Modelling the Impacts of Different Policy Scenarios on Urban Growth in Lanzhou with Remote Sensing and Cellular Automata

Xu Xibaoa,b Zhang Fengc Zhang Jianminga* aKey Laboratory of Western China’s Environmental Systems(MOE) & College of Earth and Environmental Sciences, Lanzhou University, Lanzhou 730000, P.R. China bSchool of Geography Science, Nantong University, Nantong 226007, P.R. China cUrban Studies and Planning Program, University of Maryland, MD 20742, USA *Corresponding author: Zhang Jianming, e-mail: [email protected]

Abstract—Integration of remote sensing and CA has been the understanding of the dynamics of complex urban systems. frontier edge of the urban research. This paper presents an Over the past decades, a great deal of research efforts has been application of remote sensing and cellular automata in modelling directed to develop dynamic models in urban applications [1- the impacts of different policy scenarios on urban growth in 7]. This paper integrated remote sensing and cellular automata- Lanzhou, China. SLEUTH urban growth model, was introduced based dynamic model to simulate the impacts of different and coupled with remote sensing loosely. SLEUTH was an policy scenarios on urban growth in Lanzhou, China (Figure extended cellular automata model, developed with predefined 1). SLEUTH is an acronym for the input layers that the model growth rules applied spatially to grid maps of cities, and designed uses in gridded map form: Slope, Land cover, Exclusion, to be both scaleable and universally applicable. SLEUTH is an Urban extent, Transportation, and Hillshade. It’s a cellular acronym for the input layers that the model uses in gridded map automaton urban growth model and its behavior is controlled form: Slope, Land Use, Exclusion, Urban Extent, Transportation by the coefficients of diffusion, breed, spread, slope resistance, and Hillshade. In this paper, the main built-up area of Lanzhou is chose as the study area, which is a typical valley-basin city. and road gravity. The model considers four types of growth Historical data sets derive from aerial photos of 1980 (1:10000) behavior: spontaneous neighborhood growth, diffusive growth and 2001(1:4000), and Landsat TM images collected in 1986 and and creation of new spreading centers, organic growth, and 1993.And three different policy scenarios were designed to road influenced growth [8]. Three different policy scenarios explore its different impacts on urban growth, respective to were designed for calibration and prediction of SLEUTH calibration and prediction of the model. The first scenario was model to explore its different impacts on urban growth in without any consideration given to urban planning or policy; the Lanzhou, China. second scenario considered urban planning and policies, but only 50% implementation of the overall plan or policies; the third II. DATA AND METHOD scenario considered full implementation of the urban plan and polices. The calibrations of the three scenarios all reflect the A. Study Area and Data restrictive land use for development in Lanzhou, and the intentions to control the sprawl strictly. The progressive urban Lanzhou is located near the geometric center of China, developments are projected into the future 20 years under three 35°30' 6" to 38°N and 102°18' to104°18'E, and covers five different scenarios. Compared with statistic and spatial districts and three counties, namely Chengguan District, Qilihe distribution under different scenarios, it can be concluded the District, Anning District, Xigu District, Honggu District, impacts of policies on urban growth in Lanzhou are profound, Yongdeng County, Gaolan County and Yuzhong County, for a and the government plays a very important role in urban growth. total area 13 086 km². Lanzhou is a typical valley-basin city, And the second scenario is more close to the reality, the the Yellow River runs across the built-up from the west to the implementation of urban planning and policies in the reality is east, and the South Mountains and the North Mountains stand not very good and there are still many problems about urban separately along the banks of the Yellow River. The special planning and decision-making in urban growth of Lanzhou, such physiognomy restricts the connection and interaction between as arbitrary and scientific. the main built-up areas and the counties, and the amount of land available for development is small. The study area only Keywords-policy scenarios; urban growth; remote sensing; coveres the main built-up areas, including Chengguan District, cellular automata; Lanzhou Qilihe District, Anning District and Xigu District (Figure 1), with a total area of about 303.7 km², of which only about I. INTRODUCTION 218.4 km² is suitable for development. The remainder is open Restless urban growth throughout the world has called for water area or mountainous land, so its land use is very improved methods and techniques towards a better restricted.

Figure 1 Location of the study area Historical data sources include aerial photos of 1980 Spread, Slope-Resistance and Road-Gravity coefficients (1:10000) and 2001(1:4000), and Landsat TM images collected control the behavior of the cellular automaton, ranging from 0- in 1986 and 1993. The aerial photos were orthorectified, 100, and four types of growth behavior can take place: mosaicked and visual interpreted, and Landsat TM images spontaneous, diffusive, organic and road-influenced. Self- were geometrically corrected using topographic map of 1982, modification of the rules changes the control parameters when and categorized with supervised classification and visual modeled growth rates are exceeded, so that the model’s interpretation. Urban Extent and Transportation layers were behavior includes feedback [8]. derived from aerial photos and Landsat TM images. Slope and Hillshade layers were calculated from the DEM of Lanzhou, at The calibration is built upon on a statistical approach. Thirteen spatial metrics including Product, Compare, Pop, 20 m resolution. Excluded layer serves as the primary instrument to differentiate different policy scenarios, three Edges, Clusters, Cluster Size, Lee-Sallee, Slope, %Urban, X- mean, Y-mean, Rad and F-Match, are computed to quantify the layers are included: the open water area extracted from the 2001 aerial photo mosaics for the first scenario; urban planning historical fit between the modeled results and historical data. Product metric is the multiplier of the average weighted values maps of 1978-2000 and 2001-2020 only with 50% implementation for the second scenario, respective for of the other 12 metrics, F-Match metric is only valid when the land use layer included. In the previous application studies, one calibration and prediction of the model; urban planning maps of 1978-2000 and 2001-2020 with 100% implementation for (Lee-Sallee) or several metrics or all metrics are frequently chose to narrow the coefficients ranges, and several or all the third scenario, respective for calibration and prediction of the model. All the input layers were adjusted to have the same metrics are always average weighted. The optimal SLEUTH metric (OSM) determined with the self-organizing map extent and projection (Krasovsky_1940_Transverse_Mercator), resampled into 30m spatial resolution, and finally converted algorithm revealed that during the calibration process, only seven metrics were necessary to determine the goodness of fit into GIF format suitable for the input of SLEUTH model. The for the majority of urban systems, others have almost the same data processing was completed in the GIS environment with ArcInfo and ERDAS. effect as some of the seven metrics [14]. So we define a composite metric (CM) as OSM to calibrate the model, which is the product of the weighted mean of Compare, Pop, Edges, B. SLEUTH Clusters, Slope, X-mean and Y-mean. SLEUTH model, formerly called the Clarke Cellular Automaton Urban Growth Model, was developed for and III. RESULT tested on various cities in North America, including Washington, DC, and San Francisco [8-13]. SLEUTH is an A. Calibration acronym for the input layers that the model uses in gridded map form: Slope, Land Use, Exclusion, Urban Extent, From the table 1 can be recognized, the results of Transportation and Hillshade. The basic growth procedure in calibrations are dramatically different under three different SLEUTH is a cellular automaton, in which urban expansion is scenarios, the best values of the parameters for the first modeled in a spatial two-dimensional grid. Diffusion, Breed, scenario is Diffusion (33), Breed (15), Spread (20),

Figure 2 Predictions under different scenarios in Lanzhou, China increment would be 1849.05 ha., proximately 47.18 percent; TABLE I. CALIBRATION COEFFICIENTS FOR DIFFERENT SCENARIOS Under the second scenario, the total urban area for 2020 would be 12634.11 ha., the total net urban increment would be 751.32 Calibration Coefficients ha., about 6.32 percent; Under the third scenario, the total Items Slope- Road- Diffusion Breed Spread Resistance Gravity urban area for 2020 would be 13201.29 ha., the total net urban Scenario 1 33 15 20 1 90 increment would be 12555.9 ha.net urban increment, about 5.66 percent. Scenario 2 20 25 25 2 20 Scenario 3 5 88 70 50 62 The different impacts under the different scenarios can also be discerned from the spatial distribution of simulated urbanization. For the first scenario, the projected urban Slope-Resistance (1) and Road-Gravity (90); the second additions for the period of 2001-2020 are largely adjacent to scenario is Diffusion (20), Breed (25), Spread (25), Slope- the 2001 urban pixels, which can be viewed as a continuation Resistance (2) and Road-Gravity (20); and the third scenario is of urbanization, representing the expansion of existing urban Diffusion (5), Breed (88), Spread (70), Slope-Resistance (50) pixels. A large urban cluster appears in Anning District along and Road-Gravity (62). The values except Road-Gravity are with the high speed road, mainly due to high road gravity (73) similar between the first and second scenario. Slope-Resistance in the model parameters. For the second scenario, the projected (1 or 2) is very low reflects rightly the restrictive land use, urbanization additions are very limited. The great majority is influenced by its valley-basin topography. The diffusion value adjacent to the 2001 urban pixels, and no new large urban is gradually falling from the first scenario to the third scenario. cluster can be recognized, indicating the trend of reconfiguring It’s in line with the urban planning of 2001-2020 and polices the inner spatial structure between 2001 and 2020. For the third for controlling the sprawl, also as the shortage land for scenario, several large urban clusters appear at the margin of development. As to the third scenario, diffusion (5) is very low, 2001 urban pixels, where are not very suitable to be developed sufficiently featuring the intention of strictly controlling the in the reality. So the second scenario is more close to the sprawl. High values of breed (88), spread (70) and slope reality. resistance (50) also echo much arbitrariness of the implementation of urban planning in the reality at some extent. IV. CONCLUSION Although different parameters from calibration for different scenarios, the calibrations are satisfied and the model is As the given results, it can be concluded that the impacts of suitable for Lanzhou, a typical valley-basin city of China. policies on urban growth in Lanzhou are profound, and the government plays a very important role, affecting or even B. Prediction predominate the trend of urban growth. And among the three scenarios, the second scenario matches the reality best. It The progressive urban development as projected into the reflects the implementation of urban planning and policies in future 20 years under three different scenarios can be perceived the reality is not very good. Simultaneously, it may be inferred quite well from Figure 2. Under the first scenario, the total that there are still many problems about urban planning and urban area for 2020 would be 13731.84 ha., the total net urban

0-7803-9510-7/06/$20.00 © 2006 IEEE 1433 decision-making in urban growth of China, such as arbitrary [6] Wang, Y. and X. Zhang. A dynamic modeling approach to simulating and scientific, etc. socioeconomic effects on landscape changes, Ecological Modelling, 2001, 140, pp. 141-162. [7] Li,X., Yeh AGO. Modelling sustainable urban development by the ACKNOWLEDGMENT integration of constrained cellular automata and GIS. International This research was funded in part by Interdiscipline Youth Journal of Geographical Information Science, 2000,14 (2), pp.131-152 Innovation Research Foundation of Lanzhou University [8] Clarke, K. C., Hoppen, S., & Gaydos, L. A self-modifying cellular automaton model of historical urbanization in the San Francisco Bay (Project of Simulating Urban Sprawl of Lanzhou Metropolitan area. Environment and Planning B: Planning and Design, 1997,.24, Area Based on GIS and Cellular Automata), NSFC Innovation pp.247–261. Team Project (No.40421001), National Basic Research [9] Silva, E.A., Clarke, K.C. Calibration of the SLEUTH urban growth Program of China (973 Program) (No. 2005CB422001) and model for Lisbon and Porto, Portugal. Computers, Environment and Natural Science Foundation of China (No. 40501077). Urban Systems, 2002, 26, pp.525–52. [10] Candau, J., & Goldstein, N. Multiple scenario urban forecasting for the California south coast region. In Proceedings of the 40th Annual REFERENCES Conference of the Urban and Regional Information Systems [1] Batty, M., Xie, Y. Modelling inside GIS: Part 2, Selecting and Association, Chicago, Illinois., October 26–30, 2002. calibrating urban models using Arc/Info. International Journal of [11] Jantz, C.A., Goetz, S.J., Shelley, M .K. Using the SLEUTH urban Geographic Information Systems, 1994, 8, pp. 470–541. growth model to simulate the impacts of future policy scenarios on [2] White R. and Engelen G. Cellular Automata as the Basis of Integrated urban land use in the Baltimore/Washington metropolitan area. Dynamic Regional Modelling. Environment and Planning B, 1997, 24, Environment and Planning B, 2003, 31, pp. 251–71. pp.235-246. [12] Dietzel, C.K., Herold, M., Hemphill, J.J., Clarke, K.C. Spatio-temporal [3] Clarke, K., Gaydos, L. Loose-coupling a cellular automaton model and dynamics in California's Central Valley: Empirical links to urban theory. GIS: long-term urban growth prediction for San Francisco and International Journal of Geographic Information Science, 2005, 19 (2), Washington/Baltimore. International Journal of Geographical pp.175-195. Information Science, 1998, 12 (7), pp. 699–714. [13] Yang, X. and LO, C.P. Modelling urban growth and landscape changes [4] Wu, F. and C. J. Webster. Simulation of land development through the in the Atlanta metropolitan area. International Journal of Geographical integration of cellular automata and multicriteria evaluation, Information Science, 2003, 17(5), pp.463–488 Environment and Planning B, 1998, 25, pp.103-126 [14] Dietzel, C.K., Clarke K.C. Determination of optimal calibration metrics [5] Batty, M., Xie, Y., Sun, Z. Modelling urban dynamics through GIS- through the use of self-organizing maps. Institute for Environmental based cellular automata. Computers, Environment and Urban Systems, Studies, Amsterdam. Integrated Assessment of the Land System 1999, 23, pp.205–233. Workshop, 2004.