of the spatialdynamics of urban and regional systems. Computers Environment and UrbanSystems, 24, 383–400.

Wolfram, S. (1994). Cellularautomata and complexity: Collected papers. Reading, MA:Addison-Wesley.

World Gazetteer (2012). <http://www.world-gazetteer.com/>.Wu, Q., Li, H., Wang, R., Paulussen, J., He, Y., Wang, M., Wang, B., & Wang, Z. (2006).

Monitoring and predicting land use change in Beijing using remote sensing andGIS. Landscape and Urban Planning, 78(4), 322–333.

Zamani, B., & Arefi, M. (2013). Iranian new towns and their urban managementissues: A critical review of influential actors and factors. Cities, 30, 105–112

Zanganeh Shahraki, S., Sauri, D., Serra, P., Modugno, S., Seifolddini, F., & Pourahmad,A. (2011). Urban sprawl pattern and land-use change detection in Yazd, Iran.Habitat International, 35(4), 521–528.

Zebardast, E. (2006). Marginalization of the urban poor and the expansion of thespontaneous settlements on the Tehran metropolitan fringe. Cities, 23(6),439–454. http://dx.doi.org/10.1016/j.cities.2006.07.001.