Cities 31 (2013) 105–113
Contents lists available at SciVerse ScienceDirect
Cities journal homepage: www.elsevier.com/locate/cities
Agent-based modeling of urban land-use development, case study: Simulating future scenarios of Qazvin city Farhad Hosseinali a,⇑, Ali A. Alesheikh a,1, Farshad Nourian b,2 a b
Faculty of Geodesy and Geomatics Engineering, K.N. Toosi University of Technology, ValiAsr Street, Mirdamad Cross, 19967-15433 Tehran, Iran School of Urban Planning, University of Tehran, Enghelab Avenue, 14155-6135 Tehran, Iran
a r t i c l e
i n f o
Article history: Received 31 March 2012 Received in revised form 3 July 2012 Accepted 6 September 2012 Available online 6 October 2012 Keywords: Urban land-use development Agent-based modeling Development policy Qazvin
a b s t r a c t Urban land-use development is a problematic phenomenon in developing countries. Modeling this phenomenon is of considerable interest to urban planners and city managers. Several methods have been developed to simulate the dynamics of land-use changes. However, the complexity of urban growth is considered a factor that impedes the usefulness of such simulation methods. Among the available methods, those considered ‘‘agent-based models’’ have found popularity in simulating land-use development and urban sprawl modeling. These methods use a dynamic bottom-up approach with the actors in landuse development as their basic components. In this paper, a new agent-based model is introduced. This model is equipped with new methods for modeling the movements of agents and competition among agents. The model is used to simulate urban land-use development in the Qazvin province of Iran, which covers an area of 36 45 km. The model is ﬁrst calibrated with existing data and is then used to predict future land-use development. To test development policies, four scenarios are deﬁned. The ﬁrst scenario reﬂects the current pattern of development, which is evaluated using the calibrated model. The second and third scenarios examine different policies, including those that act as ‘‘incentive’’ strategies and those that are ‘‘punitive.’’ The fourth scenario focuses on changes to the demographic population of agents. The results reveal that the current trend in urban growth tends to be dispersed in the study area. However, different policies tend to produce different results: in areas in which an incentive policy is in place, 140 clusters of development were detected, while in areas in which a punitive policy is in place, 180 clusters were detected. The incentive strategy is concluded to be more successful than the punitive strategy in reducing the dispersion of development. Change in the population demography is observed to be more efﬁcient in areas of development than in those of dispersion. Ó 2012 Elsevier Ltd. All rights reserved.
Introduction As a developing country, Iran is now witnessing continual largescale urbanization (Raﬁee, Salman Mahiny, Khorasani, Darvishsefat, & Danekar, 2009). The number of towns and cities in Iran has also increased signiﬁcantly, from a total of 199 towns in 1956– 1200 in 2012 (Statistical Centre of Iran). This rapid urbanization pattern can be seen in most major cities in Iran. As an example, the city of Qazvin, an ancient city located 150 km west of Tehran, has witnessed rapid growth both in size and population in the last two decades (Housing and Urban Development Organization of Qazvin). As an industrial and agricultural city, the city’s population increased from 291,117 in 1996 to ⇑ Corresponding author. Tel.: +98 21 8878 6212; fax: +98 21 8878 6213. E-mail addresses: [email protected]
(F. Hosseinali), [email protected]
(A.A. Alesheikh), [email protected]
(F. Nourian). 1 Tel.: +98 21 8878 6212; fax: +98 21 8878 6213. 2 Tel.: +98 21 8896 2743. 0264-2751/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.cities.2012.09.002
349,821 in 2006 (Statistical Centre of Iran). During the past 50 years, a number of formal and informal settlements have also formed around the city. Our study area covers the city of Qazvin itself and ﬁve other towns as well as a number of villages and industrial settlements. The rapid expansion of residential land-use endangers agriculture and environmental resources. Hence, simulation of urban land-use change and development is vitally important for municipalities responsible for planning for the future. Simulation provides users with practical feedback when planning real-world systems (Zhang, Ban, Liu, & Hu, 2011). This allows planners to determine the suitability and efﬁciency of a plan before the plan is implemented. A simulation is deﬁned here as a process of changing one or more variables in a model and observing the resulting developments (Banks, Carson, Nelson, & Nicol, 2004). Simulation also allows planners and city managers to study a problem at several different levels of abstraction. By approaching a system at a higher level of abstraction, planners are better equipped to
F. Hosseinali et al. / Cities 31 (2013) 105–113
understand the behaviors and interactions of all of the high-level components within the system and are therefore prepared to counteract the complexity of the overall system (Santé, García, Miranda, & Crecente, 2010). Simulation of land-use development can potentially represent the consequences of current planning policies. Without using models that embrace the complexity of the urban system, it would be difﬁcult to simulate and predict the future of urban growth (Batty, 2005). Urban land-use development is one of the most profound human-induced alterations in the Earth’s system (Le, Park, Vlek, & Cremers, 2008; Vitousek, Mooney, Lubchenco, & Melillo, 1997). The changes that occur in land-use due to urbanization promote a complex process caused by the interaction between natural and social systems at different spatial scales (Rindfuss, Walsh, Turner, Fox, & Mishra, 2004; Valbuena, Verburg, & Bregt, 2008). Traditionally, two approaches have been proposed to characterize and understand these changes: (1) a bottom-up, anthropological, process-oriented approach based on household surveys and a resource base inventory and (2) a top-down, land evaluation, pattern-oriented approach based on remote sensing and census data (Geoghegan et al., 1998). Another group of models has recently emerged and gained popularity in the urban-related scientiﬁc community. These models use the real actors of land-use change (individual or institutions) as objects of analysis and of simulations, and pay explicit attention to interactions among these ‘‘agents.’’ Therefore, such models are commonly referred to as agent-based models (Castella & Verburg, 2007). Several characteristics deﬁne agents: they are autonomous, they share an environment through agent communication and interaction, and they make decisions that tie their behavior to the environment. Agents have been used to represent a wide variety of entities, including atoms, animals, cars, people, biological cells and organizations (Conte, Hegselmann, & Terna, 1997; Epstein, Axtell, & Project, 1996; Parker, Manson, Janssen, Hoffmann, & Deadman, 2003; Robinson et al., 2007; Weiss, 1999). Agents make inductive and evolving choices that move them toward achieving goals (Parker et al., 2003; Wooldridge, 2009). The aim of this research is to simulate urban land-use development using a newly developed agent-based model. Simulation of land-use development has been conducted using a variety of models, such as artiﬁcial neural networks (ANNs), cellular automata (CA) and regression (Batty, Xie, & Sun, 1999; Hu & Lo, 2007; Parker et al., 2003; Pijanowski, Tayyebi, Delavar, & Yazdanpanah, 2009). Some packaged agent-based or CA models are also available. For instance, Raﬁee and his colleagues calibrated the SLEUTH model to simulate future urban growth in Mashad, Iran (Raﬁee et al., 2009). SLUCE is another package widely used in this ﬁeld (Brown et al., 2008). Nevertheless, our efforts were directed toward constructing a more ﬂexible model able to mimic the process of land-use development in Iran. The ﬂexibility of our model may allow it to be used in other countries. We propose an approach that integrates land-use factors into an agent-based model for modeling future urban land-use scenarios. The model was calibrated with existing data and then a 20-year simulation was run, covering the period of 2010–2030. The goals of this model are to predict future land-use development under existing spatial policies to produce alternative planning and policy scenarios, and to compare these alternative scenarios in terms of their effects on future land-use development. Four scenarios are deﬁned in this paper. The ﬁrst scenario is in line with the current calibrated trend of urban development. The second scenario involves modifying the values of criterion maps that are used to input spatial data. In this scenario, ‘‘incentive’’ policies are used. The third scenario is similar to the second but the modiﬁcation of the values of criterion maps is performed using ‘‘punitive’’ measures. The fourth scenario involves using the change in the demo-
graphics of the agents. GIS, a versatile system and science for dealing with spatial data (Longley, 2005; Sabet Sarvestani, Ibrahim, & Kanaroglou, 2011), was used to help us collect spatial data, prepare proper maps and present the results. Methodology In our model, the agents represent land-use developers who move in the landscape seeking appropriate cells (i.e., parcels of land) to develop. The model mimics the mechanism of searching for and developing land in Iran. However, this may also be the case in many other countries. The agents are categorized into ﬁve groups with different aims (discussed in the next section). The development is divided into separate stages. Each stage corresponds to a year. In each year, the mobile agents search the landscape and record the state of the cells they visited (searching stage, Fig. 1). Next, they decide to develop the most appropriate cells among those that they visited (developing stage, Fig. 2). In the developing stage, some cells may simultaneously be chosen for development by more than one agent. In such cases, the agents compete and the winner of the competition develops that particular cell. The details of the model are explained below. Criteria for selection of a target location Many researchers have considered various factors representing decision-making criteria of agents to select targets for development (Matthews, Gilbert, Roach, Polhill, & Gotts, 2007; Parker et al., 2003). Nevertheless, residential decision criteria, such as the household stage in a life cycle, the price of property, the demographic structure of neighborhood, and public transportation, do not account more than 20–30% of development choice (Benenson & Torrens, 2004). Consequently, some researchers have stated that traditional micro-simulation modeling, which uses such databases extensively, does not account for interdependencies among the decision factors (Benenson & Torrens, 2004; Waddell, 2002). Thus, three criterion maps, namely, attractiveness, accessibility and land value, were used in this research (Figs. 5 and 7). The framework considers the landscape as a raster space on which agents act for developing particular cells. Each cell can be developed or not developed; therefore, each agent should assess the undeveloped areas and decide where to develop. Agent classes and their characteristics The categorization of agents takes place by considering the situations and goals of land-use developers. This categorization is similar to that performed by Loibl and Toetzer (2003), but has been modiﬁed to match the conditions of the executing environment in Iran. Thus, the agents are categorized into the following ﬁve types: 1. Young persons with moderate income who look for fairly affordable cells with good accessibility. 2. High-income developers seeking valuable land with acceptable attractiveness. 3. The afﬂuent who desire highly attractive cells for recreational residence. 4. Low-income people who search for the least expensive cells. 5. Moderate-to high-income people who consider the three criteria of land value, accessibility and attractiveness to have the same weight. In addition to its type, each agent has a location in a cell of the landscape, limited movement, a minimum required location change (jump) in a district, and a number of districts to search.
F. Hosseinali et al. / Cities 31 (2013) 105–113
Randomly select a search district (based on the probability of selection)
Selecting the region of search in the district Assessing the adjacent cells and moving to the best one N
Is the search finished in the region? Go to Developing stage
Is the search finished in the district? Y Sorting the visited cells
Should another district be searched?
Fig. 1. Flowchart of the model in the searching stage.
Choose one agent randomly
Pick the top scoring cell Register the number of request for the cell
Register a conflict
Have other agents chosen that cell? N Register the number of request for the cell
Determine winner and loser(s) of competition
Develop the cell and register the type of developer
Delete the agent from the list of agents
Increase Frustration Is there any other cell in the investment list?
Is there any unsatisfied agent?
N Register dissatisfaction
Delete the agent from the list of agents Fig. 2. Flowchart of the model in the developing stage.
Distribution of agents in the landscape At the initialization of the model, the maps are uploaded into the model, the parameters are set, and the agents are created. First, the agents must be located in the landscape. The landscape is not likely to be a homogeneous area, and different districts can often be searched. Thus, the agents begin their search by choosing a district. Districts have different chances of selection by different types of agents, based on the districts’ characteristics and the agents’ desires. The selection of a district is performed randomly, referring to the chosen probability distribution P ki for each agent, where k is the type of agent and i is the district. The cells adjacent to those that are already developed are potentially subject to urbanization. Thus, in each district, the agents ﬁrst go to the cells adjacent to currently developed areas. Once the agents ﬁnd their locations in the landscape, each agent moves around, and after a limited amount of movement, changes
its location (jumps) to another position in the same district. The agents only traverse the undeveloped cells. The agents might also change their districts and conduct the same activities in the new ones.
Agents’ movements Wherever the agent starts, it assesses the state of the current (standing) cell and that of its eight adjacent cells. ‘‘State’’ refers to the values of the three criteria in the given position. The agent then moves to the best neighboring cell, or if more than one neighboring cell has the same score, the agent chooses one of them randomly. The agent records the positions and the states of all of the cells that it traverses, as well as those of their undeveloped neighbors. Traversed cells and their undeveloped neighbors are called ‘‘visited cells.’’ By the time each agent ﬁnishes its search in the
F. Hosseinali et al. / Cities 31 (2013) 105–113
landscape, it has a list of the states and positions of the cells it has visited. This list can be considered the agent’s investment list. Decision to develop Once an agent ﬁnishes its search, it should decide which cells to develop. To do this, the agent sorts its investment list in descending order, with the most suitable cell at the top of the list. The agent then selects the top-scoring cell from its sorted investment list and develops it if there is no competitor. Competition It is highly probable that one cell is selected by more than one agent for development. Such a cell will certainly be developed. In such conditions, the conﬂict is resolved by a competition to identify the dominant agent. The winner of the competition is determined by the ‘‘scores’’ of competing agents. An agent’s score depends on the type of agent and the number of times that the agent has lost cells in previous competitions. The score is calculated using the following formula:
Score ¼ W Type ScoreType þ W Frustration Frustration
where ScoreType is the score assigned to each type of agent, Frustration is the number of times that an agent has lost a cell, and W Type and W Frustration are the weights considered for ScoreType and Frustration, respectively. The value of Frustration is equal to zero for all agents at the beginning. However, whenever an agent loses a cell in a competition, its Frustration value increases by one. This increase means that in the next competition the agent will be more likely to develop a cell. W Type , W Frustration and ScoreType are determined by experts by considering pair-by-pair conditions of the competitions among agents. Study area Fig. 3. The study area.
The study area is located in the Qazvin province of Iran. It is 45 km in length and 36 km in width (Fig. 3). The landscape is composed of 162,000 cells of 100 100 m. This area contains the city of Qazvin, the central city of the province, and ﬁve smaller towns located nearby. The area also includes several villages and industrial regions. Development has occurred mostly in lands around the city and the towns. Increasing demand for land, however, has led to development of a number of informal settlements. Such developments, which negatively affect agricultural and natural resources, are major challenges faced by the planners in Qazvin. Fig. 4 shows a developing area to the west of Qazvin that is legally zoned for residential development.
area of current development in that district (between 2005 and 2010) by the total area of development in the study area (Table 1). The development areas were detected using 2005 and 2010 landuse/land-cover maps obtained from the National Cartographic Center (NCC) of Iran. Next, the residential areas and other land-uses were detected. We used ArcGIS 9.3 (ESRI, 2011) for preparing the maps, conducting the analysis and representing the results. In addition, NetLogo 4.1 (Wilensky, 2009) with its GIS extension was used to develop our agent-based model.
Setting the parameters Data preparation Three criterion maps (layers), namely, land value, attractiveness, and accessibility, were used in this study. Fig. 5 shows two such maps. The land value map was generated by taking into account land price, availability of land for development (development plan map), slope, and soil quality. The accessibility map was generated by evaluating the shortest time to reach a cell from the nearest city or town. To produce the attractiveness map, proximity to green zones, views of the city and the local temperature were considered. All of the maps were normalized to have values between 0 and 1. The map of the districts, which reﬂects the probability of development for each district, was also used. The area is divided into 12 districts by the local people. To produce the map, the probability of development for each district was calculated by dividing the total
The parameters of the model and the values to which they were set before running the model are presented in Table 2. To set values of the parameters, we used three approaches: utilizing expert knowledge, using existing data, and testing the various conﬁgurations of the model. The weights given to districts were speciﬁed by experts. The experts were also asked to determine the scores of the agents (Eq. (1)). To accomplish this, the experts were asked to consider the conditions of competition among various types of agents. As a result of this exercise, values of 9 and 10 were determined for W Type and W Frustration , respectively. The number of agents for each type had to be determined as well. First, we detected a total of 1200 ha of area developed between 2005 and 2010 in our study area. Based on land value, accessibility, and attractiveness of development, 110 ha were taken as
F. Hosseinali et al. / Cities 31 (2013) 105–113
Fig. 4. Construction to the west side of Qazvin which is zoned for residential development.
Fig. 5. Maps of accessibility (left) and attractiveness (right).
Table 1 The probability of districts. District Probability (%)
Table 2 The ﬁnal parameters of the model. Type of agent Count Weight of accessibility Weight of attractiveness Weight of land value Number of searching districts Number of traverse cells Number of jumps in each district ScoreType
1 110 3 1 2 9 12 2 2
2 90 1 2 3 9 11 2 5
3 10 1 2 1 9 13 2 5
4 20 2 1 1 9 11 2 1
5 10 1 1 1 8 10 1 3
developments for type 1, 90 ha for type 2, 10 ha for type 3, 20 ha for type 4 and 10 ha for type 5 in each year. In this study, we assumed that each agent is able to develop up to one hectare each
year. To determine the other parameters, type 5 among our agents was considered the reference agent. We assumed that agents can search 50%, 70%, or 90% of the districts. Furthermore, the number of traversed cells was considered to be either10 or 15 times more than that of the developing cells. These values are presented in Table 2 for each type. Finally, the number of jumps was assumed to be equal to the number of cells each agent develops each year. The number of jumps is one more for other types than for the reference type. To compare the results, the Kappa coefﬁcient was used (Pijanowski, Pithadia, Shellito, & Alexandridis, 2005; Tian, Ouyang, Quan, & Wu, 2011). The results show that when the number of traversed cells is 10 times of the number of cells that each agent develops per year, a regular pattern is generated. The highest Kappa is realized when the agents search 70% of the districts. Consequently, the parameters listed in Table 2 were set for the model.
F. Hosseinali et al. / Cities 31 (2013) 105–113
The model is thus calibrated using the parameters. In calibration, data for year 2005 were entered into the model. Next, the results of the simulation conducted using the model were compared with data for the year 2010. We computed the Kappa statistic based on the calibrated model. The value of Kappa was 0.77, which means that there is a good agreement between the results from the model and the observed data (Tian et al., 2011). After the calibration, the model was used to simulate future land-use development using the 2010 data. By forecasting the population in 2030 and considering the current trend of urban sprawl in the study area, we evaluated 2600 hectares of urban land-use development for 2030. Therefore, the numbers of agents of type 1 to type 5 were estimated as 60, 45, 5, 15 and 5, respectively.
Deﬁning the scenarios Four scenarios were deﬁned in this research. The ﬁrst scenario, referred to here as ‘‘the status quo,’’ simulates urban land-use development using the calibrated model. In this scenario, the current 2010 data, along with criterion maps, are used, and the landuse development for 2030 is simulated. The other scenarios take into account various plans and policies for the future. Here, it is assumed that the comprehensive and detailed plans prepared and used by organizations such as the municipalities and the Ministry of Housing and Urban Development affect the future development of the study area. For example, change in land and construction regulations lead to less or more development. Thus, these factors
Fig. 6. The zones created by the incentive (left) and punitive (right) measures.
Fig. 7. The land value maps. Top left: for incentive scenario, top right for punitive scenario and bottom for status quo and reconﬁguration scenarios.
F. Hosseinali et al. / Cities 31 (2013) 105–113
Fig. 8. Results of the model in scenarios 1–4 from top to the bottom. Left: probabilities of development, right: developments with probability of more than 50%.
had to be considered in our criteria maps. The criteria maps show that changing the land value by changing regulations is the least costly and perhaps the fastest method for creating incentive for
development. A land value map is based on the legal, physical and economic conditions of the area. Thus, two scenarios are designed based on changing the regulatory status of the study area
F. Hosseinali et al. / Cities 31 (2013) 105–113
because any changes in the rules and regulations results in change in the land values. The regulatory status indicates the legal conditions that are considered in the development plan map. Thus, based on the comprehensive plan for Qazvin, the zones that are of greater priority for development should have higher land values. In this research, approximately 2600 hectares of such land were identiﬁed for the ‘‘incentive’’ scenario (Fig. 6), and their values in the development plan map increased by 50% to produce a new land value map. Referring back to the results of the ‘‘status quo’’ scenario, some zones here experienced undesirable and scattered developments. Therefore, in our third scenario, we assumed that these sites (Fig. 6) are protected by the municipalities from development using more serious control measures. Such measures reduce the values of such zones in the development plan map by 50%. Thus, a new land value map is generated using this ‘‘punitive’’ scenario. In our fourth scenario, we assumed that the demographic proﬁle of developer agents is transformed. Our assumptions are that the economic conditions improve and that the average ages of the developers increase. Thus the population of agents of types 1 and 4 is reduced, and more agents are categorized as types 2, 3 and 5. Therefore, the numbers of agents of types 1–5 are 70, 30, 10, 10 and 10, respectively. This scenario is called the ‘‘reconﬁguration’’ scenario. The land value map used by the scenarios is illustrated in Fig. 7.
Results and discussion Because of the stochasticity of our agent-based model, the model was run ten times for each scenario. Each cell may be developed in no runs or in one or more runs of the model. Hence, each cell has a probability of development that varies from 0% to 100%. The pat-
tern and direction of simulated developments can be considered a measure of the quality of our model. The most compact developments are desired. To measure the dispersion of developments, we counted the number of clusters of development. A cluster is deﬁned by adjacent developed cells. Fig. 8 shows the maps produced for the four scenarios. The left sides present the probabilities of development, while the right sides show only the cells that achieved more than 50% development. To perform a more precise evaluation of the results, Fig. 9 was generated to show two numerical results. In scenario 1, which represents the continuation of current patterns of development without change, dispersed developments occur mostly to the east of the city of Qazvin. This is more evident in the development probability map. The results show that the towns, villages and residential areas to the east of the city will be developed such that they will connect in the near future. Scenario 2 reveals that incentive policies are fairly successful in reducing scattered developments and directing development toward pre-determined zones. However, not all of the 2600 ha considered pre-determined zones for development are attractive enough to the agents. While in the incentive scenario, 140 clusters of development were detected, there were 180 clusters detected in the punitive scenario (Fig. 9). Therefore, punitive policies can be considered less successful in giving direction to development than incentive policies. Moreover, some developments are still observed in protected zones. Fig. 9 also shows that the reconﬁguration scenario does not have any signiﬁcant effect on the dispersion of development, although it has a slight effect on the direction and positions of developments (Fig. 8). Dispersed developments exist in all scenarios. Such developments occur mostly to the northeast of the city. To prevent such developments, more serious policies must be adopted. The incen-
Number of cells with probability of 50%
Number of cells with probability of 100%
Clusters with probability of more than 50%
Fig. 9. Numerical results of four scenarios, the numbers at the based of the columns correspond to the number of scenarios.
Table 3 The deﬁned scenarios and their observed results. Scenario
Observed results from simulation
Continuing the current situation of urban development Adopting policies to encourage the developments in the desired zones Adopting policies to prevent the developments in the undesired zones Improvement in economic conditions
Using current land-use developments in the year 2010 and simulating the developments for the year 2030 Changing the land value map by adding 2600 ha of desired zones
Dispersed developments in the region especially to the east of the city of Qazvin
Changing the land value map by adding undesired zones Changing the demographic proﬁle of developers
Decrease in number of development clusters which means more compact developments. Orientation of development toward the encouraged zones Less number of clusters than the ‘‘status quo’’ scenario but more than the incentive scenario. Developments are still observed in the protected zones Insigniﬁcant inﬂuence on compactness of development. The developments are slightly oriented toward regions with better climate
F. Hosseinali et al. / Cities 31 (2013) 105–113
tive scenario is the most successful in reducing dispersed developments. This means that incentive policies may be more efﬁcient than punitive policies. It should also be noted that, if the economic indices of the region improve, development will be biased toward those areas with a better climate, such as those north of Qazvin. Fig. 9 as a numerical result completes Fig. 8. It shows that the incentive scenario is more convergent in results than the punitive scenario (the left diagram). We conclude that the inﬂuence of environment on the pattern of land-use development is greater than the inﬂuence of the demographic proﬁle of the developers. It can also be seen that many developments will occur near the towns and villages around the city of Qazvin. Those developments, however, have a tendency to be dispersed. Both incentive and punitive policies should be used to manage those developments. Moreover, the general direction of development is to the north. When the demographic proﬁle of developers changes in scenario 4, the direction of development to the north is intensiﬁed. This is attributed to the population of agents that pay more attention to a location’s attractiveness and to the better climate in the north. Table 3 summarizes the scenarios and their observed results. Other strategies and policies, such as constructing new infrastructure, can also lead to development. For instance, constructing parks, forests and artiﬁcial lakes can increase the attractiveness of regions, which results in attracting more people. Constructing appropriate road and street networks is another important way for encouraging development toward the planned areas. Conclusions and recommendations Urban growth is a prevalent challenge in many countries. Rapid growth causes unexpected changes in the land-use of regions around cities, which can endanger the environment and natural resources. A potential way to mitigate these threats is to develop more efﬁcient urban growth strategies in which spatial planning is used to limit the effects of undesirable development. Simulation is a popular approach to testing such policies and strategies. In this study, we developed a new agent-based model for simulating future urban land-use development in our study area, located in the Qazvin province of Iran. The model was calibrated with historical data from 2005 to 2010. The urban development for the year 2030 was then simulated based on our calibrated model. To test various development strategies, four scenarios were deﬁned. The ﬁrst scenario mimicked the existing pattern of development simulated by the calibrated model. In the second and third scenarios, we examined reward and punishment strategies. The fourth scenario focused on changes in the demographic population of developer agents. In the ﬁrst scenario, dispersed development was detected. The second scenario revealed that an incentive policy, applied through increasing land value, is successful in reducing dispersed development. The third scenario, in which a punitive policy was modeled, achieved the same goal as the second one. However, the results show that punitive and protective policies are less successful in preventing dispersed development than incentive policies. It was also concluded that changes in the demographic proﬁle of the agents have little effect on the extent of dispersion of development, although demographic changes did cause modiﬁcations in the direction of development. Therefore, we suggest the use of both incentive and punitive policies to reduce dispersed development around the city and to encourage development toward planned areas. References Banks, J., Carson, J., Nelson, B. L., & Nicol, D. (2004). Discrete-Event System Simulation (4th ed.). Prentice-Hall.
Batty, M. (2005). Agents, cells, and cities: New representational models for simulating multiscale urban dynamics. Environment and Planning A, 37, 1373–1394. Batty, M., Xie, Y., & Sun, Z. (1999). Modeling urban dynamics through GIS-based cellular automata. Computers, Environment and Urban Systems, 23, 205–233. Benenson, I., & Torrens, P. M. (2004). Geosimulation: Automata-based modeling of urban phenomena. John Wiley & Sons. Brown, D. G., Robinson, D. T., An, L., Nassauer, J. I., Zellner, M., Rand, W., Riolo, R., Page, S. E., Low, B., & Wang, Z. (2008). Exurbia from the bottom-up: Confronting empirical challenges to characterizing a complex system. Geoforum, 39, 805–818. Castella, J.-C., & Verburg, P. H. (2007). Combination of process-oriented and patternoriented models of land-use change in a mountain area of Vietnam. Ecological Modelling, 202, 410–420. Conte, R., Hegselmann, R., & Terna, P. (1997). Simulating social phenomena. Berlin, Germany: Springer. Epstein, J. M., & Axtell, R.Project. (1996). Growing artiﬁcial societies: Social science from the bottom up. Brookings Institution Press. ESRI (2011). ArcGIS 9.3: The complete enterprise GIS. Geoghegan, J., Pritchard, L. P., Ogneva-Himmelberger, Y., Chowdhury, R. R., Sanderson, S., & Ii, T. B. L. (1998). ‘‘Socializing the pixel’’ and ‘‘pixelizing the social’’ in land-use/cover change. In D. Liverman, E. F. Moran, R. R. Rindfuss, & P. C. Stern (Eds.), People and pixels: Linking remote sensing and social science (pp. 51–69). Washington, DC: National Research Council. Housing and Urban Development Organization of Qazvin. . Hu, Z., & Lo, C. P. (2007). Modeling urban growth in Atlanta using logistic regression. Computers, Environment and Urban Systems, 31, 667–688. Le, Q. B., Park, S. J., Vlek, P. L. G., & Cremers, A. B. (2008). Land-Use Dynamic Simulator (LUDAS): A multi-agent system model for simulating spatio-temporal dynamics of coupled human–landscape system. I. Structure and theoretical speciﬁcation. Ecological Informatics, 3, 135–153. Loibl, W., & Toetzer, T. (2003). Modeling growth and densiﬁcation processes in suburban regions—simulation of landscape transition with spatial agents. Environmental Modelling & Software, 18, 553–563. Longley, P. (2005). Geographic information systems and science. Wiley. Matthews, R., Gilbert, N., Roach, A., Polhill, J., & Gotts, N. (2007). Agent-based landuse models: A review of applications. Landscape Ecology, 22, 1447–1459. Parker, D. C., Manson, S. M., Janssen, M. A., Hoffmann, M. J., & 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, 314–337. Pijanowski, B. C., Pithadia, S., Shellito, B. A., & Alexandridis, K. (2005). Calibrating a neural network-based urban change model for two metropolitan areas of the Upper Midwest of the United States. International Journal of Geographical Information Science, 19, 197–215. Pijanowski, B. C., Tayyebi, A., Delavar, M. R., & Yazdanpanah, M. J. (2009). Urban expansion simulation using geospatial information system and artiﬁcial neural networks. International Journal of Environmental Research, 3, 493–502. Raﬁee, R., Salman Mahiny, A., Khorasani, N., Darvishsefat, A. A., & Danekar, A. (2009). Simulating urban growth in Mashad City, Iran through the SLEUTH model (UGM). Cities, 26, 19–26. Rindfuss, R. R., Walsh, S. J., Turner, B. L., Fox, J., & Mishra, V. (2004). Developing a science of land change: Challenges and methodological issues. Proceedings of the National Academy of Sciences of the United States of America, 101, 13976–13981. Robinson, D. T., Brown, D. G., 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 building agent-based models in land use science. Journal of Land Use Science, 2, 31–55. Sabet Sarvestani, M., Ibrahim, A. L., & Kanaroglou, P. (2011). Three decades of urban growth in the city of Shiraz, Iran: A remote sensing and geographic information systems application. Cities, 28, 320–329. Santé, I., García, A. M., Miranda, D., & Crecente, R. (2010). Cellular automata models for the simulation of real-world urban processes: A review and analysis. Landscape and Urban Planning, 96, 108–122. Statistical Centre of Iran. . Tian, G., Ouyang, Y., Quan, Q., & Wu, J. (2011). Simulating spatiotemporal dynamics of urbanization with multi-agent systems—A case study of the Phoenix metropolitan region, USA. Ecological Modelling, 222, 1129–1138. Valbuena, D., Verburg, P. H., & Bregt, A. K. (2008). A method to deﬁne a typology for agent-based analysis in regional land-use research. Agriculture Ecosystems & Environment, 128, 27–36. Vitousek, P. M., Mooney, H. A., Lubchenco, J., & Melillo, J. M. (1997). Human domination of earth’s ecosystems. Science, 277, 494–499. Waddell, P. (2002). Modeling urban development for land use, transportation, and environmental planning. Journal of the American Planning Association, 68, 297–314. Weiss, G. (1999). Multiagent systems: A modern approach to distributed artiﬁcial intelligence. MIT Press. Wilensky, U. (2009). NetLogo 4.1 User Manual, Technical report. In Tutorial on agent-based modelling and simulation. Evanston, IL: Center for Connected Learning and Computer-Based Modeling, Northwestern University. Wooldridge, M. J. (2009). An introduction to multiagent systems. John Wiley & Sons. Zhang, Q., Ban, Y., Liu, J., & Hu, Y. (2011). Simulation and analysis of urban growth scenarios for the Greater Shanghai Area, China. Computers, Environment and Urban Systems, 35, 126–139.