Delaware River Basin Project
 Delaware River Basin Project | Home | Urban Growth Modeling DRAFTMODELING URBAN EXPANSION IN THE PHILADELPHIA METROPOLITAN AREAReport from the USGS Urban Dynamics Philadelphia Case StudyDalia Varanka, Research Geographer U.S. Geological Survey Mid-Continent Geographic Science Center 1400 Independence Road Rolla, Missouri 65401 Tel: 573.308.3897 Email: dvaranka@usgs.gov Database The database was built by scanning USGS quadrangle maps of the twentieth century at scales of 1:24,000 and 1:62,500, and has been organized into decades roughly corresponding to 1940 through 1980. The data themes, taken from the land cover classification system of Anderson, et al (1976) are: urban or built-up land, principle transportation, hydrography, elevation, and woodland. The collection procedure involved scanning USGS topographic maps and rescaling them to fit the fifteen-minute quadrangle series. The information from the map was separated into different thematic layers, and the images were assigned attributes, in GIS, of time and theme. The collection criteria and compilation techniques for the thematic layers is described in Bradford and Vincent (1998), which is available at the project Internet home page. In addition to scanned maps, two land cover datasets are available. One is developed from the USGS Land Use/Land Cover program of the 1970s and the other is from the Multi-Resolution Landcover Characteristic program (MRLC), 1990 - present. The database has been used in USGS Geologic Division and USGS Water Resources Division projects, among others. Modeling Researchers usually use modeling of one form or another - conceptual, mathematical, or physical - in their work. Mathematical, or quantitative models are highly controversial, much modeling is difficult to verify because it is oriented toward the future and cannot be empirically measured. Current technology may not be adequate to process large databases of multiple variables, nor may the data be available in an appropriate form. Despite these challenges, the experience and theoretical organization resulting from quantitative modeling may be equal to the benefits of their predictive powers. Besides serving as a measurement tool, models act as an organizational framework and increase the investigative understanding of an urban system, as well as expedite analysis (Zeigler 1984). The researchers at MCMC, in their participation in the Philadelphia case study found this perspective, described above, applicable to their work. Thus, the discussion within this paper will not focus only on predictive or quantitative aspects of the Clarke Urban Growth Model (UGM), but on social or theoretical, as well as computational, aspects of modeling that the Urban Dynamics research manifested. The following sections describe the project, analysis, and concluding remarks. The Clarke Urban Growth Model The Clarke Urban Growth model processes raster data files according to cellular automata (CA) growth rules, and according to other parameters as well. Multiple iterations of the process are done to produce Monte Carlo simulation. The output is compared to the historical record to select growth parameters that best describe urban growth in the metropolitan area. Originating as a mathematical game in Conroy's "Game of Life" (Conway, 1970), CA theory was first applied as a growth modeling technique in the biological sciences. Influenced by the scientific method in their thinking, urban planners brought CA theory to urban modeling with promising possibilities (Couclelis, 1997). CA models require three elements: an organizing field of action, usually consisting of a grid (raster) over which the data pixels are dispersed; starting conditions at a set point in time; and rules of growth applied to the individual data cells. Modifications for each cell depend on the status of neighboring cells. CA is believed to be applicable to any natural system. Because the resolution of the grid cells can be modified, these models are considered to be scale-independent (Batty and Xie, 1994). Although cellular-spaced modeling is characterized by homogeneity and uniformity of structure because the state of a cell is determined by the state of its neighboring cells and are determined by defined growth/death/birth rules that are applied uniformly without exception, homogeneity and uniformity of structure can be modified in CA models by adding input variables and defining the set of neighbors to cells that are not geometrically adjacent. If this is done, then the behavior of cells is no longer uniform, and each cell or group of cells requires their own unique transition functions (Zeigler, 1984, p. 86). In a similar way to this, a multi-factored CA urban growth model was accomplished in the Clarke UGM. The growth input factors and rules that were chosen for the UGM were intended to duplicate urban growth characteristics. The assumptions of the UGM growth rules that were embedded in the model structure are as follows:
Breed coefficient 01 Spread coefficient 74 Slope coefficient 06 Roads coefficient 45 The validity of the projections will be judged by pattern-matching the output calculated from earlier decades to other existing datasets of a similar time period. The calibrations for the entire Philadelphia-Trenton-Wilmington metropolitan area dataset will be done in 2001. Discussion The effectiveness of the UGM must be considered in several contexts. Some factors by which models can be tested are as follows. 1) Examine the formulation of the problem, 2) question whether the mathematical expressions are dimensionally consistent, 3) vary input parameters and determine if model output is plausible, and 4) apply a retrospective test - compare output to historically reconstructed data (Rubinstein 1981, p.5). All of these were interrelated factors in an operational tension that was found inherent to the UGM. As a result of our experience working with the Clarke UGM, a number of issues became apparent as areas for further research. The problem, as formulated by the UGM, was to predict future growth with relative accuracy, or to duplicate certain 'realism.' To model is to explore the realism of the representation. From some viewpoints, realism is not required in simulation; what matters most is that a correlation can be made between results and tests. Such results were demonstrated for the UGM (Clarke and Gaydos, 1998). There also is an implied or inherent claim that the model addresses factors of urban growth. The UGM addresses realism with input factors of existing urbanization, roads, excluded areas (most often hydrography) and topographical slope, combined with non-quantitative random deterministic CA expansion. Thus, it is fair to question whether the growth rules adequately include or describe urban decentralization. The input factors are plausible, but no attempt is made to realistically quantify or locate growth calculations. For example, one of the problems with the model is that it assumes that previous growth persists, when in fact the built-up environment experiences life cycles that lead to the deterioration of older urban land cover forms under certain management conditions. Clarke's UGM lacks cell 'deaths,' which were an original assumption of Conroy's "Game of Life." The topographical input factors strongly emphasize the importance of topography, relative to social factors, for growth scenarios. The background to such a conceptual foundation, that geographic phenomena are predominantly products of the physical environment, dates to the nineteenth century when urban geographers distinguished between topography and geography, the first being strictly environmental controls and the latter being situational. The roots of the topographic perspective on urban growth are in environmental determinism (Berry and Kasarda, 1977, p. 8-9). The input factors selected for the model would suggest traditional theory of urban growth, drawn from theories of urban ecology, or the 'Chicago School.' This perspective (theory) of urban expansion draws from environmental principles in using controls from environmental features and ecological processes. But later in the twentieth century, even these approaches maintained that urban studies cannot exempt social factors. One summary of contemporary urban ecological inquiry is as follows: The concern of the ecological approach with social system [city] growth and development may be seen in the contributions contemporary ecologists have made toward understanding the process of expansion... If system change is to be cumulative, it is necessary that culture, population, territory, and organization all advance together (Berry and Kasarda, 1977, p. 15). One variation between the Clarke model and the Chicago school is that the Chicago School assumes growth expands from a single city center. The Clarke model creates new clusters of urban growth, similar to the alternative scenario of polycentric urbanization. However, polycentricism most often occurs when existing urban settlements expand, and not by the boom growth of new clusters. 2. The use of CA by the UGM assumes CA applies to urban growth modeling. Based on this assumption, urbanization must be regarded as a natural system to which CA is applicable, but the UGM has offered a limited set of mathematical expressions with which to compare to the mathematical expressions of natural systems. 3. The UGM cannot vary input factors to test for modeling output plausibility, only for their best-fit, calculation parameters. Since the calibration of the UGM selects parameters that best fit the documented historical record, and the Monte Carlo method inherently rejects alternative outcomes, resulting, for example, from behavior change. Thus, the model does not allow variability, either in the form of input factors or non-equilibrium change. Because it is rule-based and not behavior oriented, the model is more likely to replicate existing conditions than to extrapolate future possible conditions that may result from the interaction between systems and choices. The calibration method proves that CA parameters can work if the future is a true replica of the past, but it does not necessarily work if the system changes. 4. The UGM employs a retrospective test in which output is compared to historically reconstructed data. The calibration of the model is largely a historical process. Although it basically begins with matching the pixels of urban land cover from a representative and a projected data layer, the assumptions interpolated from the match affect the control parameters and evaluated logic of the conclusions. Although done digitally, these are all subjective historical and perceptual interpretations from secondary sources; however, the modeling methodology omits any mention of social science factors in the process as well. The introduction of information and communication technology, and the institutional restructuring of capitalist institutions could be called the transition from an Industrial to Information Age. These factors are changing the spatial patterns of cities (Castells 1989). The UGM is non-behavioral and non-dynamic, and thus allows for no exploration of options, a critical modeling function. This is a weakness of the historically calibrated model that other spatial urban models try to avoid (Landis, 1994). Despite the weakness of urban theory in the UGM, environmental researchers seeking an urban growth model to which to link local environmental change models due to land use change have found the Clarke UGM acceptable for their work. An example of this can be seen in the work of Traci Arthur, of the Pennsylvania State University. (See: http://www.essc.psu.edu/~dajr/chester/index.htm.) Further research MCMC investigators plan to further this research project by comparing modeling projection output of earlier decades to alternative datasets to estimate the degree of similarity between urban growth projections and available historical records. It is expected that the modeling projections generated with the Clarke model will be used by a Federal environmental monitoring and research partnership to consider future depletion of natural resources in the Delaware River basin. The urban growth modeling output will be linked with process-based hydrological and forest ecosystem models, particularly carbon-cycle models. The UGM projections will also be compared to growth forecasts made by other processes. As a result of this three-year pilot project, further research on urban growth modeling continues at the USGS Mid-Continent Mapping Center for the purpose of improving USGS environmental research. Research intended to further these objectives will involve the design of simulation techniques, the presentation of simulation experiments and their results, and discussions of analysis and conclusions, including the range of applications of models, their validity, and running cost. Acknowledgements Extensive credit is due to Brian Maddox and Chris Bilderbeck for implementing and improving the modeling of the Philadelphia area data. Leona Rosenow provided various contributions, including modeling. REFERENCES Batty, M. and Xie, Y. (1994). From cells to cities. Environment and Planning B: Planning and Design 21: 531-548. Berry, B.J.L. and Kasarda, J. D. (1977). Contemporary Urban Ecology. Macmillan Publishing Co., Inc. New York. Bradford, R. and Vincent, D. (1998) Development of Temporal Mapping Techniques to Support Urban Retrospectives Database. Cadwallader, M. (1996). Urban Geography, An Analytical Approach. Prentice-Hall, Inc., Upper Saddle River, NJ. Castells, M. (1989). The Informational City; Information Technology, Economic Restructure, and the Urban-Regional Process. Basil Blackwell, Cambridge Ma. Clarke, K.C., and Gaydos, L. J. (1998). Loose-coupling a cellular automaton model and GIS: long-term urban growth prediction for San Francisco and Washington/Baltimore, International Journal of Geographical Information Systems 12: 699-714. Conway, John. (1970). Scientific American. April: 120. Couclelis, H. (1997). From cellular automata to urban models: new principles for model development and implementation Environment and Planning B - Planning and Design 24: 165-174. Landis, J.D. (1994). The California Urban Futures Model: a new generation of metropolitan simulation models. Environment and Planning B: Planning and Design 21(4). Orfield, M. (1997). Metropolitics, A Regional Agenda for Community and Stability. Washington, D.C. The Lincoln Institution of Land Policy, Brookings Institution Press, and Cambridge, Ma. Rubinstein, Ruven Y. (1981). Simulation and the Monte Carlo Method. John Wiley & Sons, New York. Varanka and Maddox. (2000). Additional Perspectives on Urban Growth Modeling: Some Research Results from the Urban Dynamics Program. NMD Research Symposium 2000. 158-167. Zeigler, B. (1984). Theory of Modelling and Simulation Robert E. Krieger Publishing Company, Inc., Malabar, Fl. top |
U.S. Department of the Interior |
U.S. Geological Survey
URL: http://mcmcweb.er.usgs.gov/de_river_basin/phil/ud_proj_report1200.html
Page Contact Information: Contact USGS
Page Last Modified: Wednesday, 09-Jan-2008 14:02:12 EST
