Modelling and Simulating Urban Residential Land Development in Jiading New City, Shanghai
- 123 Downloads
This study develops an agent-based and spatial genetic algorithm framework called Population-driven Urban Land Development (PDULD) to simulate urban land development and population dynamics. In the model, household-life cycles promote their location and relocation desires and, thus, form local housing market demand. Land developers and local governments make optimal use of current land reserves to meet housing demands. Land development in an area is treated as a multi-goal optimization activity. Community cohesion theory is introduced into the model to illustrate the influence of the population on the spatial structure of urban land use. The study uses the Spatial Genetic Algorithm to help find the best land development choices to achieve social, economic, and environmental goals. The results show that the model simulates population distribution quite well and interprets the real land use at a neighborhood level with a reasonable accuracy. A historic data comparison indicates that government policies and increasing land prices have dominated the process of land development in Shanghai based on a case study of Jiading New City.
KeywordsUrban land development Urban modelling Population dynamics Cellular automata Agent-based modelling Shanghai
Cities are emerging as humanity’s engines of creativity, wealth creation and economic growth (Bettencourt and West 2010). Despite the increasing importance of cities in human society, our ability to understand and manage them scientifically is still limited. While a variety of qualitative and quantitative methods have been adopted by urban planners and city administrators to generate plans and policies for urban resource allocation, imperative gaps still exist because of a lack of behavioral and dynamic realism in urban modelling (Batty 2008b; Heppenstall et al. 2016).
To address the gaps, urban geographers, computer scientists, and interdisciplinary researchers have made significant efforts to understand dynamic mechanisms behind urban growth and evolution by constructing diverse urban models. Due largely to the rapid advancement of computer technology, urban modellers have found that dynamic simulation models may have the power to delineate the process of urban structural evolution (Batty 1971; Couclelis 1986; White and Engelen 1993). Spatial simulation techniques such as Cellular Automata (CA) and Agent-based Modelling (ABM), have been introduced into urban land-use dynamics studies (Wahyudi and Liu 2016). The advantages of these methods are manifold, including their abilities to incorporate nonlinear, unorganized processes and cross-scale environments that can exhibit emergent properties, dynamic local interactions with spatial references, and indirect impacts and pattern-process linkages (Parker et al. 2003; Feng et al. 2016).
To date, studies on urban dynamics modelling can be classified into three main domains: theoretical modelling based on classical urban land-use theory, hybrid modelling combining land-use theories and empirical findings, and micro-simulation modelling driven by empirical data and specific case study (Huang et al. 2014). In terms of the theoretical modelling approach, researchers often adopt classic urban land-use theories to simulate the growth and expansion of urban land (Batty 2008a; Clarke et al. 1997; Clarke and Gaydos 1998; Batty and Xie 1994; White and Engelen 1993). For instance, Xu et al. (2007) built a CA model on the basis of the work of Dietzel et al. (2005) to test diffusion and coalescence theory in urban evolution; Sasaki and Box (2003) developed an agent-based model to verify von Thünen’s location theory; and Filatova et al. (2009) established a bilateral agent-based land market model using the land-market and land price theory to test Alonso’s mono-centric land price theory. Most of these studies are rooted in the classic land-use theories but fail to include social, political, and cultural factors in modelling individual decision-making processes in land development (SU 1998).
To address this deficiency, some researchers try to integrate empirical land development phenomenon, such as urban sprawl, urban gentrification, and various administrative policy into modelling and simulating land development processes, in addition to conventional land availability and suitability factors (Wu 1998; Li and Yeh 2000; Wu 2002; Yeh and Li 2002). In these models, local, regional and global constraint scores are used to re-estimate the development probabilities that are calculated from standard dynamic models. These models address the problem of neglecting the spatiality of institutional policies, planning and spatial administrative regulations, but they still overlook the heterogeneity characteristics in residential location and housing choice, local corporate land development strategy, and the involvement of government agents.
Microsimulation modelling method was firstly introduced in 1950s by Orcutt (1957) in an attempt to delineate the diversity of the US economic system. The technique operates at individual unit level and is capable of modeling heterogeneous behavioral activities. Using this method, researchers can incorporate demographic factors and integrate individual housing decision making into urban land development modelling (Waddell 2002; Waddell et al. 2003). It is well recognized that variations in housing preferences among individuals can give rise to an uneven spatial distribution in housing demand across a city space. Experimental studies have found a clear match between the revealed housing preferences and residence location choices (Levine et al. 2005). For example, to interpret residence choice effects on land development, Waddell developed the UrbanSim model (Waddell 2002; Waddell et al. 2003) to simulate the land-market interactions of households, firms, developers and public actors. Xie et al. (2007) constructed a dynamic household model to simulate how local household development leads to global urban landscape transformations, White et al. (2012) employed historic population and land-use data to evaluate neighborhood influences on urban land use dynamics. While these models all attempt to investigate the relationships between behavioral mechanisms of individual residents and land development processes across different scales, however, dynamic socio-economic characteristics of neighborhoods and decision mechanisms of households are rarely considered in the models. For instance, it is indicated that the existence of neighborhood spillover effects can limit the pace of urban land development (Wang 2016). In essence, these models fail to delineate the co-evolutionary processes of demographic transformation and land-use changes.
To solve the gaps in current urban modelling and improve urban land-use simulations, this study develops a theoretic model for the simulation of urban land-use growth that combines CA, ABM and spatial genetic algorithm (SGA) methods. In the model, citizen agents make their location and allocation decisions based on their socioeconomic characteristics. This decision-making process leads to changes in neighborhoods’ social, economic, and environmental statuses. Government agents make development strategies and policies to balance citizens’ needs and local sustainable development goals based on regional social, economic, and environmental conditions. The development strategies are then incorporated into multi-objective planning problems that are solved using optimization methods. SGAs are used to optimize those development strategies spatially, which determines urban land use development and redevelopment. Jiading New City, which was planned as a satellite city in metropolitan Shanghai, is selected to implement and verify the model in this study.
Following this introduction, a theoretical model of population-driven urban housing and land development is presented in section 2 and 3. The study will then introduce the study area and methods for data collection and analysis in section 4, followed by descriptions of model implementation, parameters estimation, and model verification in section 5. Model simulation results will be presented in Section 6. The paper concludes by discussing the findings and their implications.
Population-Driven Urban Land Development (PDULD) Framework
For land resource demand and supply, several processes need to be considered. Firstly, the lifecycles of households initiate the formation of a general housing demand market (Mulder and Hooimeijer 1999). The varying residential family structures and financial resources present diverse requirements in the formed market (Logan et al. 2002). Finally, the household location and relocation decision-making activities will lead to the differentiation of housing needs in space (Carrion-Flores and Irwin 2004; Clark and Dieleman 1996).
In addition to the diverse demands generated by households, housing supplies are further shaped by land and housing developers and local governments. While land and housing developers attempt to maximize their economic returns by supplying sufficient housing units, local governments assess local land reserves for developing residential land and build houses to meet the demands to optimize regional land use through land use planning and land-development goals (Wei 2002; Wei 2013) (Fig. 1). Therefore, the logic of modelling new land development from a housing market perspective needs to consider these factors in order to build a comprehensive and reasonably accurate simulation model. The following sections detail each element of the model.
Household Life Cycle Stage Model
Numerous measures can be employed to depict the social and financial status of an individual household. Age and revenue are two important ones (Mulder and Hooimeijer 1999; Abdullah et al. 2013). In the model, the age of a household head (A) increases with the lapse of simulation time. Assuming that an increase in household members may eventually create a need for a new or additional house unit, the probability of purchasing a house increases over time along with increasing age of household head. The yearly household revenue (R) also increases at a defined ratio (θ) with time lapse due to increasing seniority and, consequently, human capital in local labor markets.
Household Location/Relocation Desire Model
Household Location/Relocation Model
In addition to neighborhood cohesion, physical environment is another crucial factor that a household will consider in selecting a potential housing location. The physical environment may include transportation access, available public facilities, and housing quality (Levy and Lee 2011). Public transportation is important in metropolitan areas. Many cities around the world, especially in developing countries, operate under what is called transit-oriented development (Cervero and Day 2008). In addition, a city with a high development ratio usually has better public facilities and access to employment than do other areas (Hansen 1959). To consider these factors, this study uses the distance to metro as a transportation access indicator and the neighborhood development ratio (developed land divided by the total developable land in a neighborhood) as a public facility indicator.
Land Supply Model
Land-Use Optimization Model
In a city, local governments and land use management specialists need to analyze the available land resources and make optimal area development strategies and plans based on social, economic, environmental, and ecological conditions. Urban-land owners (either government agents or private developers) can be assumed to be rational economic human beings who want to achieve the highest land sale revenue (Potepan 1996). According to the classic urban land-use theory, the closer a piece of land is to the central area, the higher its appraisal price (Alonso 1964). This is similar for main public facilities such as metro lines, schools, and hospitals. Therefore, land parcels close to urban centers and public facilities will generate higher development profits than the others, and therefore will receive higher development priority.
One key factor that urban administrators and government officials may consider in urban planning is land-development compactness. Often, low compactness is reflected in over-dispersed or leapfrog land-use patterns, also known as urban sprawl. Urban sprawl usually brings low land-use efficiency and high infrastructure and maintenance costs to a city (Brueckner and Fansler 1983). To land and housing developers, correspondingly, low compactness means high offsite and operating costs. To residents, compact land-use means low transportation costs and walkable, friendly environments (Levine and Frank 2007). Therefore, urban land-use compactness is one of the goals that both public and private agents hope to achieve in developing countries.
At the same time, a given neighborhood’s social structure will also influence land development in its neighboring areas. High neighborhood cohesion and social and financial status will help attract more households to the area, and consequently, these areas could be more heavily promoted by private developers and local development authorities (Ding and Knaap 2002). In addition to these social and economic concerns, residential environment is another highly pressing issue in the land-development process. Maintaining certain amounts of green space usually creates better residential environments.
Spatial Genetic Algorithms
As was noted in the previous section, land development requires to balance efficient resource use, environmental protection, economic development, and social equity (Cao et al. 2011). As formulated in Eq. (9), housing and land supply is proposed to optimize the development of available land to achieve development goals. Therefore, land-development models need to identify optimal solutions with multiple objectives.
Various methods and technologies, such as multi-criteria decision-based, simulation-based, and optimization-based models, have been proposed and applied to solve land allocation and optimization problems (Liu et al. 2015; Liu et al. 2014). Among them, spatial evolutionary optimization methods have the advantages of being able to incorporate many objectives and generate disparate spatial solutions (Ligmann-Zielinska and Jankowski 2010).
For instance, cells (0), (1), (2), (3), (4) in Fig. 2 are selected as the chromosomes. Within the selected parent (1), cells (0), (2), and (4) are possible land parcels for development (the same for cells (1), (2), and (3) in parent (2)). Next, to the selected chromosome in parent (1) is a developed land parcel (5). Based on the parent (1) development choice, cell (5) only has one Moore neighbor (4-neighbour) (cell (2)). After pair-cell crossover (cells (0) and (4)), cell (5) has two Moore neighbors (cells (1) and (2)). In the case of land-use, generated child (1) is apparently more compact than parent (1) before the genetic operations. This method allows for randomly generating a group of possible land-use patterns as initial parents, using multiple land-development goals as objective functions, and completing the reproduction, crossover, and mutation processes to determine the optimal land-development order.
In a next step, the proposed PDULD model and SGA method were implemented and tested in one of the fastest-developing areas in Shanghai, Jiading New City.
Study Area and Data
In the heart of Jiading district is Jiading New City (Fig. 3 left), which was planned as a satellite city as early as in the1960s. The original goal of the satellite city plan was to redistribute the over-crowded population in Shanghai’s inner city proper. The initial 1959 urban master plan for Jiading New City described the area as a satellite city with 100 to 200 thousand people, along with certain industrial land uses and independent well-built public infrastructure. In the master plan that was compiled in the 1980s, Jiading New City was re-articulated as a satellite city with 200 to 300 thousand people. The population prediction was changed again to 800 thousand to one million people in 2004 due to the rapid urbanization and suburbanization in the surrounding areas of the Jiading New City.
Inside Jiading New City (Fig. 3 right) at the north end is the Jiading Old Town, with a long history of development. In the lower middle part is a planned industrial park. The east and west sides are agricultural land that is set aside for ecological protection, and the remainder, especially along the metro line, is the planned central area of Jiading New City for residential, commercial, public facility, and government uses.
According to the Shanghai Jiading New City Master Plan 2004–2010, the region represents one of the city’s three new development hubs. It is a well-planned modern city with a rational allocation of resources and balanced strategic goals. Meanwhile, the current (2013) land-use map (Fig. 3 left) shows that most the land is developed and the area now is connected to Shanghai’s central city proper. Therefore, the study of this case area provides a promising case for understanding land development in Shanghai in recent decades.
The empirical data include time-series Landsat Thematic Mapper (TM) satellite images from 1987, 1993, 2000, and 2010, historical thematic land-use maps, urban planning data, and social and economic statistics. The 30-m TM remote sensing data were obtained from the U.S. Geological Survey website (http://www.usgs.gov/), and a Random Forest Classifier, EnMap Box (van der Linden et al. 2015), was adopted to classify the remote sensing data into urban versus non-urban land use. Land-use survey data collected in 2002, 2006, and 2013 by Shanghai Government were obtained from the Shanghai Urban Planning and Land Use Administration Bureau to verify the classification results. Around 90% classification accuracy rate was achieved for each of 1987, 1993, 2000, and 2010 TM data. Historical urban land-use data were also used to calibrate, and validate the developed model. Statistical data were obtained from the Shanghai Municipality Statistics Bureau, and the urban planning data were obtained from individual urban planning departments. In addition, detailed 2000 and 2010 neighborhood-level population census data were used to project and model household location-relocation activities. Because only 2000 and 2010 population census data were available neighborhood-level for the city, this study used year 2000 as the start year for model simulation and used the 2013 land survey data to calibrate the model.
In addition to the land use and population data, an extensive field study of the area was carried out in the summer of 2015. We interviewed 11 urban planners, urban land use and planning administrators, and government officials to understand the land-development processes, government policies, and institutional organization and mechanisms of the city and the study area (Qiu and Xu 2017). Furthermore, we also conducted a questionnaire survey of 400 citizens about their household location-relocation choices. A field study was conducted to check the on-site implementation of the Jiading Master Plan, Jiading New City Detailed Plan, Jiading North Industry Park Plan, Juyuan Detailed Plan, and Nanxiang Detailed Plan to observe and confirm the neighborhood population, land use types, amenities and lifestyles. We walked around the focal areas of these plans and checked whether the plans matched the actual land use, populations, amenities and lifestyles of the residents in these areas. Video and photo images of the landscape and everyday street life were also taken randomly for analysis. The collected information assisted in a better understanding of the case study area.
Model Parameters and Calibration
Because there were no household-level population statistic data available for the area, we projected the household groups that lived in each residential cell using neighborhood population statistic data. Firstly, the number of male citizens age 25 years and older in each neighborhood; these were treated as the household heads. Each household head was then assigned an age (A) based on the male age group data from neighborhood-level statistics; next, the household heads were evenly distributed among the residential cells within each neighborhood. Each household head is treated as an agent, generating a total of 52,422 household agents to represent the total number of households in the study area in 2000. According to the Shanghai Statistical Year Books, the area enjoyed a 12% average annual household increase during the decade since 2000.
Local household agents, once they change their personal statuses, will check how they feel about the connection with their new neighborhoods. If the difference D i, t is larger than a defined threshold, TH D , and household savings, S t , is greater than the new house down payment threshold, TH S , the household agent will consider moving out of that neighborhood to another one in the same region. Meanwhile, household agent’s action generates a random number (0 < =Rnd<=1000) that is compared with a predefined threshold, TH O , to determine whether the household will leave the region or not. In addition, each household agent will generate a probability value through Eq. (2) that can be compared with threshold, TH A , to determine whether the agent (and thus the household) will die off.
Land supply and development
With the increasing housing demand, new residential land needs to be developed. The annual land supply is defined through Eq. (7). Because the study area was divided into regular grids, the area of each cell (s i ) was defined. Moreover, Old Town Center was treated as the origin point because at the time of the study, it was still the center of the region, and the Euclidian distance (μ) was calculated for each land cell to Old Town Center. After multi-time simulations, the annual land supply Y was found to be close to the actual urban land increase rate when δ was close to the reciprocal of the maximum distance.
With all the developable land cells, the model used SGA to determine the best land development order. All the parameters in Eqs. (10), (11), (12), (13), and (14) had to be estimated. Once the most suitable land cells are chosen, houses will be built on them. To determine the number of houses that could be built on each residential cell, this study used floor area ratio data from Jiading New City Regulation Plan (2013) and kriging interpolation to generate the maximum construction volume surface of the study area. Next, the study assumed a normal apartment home size of 120 m2 and divided the maximum construction volume to generate the maximum number of houses that could be built on each residential cell. Once new houses are built, household groups that wish to move will choose to move to the neighborhood with the most similar households (the smallest D).
All the land use maps and boundary data were first digitized and stored in the GIS database as vector data and were then converted into raster grids. To simulate land-development using the proposed models, the raster data were transformed into ASCII files, and then the ArcGIS Euclidean Distance tool was used to generate the distances from each cell to the metro lines and the city center. The generated raster distance data were also transformed into ASCII files with the same cell sizes as the others. After all the data were in place, the model was coded and run in one of the most popular simulation platforms, NetLogo (Wilensky 1999). The ASCII data files were imported into the model directly using the GIS extension in NetLogo.
Predefined and estimated parameters
Simplified source codes:
a1 < − seq(0, 1, by = 0.1)➔a14 < − seq(0, 10, by = 1).
a < − rbind(a1,a2,a3,a4,a5,a6,a7,a8,a9,a10,a11,a12,a13,a14).
NLLoadModel(“… /Residential Model.nlogo”).
NLCommand(“set mu1”, a[1,×1]) ➔ NLCommand(“set Omiga4”, a[14,×14]).
GA < − ga(type = “real-valued”, fitness = function(x) + Rastrigin(x, ➔ x), min = c(1, ➔ 1),max = c(12, ➔ 12), popSize = 10,maxiter = 20, monitor = TRUE).
Residential Land Use
To further analyze the accuracy of the simulation results and investigate the land development trajectories in the study area, we compared the simulated results with the actual land use changes, specifically by using the simulated neighborhood residential land increase volume minus real neighborhood residential land increase divided by real neighborhood residential land increase to generate the simulation result index. A positive result denotes overestimation and a negative result indicates underestimation.
Household and Neighborhood Simulation Results
The Fig. 9 time series box plot graph and Fig. 10 neighborhood age maps also show that even though the general trend of average age differences across area neighborhoods decreased in the simulation, the age gaps still exist between the Old Town communities and rest of the region: The Old Town area has a relatively high average age.
Discussion and Conclusion
Land development modelling and simulation provides an important computational and visualization tool for assessing the impacts of future land demand for population and economic growth. It also provides a means for understanding complex causal relations and dynamic interactions among numerous factors ranging from economic, political, and social to environmental dimensions.
This study proposes a new simulation model of urban land-use growth and evolution that combined ABM, CA and SGA methods. The model is featured by a proposed population-driven urban land development framework. Within the model, a household group determines its housing location desires and then forms the local housing demand market, and land developers and local governments then make the optimal use of the current land reserves to meet these housing demands. In this process, land development in an area is treated as a multi-goal optimization process. A spatial genetic algorithm is used to help identify the best land development choice for achieving social, economic, and environmental goals of a given urban area.
In addition to environmental factors, this study introduces community cohesion theory into the model to illustrate the influence of populations on the spatial structure of urban land use. In the model, household heads act as autonomic agents who assess the socioeconomic status of their current neighborhoods and can choose to relocate to other neighborhoods, and a given neighborhood’s development status affects local land development activities. In this way, this study innovatively creates a dynamic evolutionary model that integrate both population and land-development dynamics in land development simulation.
Moreover, the study proposes a new parameter estimation method by using evolutionary algorithms. Sample data regression and multi-time experiments are among the main land use and cover change model parameter estimation methods (Couclelis 2001). These methods both are very time consuming and highly inaccurate. This study represents the first attempt to use evolutionary algorithms and historic data to estimate unknown parameters through multi-generation training. It opens a new window for future similar studies.
The simulation results show that the model interpret the real land use at the neighborhood level with a moderate success (53%). There are major (34%) overestimations around the eastern rural communities in the study area and the nearby outlying areas of Old Town Centre. The data comparisons show that the Old Town communities have all been built up. One of the main explanations is that these lands were dominated by commercial uses such as office buildings and other facilities, which out-compete residential land development. The neighborhood household simulation results match the current household distribution quite well. The average household revenue and cohesion of the neighborhoods in the study area diverge from 2000 to 2012, and average residential neighborhood age declines throughout the whole study area.
The modelling and simulation results in this study confirm the findings in the literature (SU 1998) that urban land use development is highly affected by a city’s household social, economic, and environmental characteristics. However, two key issues emerge from this study: government intervention and land use profit competition.
As shown in the previous section, the land-use simulation results deviate heavily from the real land use in some of the study area, even though the simulated residential spatial distributions match with the reality well. The actual residential land use increases in the study area (Fig. 6) are clustered around the outskirts of the Old Town and the planned new center area of Jiading New City. Compared with the sparsely distributed land use in the simulation results, the actual residential land uses are in the form of large blocks. One of the major reasons for this clustered form is that the area is developed according to urban planning zoning and rezoning policies. The whole area is divided into several large functional zones and then subdivided into large land-use blocks. Each of these blocks is designed for residential, commercial, industrial, or recreation use. Inside the designed residential blocks are high-density apartment buildings. The mismatch between the actual land use and the population simulation results indicate that the region is regulated artificially by administration rather than a natural growth based on market processes.
Historic data and field studies show that the coexistence of urban development and redevelopment in the area. With the increasing population and decreasing developable land, the area’s land value is increasing continuously. For instance, residential uses out-compete industrial uses in the planned industrial park, and industrial factories have had to move under both government and market pressures. The commercial uses outbid residential uses in the built-up areas around Old Town Centre in Jiading New City, as old houses were demolished or converted into uses for businesses, office buildings, and other land uses with high per-unit outputs. This process of urban redevelopment needs to be incorporated for better simulation results in the future.
- Batty, M. (2008a). Cities as complex systems: Scaling, interactions, networks, dynamics and urban morphologies. In Encyclopedia of complexity and systems science, 1041–1071. New York: Springer.Google Scholar
- Batty, M. (2008b). Fifty years of urban modeling: Macro-statics to micro-dynamics. In The dynamics of complex urban systems (pp. 1–20). New York: Springer.Google Scholar
- Clark, W. W. A., & Dieleman, F. M. (1996). Households and housing: Choice and outcomes in the housing market. New Jersey: Transaction Publishers.Google Scholar
- Couclelis, H. (2001). Why I no longer work with Agents. In Special workshop on agent-based models of land-use, land-cover change, CSISS, Irvine, available on-line at www. csiss. org/events/other/agent-based/papers/couclelis. pdf.Google Scholar
- Filatova, T., Parker, D., & van der Veen, A. (2009). Agent-based urban land markets: Agent's pricing behavior, land prices and urban land use change. Jasss-the. Journal of Artificial Societies and Social Simulation, 12, 3.Google Scholar
- Liu, Y., Liu, D., Liu, Y., He, J., Jiao, L., Chen, Y., & Hong, X. (2012). Rural land use spatial allocation in the semiarid loess hilly area in China: Using a particle swarm optimization model equipped with multi-objective optimization techniques. Science China Earth Sciences, 55, 1166–1177.CrossRefGoogle Scholar
- Liu, Y., Yuan, M., He, J., & Liu, Y. (2015). Regional land-use allocation with a spatially explicit genetic algorithm. Landscape and Ecological Engineering, 11(1), 209–219.Google Scholar
- Logan, J. R., Zhang, W., & Alba, R. D. (2002). Immigrant enclaves and ethnic communities in New York and Los Angeles. American Sociological Review, 67(2), 299–322.Google Scholar
- Mulder, C. H. & P. Hooimeijer (1999). Residential relocations in the life course. In Population Issues: An Interdisciplinary Focus (pp. 159–186). New York: Springer.Google Scholar
- Orcutt, G. H. (1957). A new type of socio-economic system. The Review of Economics and Statistics, 39(2), 116–123.Google Scholar
- Sasaki, Y., & Box, P. (2003). Agent-based verification of von Thünen’s location theory. Journal of Artificial Societies and Social Simulation, 6(2), 1–9.Google Scholar
- Schelling, T. C. (2006). Micromotives and macrobehavior. New York: WW Norton & Company.Google Scholar
- Wahyudi, A., & Liu, Y. (2016). Cellular automata for urban growth modelling. International Review for Spatial Planning and. Sustainable Development, 4, 60–75.Google Scholar
- Wu, F. L. (1998). SimLand: A prototype to simulate land conversion through the integrated GIS and CA with AHP-derived transition rules. International Journal of Geographical Information Science, 12(1), 63–82.Google Scholar
- Wu, F. L. (2002). Calibration of stochastic cellular automata: The application to rural-urban land conversions. International Journal of Geographical Information Science, 16(8), 795–818.Google Scholar