The Impact of Data Time Span on Forecast ... - Semantic Scholar
International Journal of Applied Geospatial Research, 5(3), 21-35, July-September 2014 21
The Impact of Data Time Span on Forecast Accuracy through Calibrating the SLEUTH Urban Growth Model Reihaneh Peiman, Sam Houston State University, Huntsville, Texas, USA & University of Pisa, Pisa, Italy Keith Clarke, University of California- Santa Barbara, Santa Barbara, California, USA
ABSTRACT Does the spacing of time intervals used for model input data have an impact on the model’s subsequent calibration and so projections of land use change and urban growth? This study evaluated the performance of the SLEUTH urban growth and land use change model through two independent model calibrations with different temporal extents (1972 to 2006 vs. 2000 to 2006) for the historical Italian cities of Pisa Province and their surroundings. The goal in performing two calibrations was to investigate the sensitivity of SLEUTH forecasts to longer or shorter calibration timelines, that is does calibrating the model over a longer time period produce better model fits and therefore forecasts? The best fit parameters from each calibration were then used in forecasting urban growth in the area up to the year 2027. The authors findings show that the spatial growth estimated by the model was strongly influenced by the physical landscape and road networks. The forecast outputs over 100 Monte Carlo trials reflect the start of newly formed detached settlements towards and along existing roads, i.e., classic urban sprawl. The authors conclude that the short term calibration was a better model fit compared to the long term calibration. Nevertheless, the absolute preference for the short-term calibration over long-term implies that time-sensitivity in calibration remains a challenge for SLEUTH applications. Keywords:
Calibration, Land Use Change, Pisa Province, SLEUTH, Urban Growth
INTRODUCTION Models form a bridge between description and analysis on the one hand, and explanation and prediction on the other (Clarke et al., 2007). Models are essential components
to understand the dynamics of urban growth, and one way to conduct experiments. The processes underlying land use change are able to generate aggregate structures and future patterns that are not reflected at the level of the individual system constituents, exhibiting the
DOI: 10.4018/ijagr.2014070102 Copyright © 2014, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
22 International Journal of Applied Geospatial Research, 5(3), 21-35, July-September 2014
property of emergence (Holland, 1998). Using a complex systems approach (Silva, 2010), a variety of spatial scales, including global and superregional urban networks, metropolitan agglomerations, as well as urban growth and land use change within individual cities at the local level can be addressed by urban modeling and through theories of urban dynamics (Alberti, 1999; Alberti & Waddell, 2000). The simplest models to exhibit these emergent and scaling properties are cellular automata (CA). Couclelis (1985; 1997) has strongly advocated the use of CA as potentially powerful contributors to urban systems modeling. CA operation is based on an array of identically programmed automata or cells that exist in one of a finite number of states (Park & Wagner, 1997; Clarke & Gaydos, 1998). During recent years, the CA method-in combination with the use of geographic information systems (GIS) as tools of data assimilation-has been widely applied to analyze and simulate urban growth. Despite criticism of the first generation of urban computer models (Lee, 1973; Lee, 1994), computational CA models have proven useful in urban planning. In reality, the complexity of the process underlying urban growth results in significantly different characteristics from city to city, so that modeling results derived from heterogeneous applications are generally incomparable (Benenson & Torrens, 2004). Thus good CA models are those that can be made to adapt common behaviors to the spatial heterogeneity of their application environments. The SLEUTH model employs a modified CA to simulate the spread of urbanization across a landscape (Clarke et al., 1997; Clarke & Gaydos, 1998). The model was initially applied to North American cities such as San Francisco (Clarke et al., 1997), Chicago, the Washington-Baltimore area (Clarke & Gaydos, 1998), Sioux Falls, central California (Tietz et al., 2005), and Philadelphia (Varanka, 2001). Subsequently, other examples of the application of the SLEUTH model to various worldwide case studies include: Lisbon and Porto (Silva & Clarke, 2002) in Portugal (Europe), Porto Alegre (Leao, 2002) in Brazil (South America), Cape
Town in South Africa (Watkiss, 2008), Chang Mai (Sangawongse, 2006), Thailand (Asia), and Sydney (Liu and Phinn, 2004), Australia. Over a decade of cellular modeling with SLEUTH, the model has found many improvements in its own performance to the extent that its limitations have been greatly reduced in comparison to the first applications of the model (Clarke 2008a, b). Today SLEUTH is used not only for local applications, but also for forecasting urban development at regional and continental scales, and for informal settlements (Sietchiping, 2004). The acronym SLEUTH is from the model’s required input data layers, that is, Slope, Land use, Exclusion, Urban extent, Transportation, and Hillshade, which the model uses to simulate land use dynamics as a physical process (Clarke & Gaydos, 1998). In the model, the behavior of four forms of urban growth is controlled by five parameters. The model uses brute force calibration based on a combination of Monte Carlo (MC) techniques and the hindcasting method (Morrison et al., 1992). During calibration, model forecasts are statistically and spatially associated with actual historical growth patterns. As such, SLEUTH as a spatially and temporally explicit model is capable of simulating alternate growth scenarios using a rigorous calibration process (Syphard et al., 2005), and of interpolating between past data years. As calibration ensures the function of the model, enhancing the calibration process has been the subject of recent research (e.g., Yang & Lo, 2003; Goldstein, 2004; Dietzel & Clarke, 2007; Jantz et al., 2010; Clarke-Lauer & Clarke, 2011). Briefly, calibration of the model using historical data results in the selection of final coefficients which can then be applied to forecast urban development over the coming years and even decades. With close to 100 applications of SLEUTH to cities and regions over more than a decade, SLEUTH has enjoyed a longer than typical lifetime as an urban model (Clarke, 2008a).“Traditional models of urbanization have sought to model and predict either the economic and size relationship between cities
Copyright © 2014, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
13 more pages are available in the full version of this document, which may be purchased using the "Add to Cart" button on the product's webpage: www.igi-global.com/article/the-impact-of-data-time-span-onforecast-accuracy-through-calibrating-the-sleuth-urbangrowth-model/118257?camid=4v1
This title is available in InfoSci-Journals, InfoSci-Journal Disciplines Engineering, Natural, and Physical Science. Recommend this product to your librarian: www.igi-global.com/e-resources/libraryrecommendation/?id=2
Related Content Spatiotemporal Pattern Analysis of Rapid Urban Expansion Using GIS and Remote Sensing Hyun Joong Kim (2012). Geospatial Technologies and Advancing Geographic Decision Making: Issues and Trends (pp. 188-204).
www.igi-global.com/chapter/spatiotemporal-pattern-analysis-rapidurban/63604?camid=4v1a Monitoring Urban Sprawl and Sustainable Urban Development Using the Moran Index: A Case Study of Stellenbosch, South Africa Walter Musakwa and Adriaan van Niekerk (2014). International Journal of Applied Geospatial Research (pp. 1-20).
www.igi-global.com/article/monitoring-urban-sprawl-and-sustainable-urbandevelopment-using-the-moran-index/118256?camid=4v1a Data Mining Location-Based Social Networks for Geospatial Discovery Edward Pultar (2013). Geographic Information Systems: Concepts, Methodologies, Tools, and Applications (pp. 2006-2019).
www.igi-global.com/chapter/data-mining-location-basedsocial/70547?camid=4v1a
GML as Database: Present and Future Jose E. Córcoles and Pascual González (2009). Handbook of Research on Geoinformatics (pp. 1-10).
www.igi-global.com/chapter/gml-database-present-future/20380?camid=4v1a
The Impact of Data Time Span on Forecast Accuracy through Calibrating the SLEUTH Urban Growth Model Reihaneh Peiman, Sam Houston State University, Huntsville, Texas, USA & University of Pisa, Pisa, Italy Keith Clarke, University of California- Santa Barbara, Santa Barbara, California, USA
ABSTRACT Does the spacing of time intervals used for model input data have an impact on the model’s subsequent calibration and so projections of land use change and urban growth? This study evaluated the performance of the SLEUTH urban growth and land use change model through two independent model calibrations with different temporal extents (1972 to 2006 vs. 2000 to 2006) for the historical Italian cities of Pisa Province and their surroundings. The goal in performing two calibrations was to investigate the sensitivity of SLEUTH forecasts to longer or shorter calibration timelines, that is does calibrating the model over a longer time period produce better model fits and therefore forecasts? The best fit parameters from each calibration were then used in forecasting urban growth in the area up to the year 2027. The authors findings show that the spatial growth estimated by the model was strongly influenced by the physical landscape and road networks. The forecast outputs over 100 Monte Carlo trials reflect the start of newly formed detached settlements towards and along existing roads, i.e., classic urban sprawl. The authors conclude that the short term calibration was a better model fit compared to the long term calibration. Nevertheless, the absolute preference for the short-term calibration over long-term implies that time-sensitivity in calibration remains a challenge for SLEUTH applications. Keywords:
Calibration, Land Use Change, Pisa Province, SLEUTH, Urban Growth
INTRODUCTION Models form a bridge between description and analysis on the one hand, and explanation and prediction on the other (Clarke et al., 2007). Models are essential components
to understand the dynamics of urban growth, and one way to conduct experiments. The processes underlying land use change are able to generate aggregate structures and future patterns that are not reflected at the level of the individual system constituents, exhibiting the
DOI: 10.4018/ijagr.2014070102 Copyright © 2014, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
22 International Journal of Applied Geospatial Research, 5(3), 21-35, July-September 2014
property of emergence (Holland, 1998). Using a complex systems approach (Silva, 2010), a variety of spatial scales, including global and superregional urban networks, metropolitan agglomerations, as well as urban growth and land use change within individual cities at the local level can be addressed by urban modeling and through theories of urban dynamics (Alberti, 1999; Alberti & Waddell, 2000). The simplest models to exhibit these emergent and scaling properties are cellular automata (CA). Couclelis (1985; 1997) has strongly advocated the use of CA as potentially powerful contributors to urban systems modeling. CA operation is based on an array of identically programmed automata or cells that exist in one of a finite number of states (Park & Wagner, 1997; Clarke & Gaydos, 1998). During recent years, the CA method-in combination with the use of geographic information systems (GIS) as tools of data assimilation-has been widely applied to analyze and simulate urban growth. Despite criticism of the first generation of urban computer models (Lee, 1973; Lee, 1994), computational CA models have proven useful in urban planning. In reality, the complexity of the process underlying urban growth results in significantly different characteristics from city to city, so that modeling results derived from heterogeneous applications are generally incomparable (Benenson & Torrens, 2004). Thus good CA models are those that can be made to adapt common behaviors to the spatial heterogeneity of their application environments. The SLEUTH model employs a modified CA to simulate the spread of urbanization across a landscape (Clarke et al., 1997; Clarke & Gaydos, 1998). The model was initially applied to North American cities such as San Francisco (Clarke et al., 1997), Chicago, the Washington-Baltimore area (Clarke & Gaydos, 1998), Sioux Falls, central California (Tietz et al., 2005), and Philadelphia (Varanka, 2001). Subsequently, other examples of the application of the SLEUTH model to various worldwide case studies include: Lisbon and Porto (Silva & Clarke, 2002) in Portugal (Europe), Porto Alegre (Leao, 2002) in Brazil (South America), Cape
Town in South Africa (Watkiss, 2008), Chang Mai (Sangawongse, 2006), Thailand (Asia), and Sydney (Liu and Phinn, 2004), Australia. Over a decade of cellular modeling with SLEUTH, the model has found many improvements in its own performance to the extent that its limitations have been greatly reduced in comparison to the first applications of the model (Clarke 2008a, b). Today SLEUTH is used not only for local applications, but also for forecasting urban development at regional and continental scales, and for informal settlements (Sietchiping, 2004). The acronym SLEUTH is from the model’s required input data layers, that is, Slope, Land use, Exclusion, Urban extent, Transportation, and Hillshade, which the model uses to simulate land use dynamics as a physical process (Clarke & Gaydos, 1998). In the model, the behavior of four forms of urban growth is controlled by five parameters. The model uses brute force calibration based on a combination of Monte Carlo (MC) techniques and the hindcasting method (Morrison et al., 1992). During calibration, model forecasts are statistically and spatially associated with actual historical growth patterns. As such, SLEUTH as a spatially and temporally explicit model is capable of simulating alternate growth scenarios using a rigorous calibration process (Syphard et al., 2005), and of interpolating between past data years. As calibration ensures the function of the model, enhancing the calibration process has been the subject of recent research (e.g., Yang & Lo, 2003; Goldstein, 2004; Dietzel & Clarke, 2007; Jantz et al., 2010; Clarke-Lauer & Clarke, 2011). Briefly, calibration of the model using historical data results in the selection of final coefficients which can then be applied to forecast urban development over the coming years and even decades. With close to 100 applications of SLEUTH to cities and regions over more than a decade, SLEUTH has enjoyed a longer than typical lifetime as an urban model (Clarke, 2008a).“Traditional models of urbanization have sought to model and predict either the economic and size relationship between cities
Copyright © 2014, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
13 more pages are available in the full version of this document, which may be purchased using the "Add to Cart" button on the product's webpage: www.igi-global.com/article/the-impact-of-data-time-span-onforecast-accuracy-through-calibrating-the-sleuth-urbangrowth-model/118257?camid=4v1
This title is available in InfoSci-Journals, InfoSci-Journal Disciplines Engineering, Natural, and Physical Science. Recommend this product to your librarian: www.igi-global.com/e-resources/libraryrecommendation/?id=2
Related Content Spatiotemporal Pattern Analysis of Rapid Urban Expansion Using GIS and Remote Sensing Hyun Joong Kim (2012). Geospatial Technologies and Advancing Geographic Decision Making: Issues and Trends (pp. 188-204).
www.igi-global.com/chapter/spatiotemporal-pattern-analysis-rapidurban/63604?camid=4v1a Monitoring Urban Sprawl and Sustainable Urban Development Using the Moran Index: A Case Study of Stellenbosch, South Africa Walter Musakwa and Adriaan van Niekerk (2014). International Journal of Applied Geospatial Research (pp. 1-20).
www.igi-global.com/article/monitoring-urban-sprawl-and-sustainable-urbandevelopment-using-the-moran-index/118256?camid=4v1a Data Mining Location-Based Social Networks for Geospatial Discovery Edward Pultar (2013). Geographic Information Systems: Concepts, Methodologies, Tools, and Applications (pp. 2006-2019).
www.igi-global.com/chapter/data-mining-location-basedsocial/70547?camid=4v1a
GML as Database: Present and Future Jose E. Córcoles and Pascual González (2009). Handbook of Research on Geoinformatics (pp. 1-10).
www.igi-global.com/chapter/gml-database-present-future/20380?camid=4v1a
