Urban expansion in Centre County, Pennsylvania: Spatial dynamics and landscape transformations

Applied Geography 29 (2009) 235–249

Contents lists available at ScienceDirect

Applied Geography journal homepage: www.elsevier.com/locate/apgeog

Urban expansion in Centre County, Pennsylvania: Spatial dynamics and landscape transformations Nnyaladzi Batisani a, b, *, Brent Yarnal a,1 a b

Department of Geography, The Pennsylvania State University, University Park, PA 16802, USA Department of Agricultural Engineering and Land Planning, University of Botswana, Gaborone, Botswana

a b s t r a c t Keywords: Urban sprawl Landscape Fragmentation Land use change Pennsylvania Spatial dynamics

Sprawling urban development is a major driving force of landscape fragmentation and loss of agricultural land. Despite this understanding, science has yet to realize a coherent suite of methods to analyze all circumstances of sprawl. Consequently, this paper contributes to this realization by combining three methods to address sprawl in a small US metropolitan area – Centre County, Pennsylvania: cross-tabulation to identify systematic non-random land use transitions; logistic regression to determine explanatory variables of urban land use location resulting from these transitions; and the CLUE-S regional modeling framework to project future urban land use patterns in the county. The results demonstrate the versatility of the methodology because of its ability to detect land use change despite the large proportion of the landscape that remained uncharged during the two periods under consideration, and because of its ability to distinguish systematic non-random land use transitions from random ones. The strength of the methodology is further demonstrated by its capability to allocate land use change according to change in land use location as well as to net change in land use quantity. The methodology identified soil and topography as the primary explanatory drivers of urban land use location in Centre County. Although the model is able to simulate urban land use location at the county level, it is less able to simulate these locations at the sub-county level, thereby suggesting that the explanatory variables for urban land location are not fully captured at this scale. Overall, the methodology for sprawl analyses presented in the study is robust and adds to the tools available to decision makers for assessing sprawl dynamics. Ó 2008 Elsevier Ltd. All rights reserved.

Introduction The 1990 Census showed that for the first time more Americans were living in suburbs than in central cities. About one-fifth of the nation’s prime farmland is located within metropolitan counties and, when non-metropolitan counties adjacent to metropolitan counties are included, these greater metropolitan areas contain over one-third of the nation’s prime farmland (Daniels, 1997; Mieskowski & Mills,1991). Farmland and natural lands contribute to food production, flood control, air cleansing, and water filtering; those amenities, as well as the inherent societal value of open space, are lost when these lands are developed (Nelson, 1992). Therefore, one challenge to land resource management in areas where urban development is taking place in prime agricultural lands is to achieve compact development that does not degrade these natural resources (Couch & Karecha, 2006).

* Corresponding author. Department of Agricultural Engineering and Land Planning, University of Botswana, Private Bag 0027, Gaborone, Botswana. Tel.: þ267 3650100; fax: þ267 3928753. E-mail addresses: [email protected] (N. Batisani), [email protected] (B. Yarnal). 1 Center for Integrated Regional Assessment, 2217 Earth and Engineering Sciences Building, The Pennsylvania State University, University Park, PA 16802, USA. Tel.: þ1 814 865 1585; fax: þ1 814 865 3191. 0143-6228/$ – see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.apgeog.2008.08.007

236

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

Although urban areas make about 4 percent of the Earth’s land surface area (World Resources Institute, 2003), urban sprawl can cause larger changes in environmental conditions than other land uses (Fang, Gertner, Sun, & Anderson, 2005; Folke, Jansson, Larsson, & Constanza, 1997; Lambin, Turner, & Geist, 2001). Urban development and associated changes in landscape composition and pattern set off a cascade of environmental impacts that are of growing concern (Alberti, 1999; Bartlett, Mageean, & O’Connor, 2000; McKinney, 2002; Nilsson et al., 2003). Urban growth can lead to landscape fragmentation resulting in loss of habitat and disruption of migration routes for many animal species. These environmental impacts are likely to increase in the 21st century when more than one-half of the world’s population is expected to be living in urban areas (UN 2004). The rapid pace and broad scope of urban growth are stressing the ability of land use planners and environmental resource managers to address the cumulative degradation of ecosystems (Lathrop, Tulloch, & Hatfield, 2007). Fang et al. (2005) note that in order to keep ecosystems functioning well, it is necessary for environmental researchers, managers, and decision makers to understand the spatial dynamics of urban sprawl. Complementing this idea, Geurs and van Wee (2006) stress that a comprehensive exploration of the consequences of urban growth is needed for informed decisions on sprawl patterns and its costs. Ichikawa, Okubo, Okubo, and Takeuchi (2006) conclude that an examination of future implications of urban development on ecosystems functioning is critical for an informed land use planning process to avoid ill-advised and irreversible land use decisions. While substantial research and academic discourse have addressed many of the socioeconomic issues related to sprawl, far less research has focused on developing concrete methodologies able to identify and characterize sprawl (Hasse, 2004). Lopez and Hynes (2003) further point out that lack of a coherent methodology to measure sprawl has been a major cause of contention among various investigators because of differences in defining the phenomenon. This lack of progress in understanding patterns and processes of urban growth within a landscape results from inconsistent measurements of urban land use patterns (Cutsinger, Galster, Wolman, Hanson, & Towns, 2005; Tsai, 2005). Sprawl refers to a type of spreading suburban development with negative outcomes, such as increased commuting time. The Florida Growth Management Plan (1993) sees sprawl as an unplanned suburban development allowing land use patterns that inflate facility costs and that fail to protect natural resources and agricultural lands. Benfield, Raimi, and Chen (1999) simply think of sprawl as a specific manifestation of problematic urban growth. This paper specifically defines it as the intrusion of low-density residential and non-residential development into rural and undeveloped areas (Burchell & Shad, 1999). Several researchers have developed methods for sprawl analysis at various spatial resolutions. Torrens and Alberti (2000) developed an empirical landscape approach to sprawl measurement that focuses on the characteristics of density, scatter, the built environment, and accessibility. Galster et al. (2001) measured sprawl using the dimensions of density, continuity, and compactness at a course resolution based on U.S. census data gridded into one-half-mile cells. This work was expanded to implement a larger set of spatial measures for a greater number of metropolitan areas (El Nasser & Overberg, 2001). Hasse and Lathrop (2003) emphasized that although these sprawl analyses are useful for inter-metropolitan comparisons at a national scale, methods aimed at a finer level of resolution are also needed to illuminate intra-metropolitan patterns of urban growth. Wang, Gertner, Howard, and Anderson (in press) supported this perspective by stressing the importance of analyzing land change at spatial resolutions relevant to scales of decision making. Moving away from scale and focusing on tools, Batty, Xie, and Sun (1999), Fang et al. (2005), and Wu (2002) highlighted the importance of modeling and simulation in sprawl analysis. Cutsinger et al. (2005) drew attention to the need for a comprehensive exploration of various land use dimensions to facilitate informed decision making on sprawl. In sum, this material strongly suggests the need to develop empirical analysis that quantifies land use patterns in ways that permits the unambiguous measurement of sprawl. Consequently, the goal of this paper is to provide a research framework and methodologies that contribute to the understanding of sprawl dynamics. It reaches this goal through three analyses. The first analysis addresses sprawl and landscape fragmentation through cross-tabulation, identifying the dominant and systematic land use transitions in the area that lead to the observed land use patterns. The second analysis subsequently identifies the underlying explanatory drivers (processes) of urban land use location (the dominant land use transition), through logistic regression. The third analysis projects future urban land use location based on the identified processes (drivers) through simulation modeling and subsequent validation of the simulated products. Materials and methods Study area Centre County, Pennsylvania (Fig. 1) typifies a growing debate regarding the tradeoffs between socioeconomic growth and development and their impacts on the landscape. It is the fifth largest county in Pennsylvania, but two thirds of its land area (2887 km2) is protected conservation area and agricultural easements. It has one of the highest median housing values in Pennsylvania with a single family housing median price of $156,000 in 2005. It is also rated highly as a retirement destination (Centre County, 2005). The combination of high residential land demand and limited land availability outside protected areas puts pressure on the land resources in the county. A wide range of land uses and land covers coexists within Centre County, with forests and agricultural lands being important components of the landscape. Forests are mainly concentrated where topography is steep and land is marginal for

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

237

Fig. 1. The location of Centre County in Pennsylvania and developable land in the county.

agricultural purposes, whereas agriculture is concentrated mainly in the fertile limestone and shale soils of the valleys (Centre County, 2005). Over the years the number and size of farms have decreased as the number of rural non-farm residents has increased in the county. Goetz et al. (2004) noted that the increase in the number of rural non-farm residents had lead to a loss of 1618 ha of prime farmland between 1977 and 2005 in the Chesapeake Bay watershed within which Centre County is located. Jantz, Goetz, and Jantz (2005) concurred and pointed out that between 1990 and 2000 there was an increase of 61 percent in the built environment of the Chesapeake Bay watershed and of this development 64 percent occurred on agricultural and grasslands, whereas 33 percent occurred on forested land. The county is divided into seven planning regions. The Centre Region Council of Governments is the planning region that is the most urbanized and home to The Pennsylvania State University. The Centre Region includes the State College Borough and College, Ferguson, Halfmoon, Harris, and Patton townships. This work uses the Centre Region to represent the sub-county level (or sub-county extent) for comparison with the county level (or county extent). The Mid-State and University Park airports service the county and the Keystone Shortway (Interstate 80) runs east–west across the county. This highway has greatly facilitated accessibility to major markets for the county’s products by the eight motor freight carriers that serve the area. The imminent completion of Interstate 99 is likely to increase the accessibility of the county’s products to markets in other parts of Pennsylvania and beyond. In addition to these commercial benefits brought to the county by the highways, increased ease of commuting by workers to commercial centers such as State College is likely to result in increased conversion of agricultural and forest lands to residential use (Centre County, 2005). The competition for land between residential and agricultural uses in the valleys and the anticipated housing demand increase make Centre County a good place to study sprawl and its effect on the landscape. Land use decisions in the United States are based on jurisdiction, and land use planning in Pennsylvania takes place at the local levels of township and borough, so sprawl analysis must take place at a scale consistent with this level of land use decision making. Data Land use/land cover data classified at Anderson level 1 from Landsat TM images of the county for 1993 and 2000 were available and used to parameterize and validate the simulation model respectively and were obtained from the Centre for Integrated Regional Assessment, The Pennsylvania State University. Land use maps had six land use categories: Urban (i.e., the built environment), Forest, Agriculture, Water, Rangeland, and Abandoned Mining Sites. Only images from summer scenes were used because they are subjected to the least variability due to changing meteorological conditions and because they permit a better distinction between forested, agricultural, and urban or built-up land than spring or winter scenes (Carlson & Arthur, 2000). These classifications were performed by an experienced analyst from the United States Geological Survey and were ground truthed extensively by utilizing a set of 1 m spatial resolution aerial photos and by field verification. Hence, they are considered highly reliable with producer and user accuracies of 90 and 83 percent, respectively. To correct for differences in brightness and spectral response (including atmospheric influences), an image-to-image empirical normalization procedure that compared invariant scene targets was used to normalize the 1993 TM imagery to the corresponding imagery from 2000 (Lathrop, 2004). The Water, Rangeland, and Abandoned (old mining sites) categories were aggregated into a single land use category called Others for this analysis. Rangeland consists of grassland where forest has been cleared and of land under pasture, while Agriculture refers exclusively to land used for arable farming. GIS layers of potential drivers of urban land use

238

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

location used in the simulation were obtained from the Land Analysis Lab, The Pennsylvania State University. The soil layer was obtained from the Soil Survey Geographic (SSURGO) database of The Natural Resources Conservation Service. The SSURGO database is at a scale of 1:24,000 resulting in 30,000 polygons for Centre County. Each map unit has three components, with the dominant component accounting for most of the variance in the unit. A common problem in spatially explicit analysis of complex systems is that the scale at which an analysis is conducted affects the type of explanation given to the observed phenomena (Gibson, Orstrom, & Ahn 1998; Levin, 1993; O’Neill et al., 1991; Veldkamp & Lambin, 2001). Veldkamp et al. (2001) agreed with this assertion and noted that at coarse (aggregated) scales the level of aggregation obscures the local variability but can show patterns invisible at detailed scales. They further noted that factors determining land use change can operate outside the affected area. Several studies have studied sprawl at the micro-scale of individual behavior (e.g., Bilsborrow & Ogondo, 1992; Bingsheng, 1996; Cutsinger et al., 2005; Hasse & Lathrop, 2003), while others have analyzed it at the macro-scale of the region and globe (e.g., Kok & Veldkamp, 2001; Lambin, 1994; Veldkamp & Lambin, 2001; Verburg et al., 2002; Wiens, 1989). Direct upscaling of processes and qualitative understanding of drivers of land use change obtained at the micro-scale to higher aggregation levels and vice versa (i.e., downscaling of processes and understandings garnered at the macro-scale to smaller spatial scales) are not possible because these relations are often subjected to scale dependencies. These dependencies can be a result of the nonlinearity of the studied relationships, feedbacks within the system, and interactions at the micro-scale (Rastetter et al., 1992; Veldkamp et al., 2001). Thus, it is not possible to derive the functioning of a land use system by solely studying the behavior of its micro-scale or macro-scale components. Heimlich and Anderson (2001) further emphasized the importance of analyzing sprawl at meso-scale by noting that between 1994 and 1997 lots of one acre or greater accounted for over 90 percent of new land converted for housing in the United States of America. Hence, there is a need to analyze land use change at the meso-scale to bridge the upscaling and downscaling problem between the micro- and macro-scales and to account for this critical scale of land use change. Therefore, to approximate the meso-scale, in this study the 30 m resolution land use maps were aggregated to 100 m and 250 m for sub-county and county levels, respectively, and the same was done to the potential drivers layers.

Methods Land use transition analysis A cross-tabulation matrix (Pontius, Shusas, & McEachern, 2004) was used to assess land use transitions between the categories of Urban, Agriculture, Forest, and Others between 1993 and 2000 at sub-county and county levels according to two pairs of components: net change (in land use quantity) and swap (in land use location), and gross gains and losses. Analysis of swap was necessary because although a net change in the quantity of a category indicates a definite change on the landscape, a lack of net change does not necessarily indicate a lack of change. It is possible that change occurs in such a way that the location of a category changes between time 1 and time 2, while the quantity remains the same. For example, a given quantity of forest loss at one location can be accompanied by the same quantity of forest gain at another place. This type of change in location is termed a ‘‘swap.’’ Analysis of these components can distinguish between a clearly systematic landscape transition and a seemingly random landscape transition. The distinction between systematic and random land transitions enables scientists to focus on the strongest signals of systematic landscape transitions, which is necessary to link pattern to process. The matrix was analyzed using a modified chi-square statistic (Pontius, Shusas, et al., 2004), which compares the matrix of observed values to a matrix of expected values. The chi-square computes the expected values by assuming that each total, Piþ and Pþj, is given a priori. The expected proportion of the landscape that experiences a transition from category i to category j due to chance is Piþ  Pþj. The expected proportion of the landscape that experiences persistence of category j due to chance is Pjþ  Pþj. Eq. (1) gives the formula for the chi-square statistic, where N is the number of grid cells in the map.

c2 ¼

J X J X i¼1 j¼1

(

 N

Pij  ðPiþ  Pþj Þ   Piþ  Pþj

2 ) (1)

Evaluation of drivers of urban land use location and sprawl simulation Prediction of urban land use location was conducted within the CLUE-S modeling framework (Verburg et al., 2002). The underlying assumption of this framework is that observed spatial relations between land use types and potential explanatory factors of land use patterns represent currently active processes and remain valid in the future. The quantitative relationship between observed land use distribution and spatial variables is derived by means of multiple regression. For this reason, the CLUE-S model is generally referred to as an empirical–statistical model. Nonetheless, statistical analysis is supplemented by a set of transition rules, which additionally control the competition between land use types. Land use changes are driven by estimates of demands at the level of the analysis (Heistermann, Muller, & Ronneberger, 2006). The model is sub-divided into two distinct modules, namely a non-spatial demand module and a spatially explicit allocation procedure (Fig. 2). The non-spatial module calculates the area change for all land use types at the aggregate level. In this case study, demands were based on linear extrapolation of the 1993–2000 urban land demand. Within the second part of

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

239

Fig. 2. Overview of the modeling procedure within CLUE-S (Verburg et al., 2002).

the model, these demands are translated into land use changes at different locations within the study extent using a rasterbased system. For the land use demand module, different alternative model specifications are possible, ranging from simple trend extrapolations (this case study) to complex economic models. The choice for a specific model is very much dependent on the nature of the most important land use conversions taking place within the study area and the scenarios that need to be considered. The results from the demand module need to specify, on a yearly basis, the area covered by the different land use types, which is a direct input for the allocation module. The allocation is based upon a combination of empirical, spatial analysis, and dynamic modeling. Fig. 3 gives an overview of the procedure. The empirical analysis unravels the relations between the spatial distribution of land use and a series of factors that are drivers and constraints of land use (Fig. 4). The results of this empirical analysis were used within the model when simulating the competition between land use types for a specific location. In addition, a set of decision rules is specified by the user to restrict the conversions that can take place based on the actual land use pattern (Verburg et al., 2002). The user can specify land use type or location-specific decision rules. Location-specific decision rules include the delineation of protected areas such as nature reserves and agricultural easements. If a protected area is specified, no changes are allowed within this area. For each land use type, decision rules determine the conditions under which the land use type is allowed to change in the next time step. These decision rules are implemented to give certain land use types a resistance to change in order to generate the stability in the land use structure that is typical for many landscapes. Three different situations can be distinguished and for each land use type the user can specify which situation is most relevant for that land use type. It is unlikely for some land use types to convert to another land use type after their first conversion; for instance, when agricultural land is converted to the built environment, it is not expected to return to agriculture or convert to forest cover. This situation is dealt with by defining the relative elasticity for change (ELASu). The relative elasticity ranges between 0 (easy to convert) and 1 (very difficult to convert). Explanatory variables of urban land use location. The land use map for 1993 was reclassified by assigning urban land use a value of 1, while other land uses were assigned a value of 0. The reclassified map was then used as the dependent variable and the potential drivers as independent variables (Table 1) in stepwise logistic regression. The choice for potential explanatory variables of land use location was based on prevalent theories (Kaimowitz & Angelsen, 1998; Lambin, Geist, & Lepers 2003; Lambin et al., 2001; Turner et al., 1995) and the inclusion of zoning regulations of the different municipalities was based on discussions with Centre County planners. The variables distance from water networks and distance from roads are Euclidean distances of the dependent variable (a 100 m2 array of cells at sub-county level and 250 m2 array of cells at county level) to the nearest water line and road. Distance from urban centers is the Euclidean distance of the dependent variable to the central business district (CBD) of each municipality. The variables are slope and elevation values within each land parcel and the variables soils suitable for agriculture (soil fertility) and soils suitable for septic works are the suitability ranking of soils within each land parcel for agricultural production and septic works based on the USDA soil capability classification system.

Spatial pattern of land use and driving forces

Logistic regression

Changes in driving forces

Probability Surface for all land use types

Demand for all land use types

Allocation of change

Actual land use

Decision rules and land use conversion elasticities

Fig. 3. Schematic representation of the procedure to allocate changes in land use to a raster-based map (Verburg et al., 2002).

240

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

Table 1 Explanatory variables of urban land use location. Variables

Explanation (units)

Population density Elevation Slope Distance from roads Distance from water networks Distance from sewer networks Distances from CBD Soils suitable for agriculture (soil fertility) Soils suitable for septic works Zoning

Inhabitants (km2) Digital elevation model (m) Derived from DEM (%) (m) (m) (m) (m) * * *

*Categorical variable.

Distances were log transformed before analysis to increase normality. To reduce the potential effects of spatial autocorrelation, a sample of 25 percent of the observations (all array of cells) was used in the logistic regression to reduce the number of direct neighbors in the dataset (Wassenaar et al., 2006). Model variables were selected by entry testing based on the significance of the score statistic, and removal testing was based on the probability of the Wald statistic. Probability for entry and removal was respectively set to 0.05 and 0.10. Collinearity was accounted for by eliminating the variable with the least significant Wald statistic contribution to the model (Wassenaar et al., 2006). The performance of the resulting regression models was evaluated by the relative operating characteristic (ROC) (Pontius & Schneider, 2001). In ordinary least squares regression, the coefficient of determination (R2) gives a measure of model fit, but there is no equivalent for logistic regression. Instead, the goodness of fit can be evaluated with the ROC method, which evaluates the predicted probabilities by comparing them with the observed values over the whole domain of predicted probabilities instead of only evaluating the percentage of correctly classified observations at a fixed cut-off value. Only variables significant at the 1 and 5 percent levels of significance are reported in the results (Table 6). A sample of urban land use location simulations is presented in the results. Validation of urban land use location simulation results. Validation techniques (Pontius, 2000, 2002; Pontius, Huffaker, & Denman, 2004) were used to determine the agreement between the 2000 urban land use reference map and the 2000 simulated urban land use location map. Furthermore, the techniques were used to compare the agreement between the 2000 reference map and the 2000 simulated map with the null model (i.e., agreement between the 1993 reference map and the 2000 reference map). Specifically, the validation technique (a) budgets sources of agreement and disagreement between the simulated map and the reference map and among location, quantity, and chance, (b) compares the predictive model to a null model that predicts pure persistence, and (c) evaluates the goodness of fit at multiple resolutions to see how scale influences the assessment. Simulation models of land use and land cover change typically examine a landscape at initial points in time t0 and t1 and then predict the change from t1 to some subsequent point in time t2 in order to evaluate the performance of the simulation model. If the predicted map of t2 appears similar to the reference map of t2, it is concluded that the simulation model performed well. However, a strong agreement between the predicted map of t2 and the reference map of t2 does not necessarily indicate that the simulation model provides additional information beyond. If there were no simulation model, then the best prediction of t2 would probably be the map of t1. Therefore, a null model would predict pure persistence (i.e., no change) between t1 and t2 (Pontius, Huffaker, et al., 2004; Pontius, Shusas, et al., 2004). Validation of urban land use location was based on the Kappa index, which is used to compare the reference map with the simulated map or to compare two reference maps. Several measures of agreement between two or more maps have been introduced into the applied statistics literature. The collection known as Kappa coefficients comes from the notion initiated by Scott (1955) that the observed cases of agreement between two maps include some cases for which the agreement was by chance alone. The Kappa statistic is a measure of accuracy that ranges between 0 (completely inaccurate) and 1 (completely accurate) and measures the observed agreement between the classification (or simulation) and the reference map and the agreement between maps that might be attained solely by chance (Munroe, Southworth, & Tucker, 2002). The original form of the definition of a Kappa coefficient is

k¼

Po  Pe 1  Pe

(2)

where Po is the probability that a pixel will be placed in the same land use category in two different maps, while Pe is the probability that a pixel will be placed in the same land use category in two different maps by chance. Therefore, Kappa should be the fraction of all pixels not classified the same in two maps by chance (Aickin, 1990). If the agreement between two maps is perfect, then Kappa ¼ 1; if the observed proportion of pixels in agreement between two maps is greater than expected

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

241

Table 2 Land use transitions at the sub-county level (percent of landscape). 1993 Land cover

2000 Land cover

Total 1993

Loss

0.00 0.00 (0.00) [0.00]

15.39 15.39 (0.00) [0.00]

0.00 0.00 (0.00) [0.00]

0.00 1.13 (1.13) [1.00]

0.00 0.02 (0.02) [1.00]

33.93 33.93 (0.00) [0.00]

1.54 1.54 (0.00) [0.00]

0.00 0.31 (0.31) [1.00]

49.41 49.41 (0.00) [0.00]

0.18 0.01 (0.17) [17.00]

49.89 49.89 (0.00) [0.00]

0.48 0.48 (0.00) [0.00]

0.00 0.03 (0.03) [1.00]

0.00 0.06 (0.06) [1.00]

0.17 0.08 (0.09) [1.13]

0.62 0.01 (0.62) [620.00]

0.79 0.17 (0.62) [3.60]

0.17 0.17 (0.00) [0.00]

Total 2000

17.22 15.97 (1.25) [0.08]

32.39 32.76 (0.37) [0.01]

49.58 50.62 (1.04) [0.02]

0.81 0.03 (0.78) [26.00]

100.00 100.00 (0.00) [0.00]

2.19 2.19 (0.00) [0.00]

Gain

1.83 0.58 (1.25) [2.20]

0.00 0.37 (0.37) [1.00]

0.17 1.21 (1.04) [0.86]

0.19 0.03 (0.16) [5.33]

2.19 2.19 (0.00) [0.00]

Urban

Agriculture

Forest

Others

Urban

15.39 15.39 (0.00) [0.00]

0.00 0.00 (0.00) [0.00]

0.00 0.00 (0.00) [0.00]

Agriculture

1.54 0.39 (1.15) [2.90]

32.39 32.39 (0.00) [0.00]

Forest

0.29 0.16 (0.13) [0.81]

Others

The bold number is the actual percent of the landscape. The number in italics is the percent of the landscape that would be expected if the process of change was random. The number in round parentheses is the actual minus expected percent. The number in square brackets is the number in round parentheses divided by the number in italics.

Table 3 As in Table 2, but for the county level. 1993 Land cover

2000 Land cover

Total 1993

Loss

0.00 0.00 (0.00) [0.00]

4.61 4.61 (0.00) [0.00]

0.00 0.00 (0.00) [0.00]

0.00 0.29 (0.29) [1.00]

0.00 0.01 (0.01) [1.00]

19.41 19.41 (0.00) [0.00]

0.32 0.32 (0.00) [0.00]

0.00 0.24 (0.24) [1.00]

72.56 72.56 (0.00) [0.00]

0.28 0.04 (0.24) [6.00]

72.89 72.89 (0.00) [0.00]

0.34 0.34 (0.00) [0.00]

0.00 0.03 (0.03) [1.00]

0.00 0.12 (0.12) [1.00]

0.59 0.45 (0.14) [0.31]

2.49 2.49 (0.00) [0.00]

3.09 3.09 (0.00) [0.00]

0.60 0.60 (0.00) [0.00]

Total 2000

4.99 4.72 (0.27) [0.06]

19.09 19.45 (0.36) [0.02]

73.15 73.30 (0.15) [0.002]

2.77 2.54 (0.23) [0.09]

100.00 100.00 (0.00) [0.00]

1.26 1.26 (0.00) [0.00]

Gain

0.38 0.11 (0.27) [2.45]

0.00 0.36 (0.36) [1.00]

0.60 0.74 (0.14) [0.19]

0.28 0.05 (0.23) [4.60]

1.26 1.26 (0.00) [0.00]

Urban

Agriculture

Forest

Others

Urban

4.61 4.61 (0.00) [0.00]

0.00 0.00 (0.00) [0.00]

0.00 0.00 (0.00) [0.00]

Agriculture

0.32 0.02 (0.30) [15.00]

19.09 19.09 (0.00) [0.00]

Forest

0.06 0.06 (0.00) [0.00]

Others

242

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

Table 4 Land use transitions at the sub-county level (hectares). 1993 Land cover

Urban Agriculture Forest Others Total 2000 Gain Total change Swap Net change

2000 Land cover Urban

Agriculture

Forest

Others

5983 600 115 0 6698 715 715 0 715

0 12,594 0 0 12,594 0 600 0 600

0 0 19,215 66 19,282 66 252 133 118

0 0 70 243 313 70 137 133 4

Urban

Agriculture

Forest

Others

13,379 931 170 1 14,482 1103 1103 0 1103

0 55,440 7 0 55,448 7 950 15 935

0 10.8 210,748 1728 212,487 1738 2716 1955.5 761

0 0 799 7244 8044 799 2529 1599 929

Total 1993

Loss

5983 13,194 19,400 309 38,888

0 600 185 66

Total 1993

Loss

13,379 56,383 211,725 8973 290,462

0 942 977 1729

Table 5 As in Table 4, but for the county level. 1993 Land cover

Urban Agriculture Forest Others Total 2000 Gain Total change Swap Net change

2000 Land cover

Table 6 Beta values of the explanatory variables of urban land use location. Explanatory variable

Sub-county Beta

Distance from water networks Slope Elevation Distance from roads Distance from State College township Distance from Bellefonte township Distance from Milesburg township Distance from PortiMatilda township Soils suitable for agriculture Soils suitable for septic works

County Constant

ROC

1.510

0.88

Beta

Constant

ROC

1.21

0.90

0.001 0.045 0.006 0.001 0.001

0.080 0.397

0.002 0.001 0.001 0.152

All variables significant at p < 0.01.

Table 7 CLUE-S and null model validations at the sub-county level. Index

2000 Reference map and 2000 simulated map

2000 Reference map and 1993 reference map

Kno Klocation Kquantity Kstandard

0.94 0.93 0.99 0.93

0.97 0.99 0.94 0.96

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

243

Table 8 CLUE-S and null model validations at the county level. Index

2000 Reference map and 2000 simulated map

2000 Reference map and 1993 reference map

Kno Klocation Kquantity Kstandard

0.72 0.76 0.88 0.58

0.61 0.49 0.76 0.44

proportion correct due to chance, then Kappa > 0; if the observed proportion of pixels in agreement is equal to the expected proportion agreement due to chance, then Kappa ¼ 0; and if the observed proportion in correct agreement is less than the expected proportion due to chance, then Kappa < 0. A variation of the Kappa statistic (Pontius, 2002) is

k¼

Po  Pc Pp  Pc

(3)

where Po is the observed proportion correct, Pc is the expected proportion correct due to chance, and Pp is the proportion correct with perfect match between two maps. In addition to the standard Kappa index of agreement, Pontius (2000, 2002) defines three variations: Kappa for no information (Kno), Kappa for location (Klocation), and Kappa for quantity (Kquantity). Kno is an overall index of agreement, Klocation is an index that measures the agreement in terms of location only, and Kquantity measures the agreement in terms of quantity. According to Pontius (2000), a Kappa value higher than 0.5 can be considered ‘‘satisfactory’’ for land use change modeling. Similarly, Landis and Koch (1977) characterize agreement as follows: values > 0.75 are very good to excellent, values between 0.4 and 0.75 are fair to good, and values of 0.4 or less indicate poor agreement. Results Land use transitions as proportions of the landscape at sub-county and county levels are shown in Tables 2 and 3, whereas area transitions between different land use categories at sub-county and county levels for the period 1993–2000 are shown in Tables 4 and 5. Ninety-eight percent of the landscape in Centre County persisted between 1993 and 2000 (Table 3) and all land use transitions are systematic except Forest to Urban at the county level (Tables 2 and 3). The proportion of the landscape under urban land use increased from 15.4 to 17.2 percent, an area increase of 715 ha at the sub-county level mainly from agricultural land use loss (Tables 2 and 4). In contrast, increase in urban land use at the county level was only 0.5 percent of the landscape, but an area increase of 1103 ha (Tables 3 and 5). Area under forest land use increased by 119 and 761 ha at subcounty and county levels, respectively, from the Others land use category. Forest land use experienced positional swap with

Cell specific characteristics

General land use characteristics LU conversion elasticities ELASu

Decision rules (protected area)

Allocation Total probability TPROPi.u

Demanded area per land use type

Iterarion loop Probability Pi.u

Iteration variable ITERu

Fig. 4. Representation of the iterative procedure for land use change allocation (Verburg et al., 2002).

244

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

Fig. 5. Urban land use location simulation results at the sub-county level.

the Others land use category of 134 ha at sub-county while Agriculture, Forest and the Others land use categories experience positional swaps of 15, 1955, 1599 ha respectively at county level (Tables 4 and 5). The transition of agricultural land to urban (built environment) is the dominant land use change in Centre County. Explanatory variables (underlying drivers) of urban land use location in Centre County are shown in Table 6. Beta values are logistic regression standardized coefficients of the independent variables. Slope has a negative effect on urban land use at the sub-county level but not at the county level. In contrast, soil suitability for agricultural production is a positive determinant of urban land use location at sub-county and county levels, while soil suitability for septic works is a determinant of urban land use location at the sub-county level. The results support the conclusion that land use transition occurs where urban land use gain was mainly from agricultural land that is located in fertile valleys. Topography was not a limiting factor of urban land use location at the county level, implying that pressure from urban development is lower at this scale when compared to the sub-county level; hence, there is no need to develop land in steep terrain. Because urban development was a small proportion of the landscape over the study period, the simulated 2000 urban land use location appears to be similar to the 2000 reference map of urban land use location. Nevertheless, the future urban land use location increase in Centre County is identifiable (Figs. 5 and 6). Therefore, the explanatory variables of urban location explain urban land use location at county level. The null model performs better than the simulation model in all aspects but

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

245

Fig. 6. Urban land use location simulation results at the county level.

Kquantity at the sub-county level, while in contrast the simulation model performs better than the null model at the county level (Tables 7 and 8) with 39 percent of the landscape showing location agreement between the 2000 reference map and the simulated map of 2000 land use location, and 42 percent of the landscape showing location agreement between the 1993 and the 2000 reference maps at the sub-county level (Figs. 7 and 8). Thirty percent of the landscape shows location agreement between the 2000 reference map and the simulated map of 2000 land use location; 20 percent of the landscape shows location agreement between the 1993 and the 2000 reference maps at the county level (Figs. 9 and 10). Therefore, for the study period, the 1993 reference map would have been a good proxy of 2000 urban land use patterns at the sub-county level, whereas the 2000 simulated urban land use location at the county level would have been a better representation of 2000 urban land use location. Discussion Land change, predominantly agriculture to urban, was only 2 percent of the Centre County landscape for the period 1993– 2000. This result is also consistent with findings by Chen et al. (2002), Geoghegan et al. (2001), Mertens and Lambin (2000), and Schneider and Pontius (2001), all of whom reported high persistence and limited change in land cover/land use change

246

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

100

Percent of Landscape

90 80 70 60 50

Disagreement due to quantity Disagreement due to location Agreement due to location

40 30 20

Agreement due to quantity Agreement due to chance

10 0 Fig. 7. Agreement and disagreement between the 2000 reference map and the 2000 simulated map at the sub-county level.

studies. The result further supports the assertions by Pontius, Huffaker, et al. (2004), Pontius, Shusas, et al. (2004), Yang (2002), and Yang and Lo (2002) that even fast-growing urbanizing areas, such as the Atlanta Metropolitan Area of the United States, have experienced only 25 percent of land change to urban over the last decades although most of the change was located in prime agricultural areas. Urban land gain from agriculture was the dominant land use transitions in Centre County, which is consistent with findings by Hill (1986), United Nations (1995), and Verburg, Veldkamp, and Bouma (1999), among others. Doos (2002) notes that urban expansion is an ongoing threat to farmland because urban areas tend to have been founded in agricultural areas. The methodology further provides information on swap in location within and between land use categories, thereby completing the traditional information on net change. Area under forest showed some increase due to regeneration of previously cleared areas (grasslands) and due to aforestation of abandoned mining sites. The results demonstrate that biophysical factors such as topography, soil suitability for agricultural production and septic works, and proximity to population are the major explanatory variables of urban land use location in the county, whereas zoning and population density are not. This finding agrees with the assertion by Pontius and Spencer (2005) that whatever the variation in economic, social, and legal explanatory variables across time, topography and geologic characteristics are supreme in determining land use location. Lambin et al. (2003) observed that biophysical explanatory variables define the natural capacity or predisposing conditions for land use change in a given locality. Smith and Reynolds (2002) also noted that, for any given human–environment system, a limited number of explanatory variables are required for predicting the general trend in land use. Here, soil factors emerged as one of the main explanatory variables of urban land use location, which agrees with reviews by Rudel (2005) and Wood and Porro (2002). Levia (1998), Nelson (1992) and Platt (1985) found that much of the land lost to urbanization is in prime agricultural land located on coastal plains and river valleys. In Centre County, however, there were differences in explanatory variables of urban land use location at sub-county and county levels: topography had a negative effect on urban land use location at the sub-county level, but had no effect at the county level. This finding underscores the need to analyze sprawl at multiple resolutions in order to understand how land use change occurs over distance. Simulated patterns of urban land location in Centre County for 2000 are visually similar to the 2000 urban land use reference map, thus emphasizing the significance of land use change modeling and simulation in sprawl analysis and management. This similarity is further supported by validation results that show high Kappa indexes between the reference 100

Percent of Landscape

90 80 70 60 50

Disagreement due to quantity Disagreement due to location Agreement due to location

40 30 20

Agreement due to quantity Agreement due to chance

10 0 Fig. 8. Agreement and disagreement between the 2000 and 1993 reference maps at the sub-county level.

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

247

100

Percent of Landscape

90 80 70 60 50

Disagreement due to quantity Disagreement due to location Agreement due to location

40 30 20

Agreement due to quantity Agreement due to chance

10 0 Fig. 9. Agreement and disagreement between the 2000 reference map and the 2000 simulated map at the county level.

100

Percent of Landscape

90 80 70 60 50

Disagreement due to quantity Disagreement due to location Agreement due to location

40 30 20

Agreement due to quantity Agreement due to chance

10 0 Fig. 10. Agreement and disagreement between the 2000 and 1993 reference maps at the county level.

maps and the simulated maps, although the null model is better than the simulation model at the sub-county level. Better performance of the null model when compared to the simulation model at the sub-county level is consistent with findings by Brown, Goovaerts, Burnicki, and Li (2002), Chen et al. (2002), Geoghegan et al. (2001), Lo and Yang (2002), and Schneider and Pontius (2001). These authors all reported greater agreement between the reference map of t1 and reference map of t2 than the agreement between the predicted map of t2 and the reference map of t2 because of high persistence in the landscape. The null model outperforms the simulation model at the sub-county level, where a significant proportion of the landscape underwent change when compared to the county level. The inability of the simulation model to capture land use change at the sub-county level suggests that the spatial resolution of analysis was too coarse to capture the conversion of land to the built environment that is likely to take place in small lot sizes in this highly urbanized landscape. In contrast, the county level includes suburban and rural area land conversion, which involves large lot sizes with horizontal dimensions greater than 250 m, and is therefore able to capture this process. Conclusions In a contribution to the realization of a multi-tool methodology for sprawl analysis, this paper first applies a crosstabulation matrix to assess land use change according to two pairs of components: net change in land quantity and swap in land location, as well as gross gains and gross losses. This analysis allowed the differentiation of land use change from landscape persistence and enabled the extraction of the small proportion of the landscape that underwent change from the large proportion that was static. The analysis further identified urban (the built environment) as the land use that experienced the most dominant and systematic land gain in the study area. The paper further determined explanatory variables of urban land use location through logistic regression and subsequently used the explanatory variables to project urban land use patterns into the near future through simulation modeling using the CLUE-S modeling framework. The results showed that land use change and landscape fragmentation in Centre County were dominated by transitions from agricultural land use to urban use and by swaps in location between Forest and Others land use categories. Biophysical factors, such as soil suitability for agricultural production and topography, were key determinants of urban land use location in Centre County. The CLUE-S model was able to simulate urban land use location at the county level, but simulations at the sub-county level were wanting.

248

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

In short, the methodology presented in this paper is versatile and adds to the array of methods available for sprawl analysis. Its success suggests that it could be possible to develop analytical frameworks that would allow researchers not only to slide easily from local to regional and national scales, but also to link these scales to establish a more holistic understanding of urban sprawl.

References Aickin, M. (1990). Maximum likelihood estimation of agreement in the constant predictive probability model, and its relation to Cohen’s Kappa. Biometrics, 46, 293–302. Alberti, M. (1999). Urban patterns and environmental performance: what do we know? Journal of Planning Education and Research, 19, 151–163. Bartlett, J. G., Mageean, D. M., & O’Connor, R. J. (2000). Residential expansion as a continental threat to U.S. coastal ecosystems. Population and Environment, 21, 243–258. Batty, M., Xie, Y., & Sun, Z. (1999). Modeling urban dynamics through GIS-based cellular automata. Computers, Environments and Urban Systems, 23, 205–233. Benfield, F. K., Raimi, M. D., & Chen, D. D. (1999). Once there were green fields: How urban sprawl is undermining America’s environment, economy and social fabric. Washington, DC: Natural Resources Defense Council. Bilsborrow, R. E., & Ogondo, H. O. (1992). Population-driven changes in land use in developing countries. Ambio, 21, 37–45. Bingsheng, K. (1996). Regional inequality in rural development. In R. Garnaut, G. Shutian, & M. Guonan (Eds.), The third revolution in the Chinese countryside. Cambridge University. Brown, D. G., Goovaerts, P., Burnicki, A., & Li, M. Y. (2002). Stochastic simulation of land-cover change using geostatistics and generalized additive models. Photogrammetric Engineering & Remote Sensing, 68, 1051–1061. Burchell, R. W., & Shad, N. A. (1999). The evolution of the sprawl debate in the United States. Hastings West-Northwest Journal of Environmental Law and Policy, 5(2), 137–160. Carlson, T. N., & Arthur, S. T. (2000). The impact of land – cover changes due to urbanization on surface microclimate and hydrology: a satellite perspective. Global and Planetary Change, 25, 49–65. Centre County Planning Commission. 2005. Annual Report. Bellefonte, PA. Chen, J., Gong, P., He, C., Luo, W., Tamura, M., & Shi, P. (2002). Assessment of the urban development plan of Beijing by using CA-based urban growth model. Photogrammetric Engineering & Remote Sensing, 68, 1063–1071. Couch, C., & Karecha, J. (2006). Controlling urban sprawl: some experience from Liverpool. Cities, 23, 353–363. Cutsinger, J., Galster, G., Wolman, H., Hanson, R., & Towns, D. (2005). Verifying the multi-dimensional nature of metropolitan land use: advancing the understanding and measurement of sprawl. Journal of Urban Affairs, 27, 235–259. Daniels, T. (1997). Where does cluster zoning fit in farmland protection? Journal of the American Planning Association, 63, 129–137. Doos, B. R. (2002). Population growth and loss of arable land. Global Environmental Change, 12, 303–311. El Nasser, H., & Overberg, P. (22 February, 2001). A comprehensive look at sprawl in America. USA Today 1. Fang, S., Gertner, G. Z., Sun, Z., & Anderson, A. A. (2005). The impact of interactions in spatial simulation of the dynamics of urban sprawl. Landscape and Urban Planning, 73, 294–306. Florida Division of Community Affairs. (1993). Local government and comprehensive planning land development regulation act of 1985. Tallahassee, FL. Folke, C., Jansson, A., Larsson, J., & Constanza, R. (1997). Ecosystem appropriation by cities. Ambio, 26, 167–172. Galster, G., Hanson, R., Ratcliffe, M., Wolman, H., Coleman, S., & Freihage, J. (2001). Wrestling sprawl to the ground: defining and measuring an elusive concept. Housing Policy Debate, 12, 681–718. Geoghegan, J., Villar, S. C., Klepeis, P., Mendoza, P. M., Ogneva-Himmelberger, Y., Chowdhury, R., et al. (2001). Modeling tropical deforestation in the southern Yucatan peninsula region: comparing survey and satellite data. Agriculture, Ecosystems & Environment, 85, 25–46. Geurs, K. T., & van Wee, B. (2006). Ex-post evaluation of thirty years of compact urban development in the Netherlands. Urban Studies, 43, 139–160. Gibson, C., Orstrom, E., & Ahn, T. K. (1998). Scaling issues in the social sciences. IHDP Working Paper No. 1. Bonn. Goetz, S. J., Jantz, C. A., Prince, S. D., Smith, A. J., Wright, R., & Varlyguin, D. (2004). Integrated analysis of ecosystem interactions with land use change: the Chesapeake Bay watershed. In R. S. Defries, G. P. Asner, & R. A. Houghton (Eds.), Ecosystems and land use change. Geophysical monograph series (pp. 263–275). Washington, DC: American Geophysical Union. Hasse, J. E. (2004). A geospatial approach to measuring new development tracts for characteristics of sprawl. Landscape Journal, 23, 52–67. Hasse, J., & Lathrop, R. G. (2003). A housing-unit level approach to characterizing residential sprawl. Photogrammetric Engineering & Remote Sensing, 69, 1021–1030. Heimlich, R. E., & Anderson, W. D. (2001). Development at the urban fringe and beyond: Impacts on agricultural and rural land. Economic Research Service, U.S. Department of Agriculture. Agricultural Economic Report No. 803. Heistermann, M., Muller, C., & Ronneberger, K. (2006). Achievements, deficits and potentials of continental to global scale land-use modeling. Agriculture, Ecosystems & Environment, 114, 141–158. Hill, R. D. (1986). Land use change on the urban fringe. Nature and Resources, January–June, 24–33. Ichikawa, K., Okubo, N., Okubo, S., & Takeuchi, K. (2006). Transition of the satoyama landscape in the urban fringe of the Tokyo metropolitan area from 1880 to 2001. Landscape and Urban Planning, 78, 398–410. Jantz, P., Goetz, S., & Jantz, C. (2005). Urbanization and the loss of resource lands in Chesapeake Bay watershed. Environmental Management, 36, 808–825. Kaimowitz, D., & Angelsen, A. (1998). Economic models of tropical deforestation – A review. Bogor Centre for International Forest Research. Kok, K., & Veldkamp, A. (2001). Evaluating impact of spatial scales on land use pattern. Agriculture, Ecosystems & Environment, 85, 205–221. Lambin, E. F. (1994). Modelling deforestation processes: a review. Research Report No. 1. Trees, tropical ecosystem environment observation by satellites. European Commission Joint Research Centre/European Space Agency. Lambin, E. F., Geist, H. J., & Lepers, E. (2003). Dynamics of land-use and land-cover change in the tropical regions. Annual Review of Environment and Resources, 28, 205–241. Lambin, E. F., Turner, B. I., & Geist, H. J. (2001). The causes of land-use and land-cover change: moving beyond the myths. Global Environmental Change, 11, 261–269. Landis, J. R., & Koch, G. G. (1977). The measurement of observer agreement for categorical data. Biometrics, 33, 159–174. Lathrop, R. G. (2004). Land use/land cover update to year 2000/2001. Trenton, NJ: New Jersey Department of Environmental Protection. http://crssa.rutgers. edu/projects/lc/download/reportsdata0001/landuselandcov web.pdf. Accessed June 2008. Lathrop, R. G., Tulloch, D. L., & Hatfield, C. (2007). Consequences of land use change in the New York–New Jersey highlands, USA: landscape indicators of forest and watershed integrity. Landscape and Urban Planning, 79, 150–159. Levia, D. F. (1998). Farmland conversion and residential development in North Central Massachusetts. Land Degradation and Development, 9, 123–130. Levin, S. A. (1993). Concepts of scale at the local level. In J. R. Ehleringer, & C. B. Field (Eds.), Scaling physiological processes. Leaf to globe. San Diego: Academic Press. Lo, C. P., & Yang, X. (2002). Driver of land-use/land-cover changes and dynamic modeling for the Atlanta, Georgia Metropolitan area. Photogrammetric Engineering & Remote Sensing, 68, 1073–1082. Lopez, R., & Hynes, H. P. (2003). Sprawl in the 1990s: measurement, distribution, and trends. Urban Affairs Review, 38, 325–355. McKinney, M. L. (2002). Urbanization, biodiversity and conservation. Bio-science, 52, 883–890.

N. Batisani, B. Yarnal / Applied Geography 29 (2009) 235–249

249

Mertens, B., & Lambin, E. (2000). Land-cover-change trajectories in southern Cameroon. Annals of the Association of American Geographers, 90, 467–494. Mieskowski, P., & Mills, E. (1991). The causes of metropolitan suburbanization. Journal of Economic Perspectives, 7, 135–147. Munroe, D. K., Southworth, J., & Tucker, C. M. (2002). The dynamics of land-cover change in western Honduras: exploring spatial and temporal complexity. Agricultural Economics, 27, 355–369. Nelson, A. (1992). Preserving prime farmland in the face of urbanization: lessons from Oregon. Journal of the American Planning Association, 8, 467–488. Nilsson, C., Pizzuto, J. E., Moglen, G. E., Palmer, M. A., Stanley, E. H., Bockstael, N. E., et al. (2003). Ecological forecasting and the urbanization of stream ecosystems: challenges for economists, hydrologists, geomorphologists and ecologists. Ecosystems, 6, 659–674. O’Neill, R. V., Turner, S. J., Cullinan, V. I., Coffin, D. P., Cook, T., Conley, W., et al. (1991). Multiple landscape scales: an intersite comparison. Landscape Ecology, 5, 137–144. Platt, R. H. (1985). The farmland conversion debate: NALS and beyond. Professional Geographer, 34, 433–442. Pontius, R. G. (2000). Quantification error versus location error in comparison of categorical maps. Photogrammetric Engineering & Remote Sensing, 66, 1011–1016. Pontius, R. G. (2002). Statistical methods to partition effects of quantity and location during comparison of categorical maps at multiple resolutions. Photogrammetric Engineering & Remote Sensing, 68, 1041–1049. Pontius, R. G., Huffaker, D., & Denman, K. (2004). Useful techniques of validation for spatially explicit land-change models. Ecological Modelling, 179, 445–461. Pontius, R. G., & Schneider, L. (2001). Land-cover change model validation by a ROC method for the Ipswich watershed, Massachusetts, USA. Agriculture, Ecosystems & Environment, 85, 239–248. Pontius, R. G., Shusas, E., & McEachern, M. (2004). Detecting important categorical land changes while accounting for persistence. Agriculture, Ecosystems & Environment, 101, 251–268. Pontius, R. G., & Spencer, J. (2005). Uncertainty in extrapolations of predictive land-use change models. Environment and Planning B: Planning and Design, 32, 211–230. Rastetter, E. B., King, A. W., Cosby, B. J., Hornberger, G. M., O’Neill, R. V., & Hobbie, J. E. (1992). Aggregating fine-scale ecological knowledge to model coarserscale attributes of ecosystems. Ecological Applications, 2, 55–70. Rudel, T. K. (2005). Tropical forests: Regional paths of destruction and regeneration in the late twentieth century. New York: Columbia University Press. Schneider, L., & Pontius, R. G. (2001). Modeling land-use change in the Ipswich watershed, Massachusetts, USA. Agriculture, Ecosystems & Environment, 85, 83–94. Scott, W. A. (1955). The reliability of content analysis: the case of nominal scale coding. Public Opinion Quarterly, 19, 321–325. Smith, S., & Reynolds, J. F. (2002). Desertification: a new paradigm for an old problem. In J. F. Reynolds, & D. M. Stafford Smith (Eds.), Global desertification: Do humans cause deserts? (pp. 403–424). Berlin: Dahlem University Press. Torrens, P. M., & Alberti, M. (2000). Measuring sprawl. Paper 27. London, UK: Center for Advanced Spatial Analysis, University College London. p. 43. Tsai, Y. (2005). Quantifying urban form: compactness versus ‘sprawl’. Urban Studies, 42, 141–161. Turner, B. L., II, Skole, D. L., Sanderson, S., Fischer, G., Fresco, L. O., & Leemans, R. (1995). Land-use and land-cover change. Science/research plan. IGBP Report No. 35 and HDP Report No. 7. Stockholm/Geneva, 132 pp. United Nations. (1995). The challenge of urbanization: The world’s cities. New York: United Nations. United Nations. World urbanization prospects: The 2003 revision data tables and highlights. New York: United Nations; 2004. Veldkamp, A., & Lambin, E. F. (2001). Predicting land-use change. Agriculture, Ecosystems & Environment, 85, 1–6. Veldkamp, A., Verburg, P. H., Kok, K., de Koning, G. H. J., Priess, J., & Bergsman, A. R. (2001). The need for scale sensitive approaches in spatially explicit land use change modeling. Environmental Modeling and Assessment, 6, 111–121. Verburg, P. H., Soepboer, W., Veldkamp, A., Limpiada, R., Espaldon, V., & Mastura, S. A. (2002). Modeling the spatial dynamics of regional land use: the CLUES model. Environmental Management, 30, 391–405. Verburg, P. H., Veldkamp, A., & Bouma, J. (1999). Land use change under conditions of high population pressure: the case of Java. Global Environmental Change, 9, 303–312. Wang, G., Gertner, G., Howard, H., & Anderson, A. (2008). Optimal spatial resolution for collection of ground data and multi-sensor image mapping of soil erosion cover factor. Journal of Environmental Management, 88, 1088–1098. Wassenaar, T., Gerber, P., Verburg, P. H., Rosales, M., Ibrahim, M., & Steinfeld, H. (2006). Projecting land use changes in the Neotropics: the geography of pasture expansion in forest. Global Environmental Change, 17, 86–104. Wiens, J. A. (1989). Spatial scaling in ecology. Functional Ecology, 3, 385–397. Wood, C. H., & Porro, R. (Eds.). (2002). Deforestation and land use in the Amazon. Gainesville, FL: University Press of Florida. World Resources Institute. (2003). Domesticating the world: Conversion of natural resources. Wu, F. (2002). Calibration of stochastic cellular automata: the application to rural–urban land conversions. International Journal of Geographical Information Science, 16, 795–818. Yang, X. (2002). Satellite monitoring of urban spatial growth in the Atlanta metropolitan area. Photogrammetric Engineering & Remote Sensing, 68, 725–734. Yang, X., & Lo, C. P. (2002). Using a time series of satellite imagery to detect land use and land cover changes in the Atlanta, Georgia metropolitan area. International Journal of Remote Sensing, 23, 1775–1798.