Simulate Urban Growth Based on RS, GIS, and SLEUTH Model in Shenyang-Fushun Metropolitan Area Northeastern China

2009 Urban Remote Sensing Joint Event

Simulate urban growth based on RS, GIS, and SLEUTH model in Shenyang-Fushun metropolitan area northeastern China

Fengming Xi Yuanman Hu Institute of Applied Ecology, Chinese Academy of Sciences Institute of Applied Ecology, CAS Shenyang, China Shenyang, China Graduate University of CAS [email protected] Beijing, China [email protected] Xiaoqing Wu Institute of Coastal Zone Research for Sustainable Hong S He Development, CAS Institute of Applied Ecology, CAS Yantai, China Shenyang, China [email protected] School of Natural Resources at the University of Missouri Columbia MO USA [email protected] Rencang Bu, Yu Chang, Miao Liu Institute of Applied Ecology, CAS Shenyang, China

Jing Yu Urban and Rural Planning Administration Center, Ministry of Housing and Urban-Rural Development of China Beijing, China

Abstract—Shenyang and Fushun are two most nearest mega growth pattern and most diffused growth appears at south of cities in China. Integration of the two cities as one sub- region. Edge growth and road gravity growth are the main administrative economic region is a state and province policy to growth types in the future. The accelerated urban development promote economy development of Liaoning province. How the scenario shows the most urban growth area. The limitative urban urban patterns of the mega cities will grow is interested to city growth scenario shows the least urban growth area. The planners, decision-makers, land managers, ecologists, protected urban development scenario shows moderate urban geographers, and resource managers because it’s special policy growth area and good protection to other land resources. The and spatiotemporal dynamic complexity. This study explore the urban land of two mega cities will connect firstly to a whole on combined application of remote sensing, geographical the south bank of Hun River about in 2040 in the accelerated information system and SLEUTH urban growth model to urban development and the current trends scenarios, and will not analyze and model urban growth pattern in Shenyang-Fushun connect on the other two scenarios until 2050. The combined metropolitan area northeastern China. The sequential RS images method using remote sensing, geographical information system can give quantitative descriptors of the geometry of urban form and SLEUTH urban growth model is powerful for to be computed and compared over time. The investigation is representation, modeling and prediction of the spatiotemporal based on a 16-year time series data set compiled from interpreted urban growth, and useful for understanding the alternative historical TM satellite imagery. The SLEUTH model was future planning scenarios, but location accuracy and scenarios calibrated using the mutil-temporal data set for the 6391.12km2 design must be further considered for local application. area where has experienced rapid urbanization in recent years. The model allowed a spatial forecast of urban growth, and future growth was projected out to 2050 assuming four different policy 1 INTRODUCTION scenarios: (1) current trends (CT), (2) accelerated urban Urban growth has been a hot topic not only in the development (AUD), (3) protected urban development (PUD), management of sustainable development but in the fields of and (4) limitative urban growth (LUD). The predicted urban remote sensing and geographic information science [1, 2, 3] growth shows similar compact pattern under each scenario The fastest urbanization appears in developing countries except the current trends scenario that shows diffused urban nowadays [4], especially China that has the most population in

978-1-4244-3461-9/09/$25.00 ©2009 IEEE 2009 Urban Remote Sensing Joint Event the world. Chinese urban planners are facing a huge challenge explore the followings: What the future area, location, and to rapid physical and socio-economic restructuring of Chinese orientation of urban growth pattern in Shenyang-Fushun cities on the background of rapid urbanization and multi-scale metropolitan area under different planning scenarios? When administrative policies from different level of government, will the two mega cities connect spatially as a whole? Location which also have been attracting more and more attentions of accuracy and scenarios design are discussed for local not only Chinese scholars but international urban researchers as application. well [5, 6, 7, 8, 9]. City has spatiotemporally dynamic complexity [10]. The 2 MATERIALS AND METHODS spatial pattern of a city is a consequence of the interaction of various kinds of driving forces including natural and 2.1 Study area socioeconomic factors [11, 12]. Area, location, and orientation Shenyang-Fushun metropolitan area is located on the of urban growth pattern are very important to urban planners, transition zone from the branch-range of Changbai Mountain to conservationists, economists, ecologists, and land resource the flood plain of Liao River in northeast China (Fig 1), which managers in China who accounting for the urban spatial one of fastest urbanizing, and the most typical old heavy- planning, ecological protection, economic sustainable industry metropolitan areas in northeastern China. Fusun is development, and land resources sustainable use. The multi- located the upper reach of Hunhe River and Shenyang is sources, multi-temporal, and multi-scale sequential RS images located middle reach of Hunhe River. Most area of Shenyang is can give quantitative descriptors of the urban pattern to be plain. Most area of Fushun is mountain and with a large computed and compared over time, which is useful to explore reservoir in upper reach of Hunhe River. Fushun is a famous the law of urban growth [13, 14,15,16,17]. The dynamic spatial city for its coalfield and petrochemistry. Fushun is the fourth simulation models is a good way to understand the dynamics of largest city in Liaoning province. Shenyang is the capital and complex urban system, evaluate the impact of urban growth on the first largest city of Liaoning province and communication, environment, and be a planning tool for planner and decision commerce, science, and culture center of Northeast China. makers [18, 19, 20, 21, 22, 23, 24, 25, 26, 27]. The combined Shenyang and Fushun have been two of the fastest growing and method using remote sensing, geographical information system the nearest mega cities in China, driven by policies and urban growth model is powerful for representation, encouraging industrialization at multiple government levels. modeling and prediction of the spatiotemporal urban growth The total population of study area for this research is more than pattern, and useful for understanding the alternative future 6 530 000. There is a state and province policy that will planning scenarios [28, 29, 30]. combine Shenyang and Fushun as one sub-administrative Shenyang and Fushun are two most nearest mega cities in economic region, the first time in China two proximate heavy China. Integration of the two cities as one sub-administrative industrial cities have been combined to share resources, economic region is an important state and province policy to comparative advantages, and ecological protection [31]. promote economy development of Liaoning province. There are mainly four different policy scenarios about integration of 2.2 SLEUTH dynamic urban growth model the two mega cities: (1) current trends scenario (CT), the two SLEUTH is a self-modifying probabilistic cellular automata mega cities keep the current policies and urban growth trend as model originally developed by Keith Clarke at the University history; (2) accelerated urban development scenario (AUD), of California at Santa Barbara [22, 32, 33]. It’s an initial take urban development policies into account and give lower abbreviation for the input data layers: Slope, Land Use, protective probability in all land types of urban planning area Exclusion, Urban, Transportation, and Hill Shade. The version and urban development area, especially the connective zone of used in this application, SLEUTH 3.0Beta, consists of an urban two cities, and give higher protective probability in other area; growth model (UGM) and an optional Land Cover Deltatron (3) protected urban development (PUD), give more protection Model (LCDM). The UGM simulates urban growth, which is a to important land sources such as natural water, forest, and non-urban cell will be converted to an urban cell. The LCDM, arable land, and the areas that are restrictive to urban growth; which is driven by the UGM, simulate land use and land cover and (4) limitative urban growth (LUD), give more stringent change. In this application, only the urban module was utilized, protection to land sources than PUD scenario and limit the so each cell in the study area had only two possible states: urban growth. Specifically, the objectives of this paper are to urbanized or nonurbanized. SLEUTH simulates four types of urban growth (Spontaneous growth, New spreading center growth, Edge growth, and Road-influenced growth) controlled through the interactions of five growth coefficients (ranging from 0 to 100): Dispersion, Breed, Spread, Road gravity, and Slope (Table ). We use UGM only. The four urban development types in UGMdescribe briefly below [22, 29, 32]: Spontaneous growth models the random urbanization of single pixels in undeveloped areas, which has the potential to Figure 2. Study area within the Liaoning province northeastern China capture low-density development patterns and is not dependant

978-1-4244-3461-9/09/$25.00 ©2009 IEEE 2009 Urban Remote Sensing Joint Event on proximity to existing urban areas or the transportation cannot be urbanized). The permutation and combination of the infrastructure. The overall probability that a single five parameters derived from calibration [33] are tested for their ability to replicate historic growth patterns and landscape and predict future urban growth patterns and landscape TABLE I. LIST OF INPUT DATA FOR SLEUTH MODEL IN SHENYANG- FUSHUN METROPOLITAN AREA changes. SLEUTH also maintains an optional self-modification rule. Theme year sources Urban 1988 Classified from Landsat5 TM satellite images Self-modification (if used) alters the diffusion, breed, and extent (1988, 1992, 1997, 2000, and 2004). The ancillary spread coefficient values to simulate accelerated growth (boom 1992 data: the Topographical map (1:100,000) of 1981; condition) or depressed growth (bust condition). At the end of 1997 the Shenyang and Fushun central city maps derived a year, the amount of growth is compared to a set of limits from air photos (1:10,000) of 1997 and of 2004 2000 (critical_high, critical_low). If the amount of growth exceeds (1:15,000) and the ground survey information (GPS the critical_high limit, the three coefficients are multiplied by a 2004 points). Overall Accuracy (OA) = 88.5%, Kappa Index (KI) = 0.847. value greater than one, increasing growth. If the amount of Roads 1988 Transportation Maps of Liaoning Province (1:550 growth is less than the critical_low limit, the three coefficients 1992 000) in 1988, 1992, 1997, 2000, and 2004. are multiplied by a value less than one, depressing growth. The 1997 Landsat5 TM images in 1988, 1992, 1997, 2000, coefficient changes take effect during the next year of growth. 2000 and 2004. The critical_high, critical_low, and multiplier values for boom 2004 and bust are defined in a scenario file. The road_gravity factor Exclusion 1988 Classified from Landsat5 TM satellite image of is increased as the road network enlarges, promoting a wider 2004 1988. Conserves in China band of urbanization around the roads. The slope_resistance Slope 1981 Digitalized from the Topographical map factor is increased, allowing expansion onto steeper slopes. (1:100,000) of 1981 When new growth in a time cycle takes place on steeper slopes, Hillshade 1981 Digitalized from the Topographical map the spread factor is increased which accelerates urban (1:100,000) of 1981 expansion on flat land. The self-modification rule keeps the nonurbanized cell in the study area will become urbanized is urban growth a nonlinear process [32, 34, 35, 36]. determined by the dispersion coefficient. SLEUTH enables users to define future urban growth and New spreading center growth simulates the emergence of landscape changes as a projection of past, design future new urbanizing centers by generating up to two neighboring alternative urban growth and land management scenarios [22, urban cells around areas that have been urbanized through 27,32, 37], as well as estimate effects of increased urbanization spontaneous growth. The breed coefficient determines the on local environment [30, 35, 38]. overall probability that a pixel produced through spontaneous growth will also experience new spreading center growth. 2.3 Landscape metrics To quantify and examine spatiotemporal urban growth Edge growth promotes the expansion of established urban pattern, six landscape metrics were chosen on the basis of cells to their surroundings. Only undeveloped cells that have at published research [28, 39, 40, 41, 42, 43, 44, 45, 46, 47]. least three urban neighbors and pass the spread coefficient and These were Number of Patches (NP), Class Area (CA), Largest slope resistance tests become a new urban location. Spread Patches Indexes (LPI), Euclidian mean nearest neighbor coefficient control the probability that a nonurban cell will distance (MNN), Area weighted mean patch fractal dimension become urbanized. (AWMPFD) and Contagion (Contag). These values describe Road-influenced growth simulates the influence of the urban growth structure and pattern (Table). The SLEUTH transportation network on growth patterns by generating simulations GIF graphical outputs were converted to TIFF spreading centers adjacent to roads. When road-influenced using XV software, then converted TIFF to GRID. The GRID growth occurs, newly urbanized cells are randomly selected at maps were input the Fragstats solfware [48] to calculate the a probability level determined by the breed coefficient. For landscape metrics. each selected cell, the existence of a road is sought within a search radius defined by the road-gravity coefficient. If roads 2.4 Input data are found near the selected cell, a temporary urban cell is SLEUTH uses historical data sources for calibration. It placed at the closest location adjacent to a road. This temporary needs at least four different periods of urban extents, two urban cell then searches along the road for a permanent periods of transportation, the slope, exclusion area, hill shade location. The direction of the search along the road is random and two different periods land cover if land use change is and the search distance is determined by the dispersion simulated. In this application, five urban extents and two coefficient. The permanent location becomes a new spreading transportation layers were used. Other single layers included: center, so up to three cells along a road can be urbanized at this slope, hill shade, and exclusion areas, as discussed above. Five point. periods of urban extent were derived from Landsat5 TM Each of the growth rules is affected by the coefficient images from 1988 to 2004. The Landsat5 Thematic Mapper values set at the beginning of a growth cycle, as well as the image in 1997 was geometrically corrected using the slope, urban extents, land cover (if LCDM used), and excluded topographic map (1:100,000) of 1981. Then the image-to- layers (e.g. land such as water area and national parks that image method was used for the geo-referenced registration of

978-1-4244-3461-9/09/$25.00 ©2009 IEEE 2009 Urban Remote Sensing Joint Event other images with the total Root Mean Squared (RMS) error of planning, and decision making; it can be named a policy layer less than 0.5 pixels. Visual interpretation (with local to some extent. For the calibration phase, the excluded layer consisted of natural water, which was 100% excluded from urbanization, as well as TABLE II. LANDSCAPE METRICS USED IN THIS STUDY, AFTER MCGARIGAL ET AL. (2002)

Metrics Description Units Range

NP—Number NP equals the number of None NP>=1, of patches urban patches in the study no limit area. A simple measure of the fragmentation of urban patch. CA—Class CA equals the sum of the Hectares CA>0, no area areas (m2) of all urban limit patches, divided by 10,000 (to convert to hectares); that is, total urban area in the study area. LPI—Largest LPI equals the area (m2) of Percent 0 < patch index the largest urban patch LPI<=10 divided by total urban area 0 (m2), multiplied by 100 (to convert to a percentage) in this application. It quantifies the percentage of total urban area comprised by the largest urban patch. MPA—Mean MPA equals the mean area of Hectares MPA>0, Patches Area all urban patches. It is a no limit simple aggregation or fragmentation measure of urban pattern. AWMPFD— Area weighted mean value of None 1<=AW Area the fractal dimension values MPFD<= weighted of all urban patches, the 2 mean patch fractal dimension of a patch fractal equals two times the dimension logarithm of patch perimeter (m) divided by the logarithm of patch area (m2); the perimeter is adjusted to correct for the raster bias in perimeter. It reflects shape complexity of urban patches. CONTAG— CONTAG measures the Percent 0< CONT Contagion overall probability that a cell <=100 of a patch type is adjacent to cells of the same type.

Figure 2. Urban development zones, urban planning zones, and industrial knowledge) from Landsat5 TM images for the years 1988, development zones in Shenyang-Fushun China 1992, 1997, 2000, and 2004 was carried out to form a binary map of urban/non-urban classes, with the help of ancillary data national parks, scenic areas, which were 60% excluded from including the aerial photography of the central city in 1997, urbanization. For the prediction phase, all excluded areas in the 2001and 2004, the topographic map, and ground survey calibration phase were preserved and assigned a value information. The roads layers of 1988, 1992, 1997, 2000, and of100.Policy, politics, environment, planning and management 2004, including highways, railways, national roads, provincial factors were spatialized and parameterized into the excluded roads and county roads, were digitized from the Transportation layers, which come with a range of values (0-100) indicating Maps of Liaoning Province (1:550 000) of corresponding the probabilities of exclusion. In addition, we set buffer zones years, and modified by Landsat5 TM images of corresponding along rivers and around drinking water lakes, and gave 20%- years. The slope and hillshade layers were created from a 80% probabilities of exclusion. The urban development zones, digitized topographical map (1:100 000) of 1981. The excluded urban planning zones, industry development zones, dense layer consisted of natural water bodies, nature reserves, policies development zones, and potential geological disaster national parks, and scenic areas, and areas that are not allowed zones are taken into account (Fig. 2). We gave the urban for urban development (Table). The excluded layer defines development zones, industry development zones, and dense protective probability of different land use types, and is a policies development zones lower probabilities of exclusion primary tool for simulating the effects of policy support, under different planning scenarios. All input layers were

978-1-4244-3461-9/09/$25.00 ©2009 IEEE 2009 Urban Remote Sensing Joint Event rasterized at a 60m resolution using the nearest neighbour both by urban planning project and industrial development sampling method to the spatial extent of the study area except policies were given 10% or 20% probabilities of exclusion. The the roads layers, which were rasterized directly from the vector zones influenced only by urban planning project or industrial roads data. The input data were mainly prepared in a GIS development policies were given 20% to 30% probability of environment. exclusion. Easements were 80% probability of exclusion in the development policy areas, and 100% probability of exclusion 2.5 Scenario design Scenarios have been in the planner’s and manager’s toolkit as a means of representing the future for several decades. The dual functions of scenarios are bridging, which exchanges information between scenarists and scenario users and stretching, which broadens people's perspectives, contributing to a more effective decision making process [49, 50]. The design of scenarios accounts for the following aspects: 1) protective level, location and size of different land use types. For example, water, reserves, forest, and wetlands are given different exclusion probabilities at certain locations under different scenarios. (2) Whether there are urban development zones defined by urban planning project (Fig. 2a), industrial development zones and policy zones that are influenced by special national, regional and local policies (Fig. 2b). The size, location, extents, and protective level of development zones are taken into account [51]. We designed four different policy scenarios: (1) current trends (CT), (2) accelerated urban development (AUD), (3) protected urban development (PUD), and (4) limitative urban growth (LUD). The detailed setting of each excluded layer was as following (Fig 3): The CT scenario uses the same exclude layer as that in the calibration, which consisted of natural water, which was 100% excluded from urbanization. The urban growth keeps the current trend with lower protective level to farmland and forest. The AUD scenario reflects the common rapid urban growth phenomenon in the Chinese cities. There is little or no ecological protection in the urban planning zones, urban development zones (fig. 2a) and industrial development zones such as Shenbei, Shenxi, and Hunnan (fig. 2b). The zones influenced together by urban planning project and industrial development policies were given 0% or 5% probability of exclusion. The zones influenced by either urban planning projects or industrial development policies were given 10% to 20% probability of exclusion. Easements were assigned 10- 20% probability of exclusion in development policy zones, and 70% probability of exclusion outside of the zones. Forest was given 0% probability of exclusion in development policy zones, and 60-70% probability of exclusion outside of the development policy zones. Agriculture was assigned a 0 % probability of exclusion in the development policy zones, and 50-60% probability of exclusion otherwise. The streams, lakes and reserves were given 100% probability of exclusion within their buffer zones throughout the study area. Potential geological disaster zones due of mining were ignored. The planned roads were added in respective layers. The critical slope was 21% (Fig. 3a). The PUD scenario reflects a stronger commitment towards harmonizing reindustrialization, environment, and resource protection (Fig. 3b). In the excluded layer, the zones influenced

Figure 3. Excluded layers for (a) accelerated urban development (AUD), (b) urban growth. The urban growth mainly occurs on the fringe of protected urban development (PUD), and (c) limitative urban growth (LUD) urban core under each scenario. The new urban growth also scenarios Shenyang-Fushun China 80-100% probability of exclusion. appears at the Shenbei industrial development zone along the outside. Forest and agriculture were 20-35% probability of main roads and Shenfu connective zone under AUD, CT, and exclusion in development policy zones, and 50-80% LUD scenarios. Besides the edge growth around the urban probability of exclusion outside. The streams, lakes and core, the urban growth occurs mainly in west, north and reserves were given 100% probability of exclusion and areas connected zone of two cities (Shenfu urban development zone) within 120m buffer zones were given 80% probability of under the AUD scenario; the urban growth shows diffused exclusion in the whole study area. The potential geological pattern under the CUR scenario; the urban growth under the disaster zones were taken into account, and were given higher PUD scenario has the similar trend with that under the AUD level of exclusion. There are no new roads. The critical slope scenario, but the growth area is lower. Urban land of Shenyang was 21%. and Fushun will connect spatially as a whole under the AUD and CT scenarios on the south bank of Hunhe River about in The limitative urban growth (LUD) reflects a more strict set 2040, won’t connect under the LUD and PUD scenarios on the of protective policies aimed toward limited urban development south bank of Hunhe River in 2050, and won’t connect under and more natural resource protection. The data settings in the each scenario on the north bank of Hunhe River in 2050 excluded layer were similar to those in the management urban through analysis of projected future urban growth (2005-2050). development scenario (PUD), but protective levels were higher Six landscape metrics Class Area (CA), Number of Patches than that in the PUD scenario. The critical slope was 15%, preventing development entirely on steeper slopes. In addition, (NP), Largest Patch Index (LPI), Mean Patches Area (MPA), Area weighted mean patch fractal dimension (AWMPFD) and streams, wetlands, reservoirs, and lakes included larger buffer zones (Fig. 3c). Contagion (Contag) were used to compare quantitatively the urban structure and patterns (Fig. 5-10). The Shenyang-Fushun grew outward from their historical urban core under the MU 3 RESULTS and ES scenarios, while grew outward from their historical 3.1 Calibration TABLE III. CALIBRATION ROUTINES AND RESULTS FOR SLEUTH MODEL The calibration of SLEUTH is the most important phase for OF SHENYANG-FUSHUN METROPOLITAN AREA WITH LAND USE DATA. the capture of urban growth characteristics and for the success of model forecasting. The calibration approach explores the local urban growth pattern through synthesizing physical, economic, political, policy, planning, and the management factors that influence the spatiotemporal dynamics of urban growth. We chose five statistical metrics (Compare, Pop, Edges, Clusters, and Leesalee) for selecting the ‘best fit’ calibration coefficients [32, 33]. The Composite metric is product of the five statistics. The most popular calibration method for SLEUTH is “Brute Force” calibration [37], which is composed of five stages: setting constants and verification (test), coarse calibration, fine calibration, final calibration, and deriving the coefficients [33]. Other calibration methods for SLEUTH have included genetic algorithms [52] and the self- organizing map [53]. Table shows the calibration process and results. The calibration coefficients (“DNA”) of the Shenyang- Fushun metropolitan area is {Diffusion = 30, Breed = 62, Spread = 41, Slope resistance = 18, Road gravity = 10}.

3.2 Predictive urban growth structure and patterns under each scenario The prediction of SLEUTH model uses the parameters derived from calibration. The forecast of future urban growth is a probabilistic map of potential urbanized nonurban grid assuming the same growth trend as it was in the past. For the projected urban layers, 100 Monte Carlo simulations was given, and over 90% probability was used to deicide a nonurban grid cell as likely to become urbanized. Predictive urban growth structure and patterns under each scenario in 2030 and 2050 were showed in Fig. 4, which indicates great difference in area, orientation, and pattern of

Growth parameters Monte Carlo iterations = 5 and 2050 under each scenario. The most increase was under the Coarse Number of simulations = 3125 CT scenario and the urban areas were 98648.28ha in 2030 and Range Step Metrics 199097.28ha in 2050. The least increase is under the ES Diffusion 1-100 25 Compare = 0.9836 scenario. The growth rate is decrease under each scenario Breed 1-100 25 Population = 0.9323 except the CT scenario (Fig. 5). The number of patches (NP) Edges = 0.9130 Spread 1-100 25 changes greatly under each scenario. The NP of urban land is Clusters = 0.8014 199 in 2004, and decrease to 52, 74, 71, 49 under the PUD, Slope resistance 1-100 25 Leesalee = 0.7140 LUD, CT, and AUD scenarios respectively in 2030, then Road gravity 1-100 25 Composite = 0.4791 increase to 105, 5853, and 1285 under the PUD, CT, and AUD scenarios respectively in 2050. It continues to decrease to 55 Fine Monte Carlo iterations = 7 Number of simulations = 3600 under LUD scenario in 2050. The NP under the ES scenario is Range Step Metrics the highest in 2030 and the NP under the CT scenario is highest Diffusion 25-50 5 Compare = 0.9979 in 2050. So the NP of patch will coalesce in 2030 and 2050 Population = 0.9406 under the PUD and LUD scenarios, and will coalesce in 2030 Breed 25-75 10 and diffuse under CT and AUD scenarios (Fig. 6). The Largest Edges = 0.9076 Spread 25-50 5 Patch Index (LPI) has the similar increase trend as the urban Clusters = 0.8208 Slope resistance 25-75 5 area. LPI quantifies the percentage of total urban area Leesalee = 0.7211 comprised by the largest urban patch. The highest LPI is 8.36 Road gravity 1-50 10 Composite = 0.5042 in 2030 and 13.08 in 2050 under the CT scenario. The largest Final Monte Carlo iterations = 9 urban patch will occupy 13.08% of total urban area under the Number of simulations = 5832 CT scenario (Fig. 7). The highest mean patches area (MPA) is Range Step Metrics 1949.41ha under the AUD scenario, and the lowest MPA is Diffusion 25-40 3 Compare = 0.9351 1046.15ha under the LUD scenario in 2030. The highest mean Breed 55-75 2 Population = 0.9608 patches area (MPA) is the Legend as following figures Edges = 0.9097 1949.41ha under the ES scenario and the lowest MPA is Spread 35-45 2 Clusters = 0.8264 34.02ha under the CT scenario in 2050. The NP and MPA Slope resistance 25-30 3 Leesalee = 0.7439 describe the fragmentation of urban land, which show Road gravity 1-16 3 Composite = 0.5025 fragmentation of urban land decrease in 2030 and 2050 under LUD scenario, and decrease in 2030 and increase in 2050 on Driving parameter value , Monte Carlo iterations = 100 the other scenario. The urban land under LUD scenario has the Diffusion 30 most fragmentation pattern of all scenarios in 2030. The urban Breed 62 land under the CT scenario has the most fragmentation pattern, Spread 41 and then under the AUD scenario in 2050. The fragmentation Slope resistance 18 of urban land under the CT and AUD in 2050 is higher than Road gravity 10 that in 2004, but the fragmentation of urban land under the

Figure 5. The urban area under each scenario in 2004, 2030, and 2050. .

LUD and PUD in 2050 is lower than that in 2004 (Fig. 8). The Area-weighted mean patch fractal dimension (AWMPFD) Figure 4. Predictive urban growth structure and patterns under each scenario continues to decrease in 2030 and 2050 under PUD and LUD in 2030 and 2050 scenario; and decease in 2030 and increase in 2050 under CT and AUD scenarios. The highest AWMPFD is 1.1583 under urban core with many pockets of diffused urban land. The LUD scenario and the lowest is 1.1362 under the AUD urban area (CA) was 61017.84ha in2004, and increased in2030 scenario in 2030. The highest AWMPFD is 1.2504 under CT

978-1-4244-3461-9/09/$25.00 ©2009 IEEE 2009 Urban Remote Sensing Joint Event scenario and the lowest is 1.1259 under the PUD scenario in the CT scenario with the low protective level shows great area 2050. AWMPFD describes the shape complexity of urban growth and diffused pattern; the stringent protective level patches, so the shape of urban patches become complex under under the LUD scenario shows the least urban growth and the CT and AUD scenarios (Fig. 9). The Cognation is a measure of most compact pattern, but a little urban growth will block landscape complexity and fragmentation. Ares of low regional economy development; the AUD scenario with contagion represent diffused urban sprawl. The Contagion of development zones of low protective level and other land with landscape will decrease under each scenario, and decrease most higher protective level shows much urban area growth and under the CT scenario in 2030 and 2050. The highest diffused pattern in the development zones; the PUD scenario Contagion is 70.74 in 2030 and 69.94 in 2050 under ES with higher protective level than the AUD scenario in scenario. The urban pattern tends to be sprawl under CT and development zones and the other area shows moderate urban AUD scenarios in 2050 (Fig. 10). growth area and compact pattern. PUD scenario is suitable for harmonizing urban growth and land sources protection. 4 DISCUSSION AND CONCLUSION Protection of ecological land such as natural water and its The combined method using remote sensing, geographical buffer zones, conserves, wetland and so on is important for information system and SLEUTH urban growth model is urban sustainability, Where and how urban growth and powerful for representation, modeling and prediction of the landscape change influences the future environment outcomes spatiotemporal urban growth, and useful for understanding the greatly [29, 38, 53]. But these protection policies shape the alternative future planning scenarios. Landscape metrics is a urban growth area and pattern in some degree. An urban good tool to evaluate and compare the structure and pattern of development zone is a very important urban growth policy, urban growth under each scenario, which describe urban area; which affect urban area and growth pattern. Many of Chinese

7000 5853 6000 5000 4000 (NP) 3000 2000 52 74 1285 71 49 1000 199105 19955 199 199 0

the Number of urban Patches PUD LUD CT AUD Figure 6. The Number of urban Patches under each scenario in 2004, 2030, and 2050 Figure 8. The Mean Patch Area under each scenario in 2004, 2030, and 2050.

! '*! ) #$ &) ')" !#$(!%$

Figure 9. The Area-weighted mean patch fractal dimension (AWMPFD) Figure 7. The largest patch index under each scenario in 2004, 2030, and under each scenario in 2004, 2030, and 2050. 2050. cities were influenced by aforementioned two factors. Besides urban patches shape complexity, and urban fragmentation. the transportation and physical factors, the development zones Predictive area and pattern of urban growth were influenced policies and protection policies are two most important impact mainly by development zones policies and protection policies: factors of urban growth area, orientation, and pattern in China. We take the two factors into the excluded layer by spatializing

978-1-4244-3461-9/09/$25.00 ©2009 IEEE 2009 Urban Remote Sensing Joint Event the development zones policies and protection policies, and model can not do everything, but we can tell a story using a give different excluded probability from 0 to 100. The urban model properly. land of Shenyang and Fushun will connect on the south bank of Hunhe River under the AUD and CT scenarios in about 2040 ACKNOWLEDGMENT but won’t under the LUD and PUD scenarios, which is because the low excluded probability of connective zone of two cities This research was supported by the Knowledge under the AUD and CT scenarios. The urban land of Shenyang Innovation Programs of Institute of Applied Ecology, Chinese and Fushun will not connect under each scenario on the north Academy of Sciences (O6LYQY1001), the National Science bank of Hunhe River until 2050, which is because the scenic and Technology Supporting Program of China land and forest with high excluded probability exist. (2006BAJ10B05-3, 2006BAJ11B04-2) and the National Natural Sciences Foundation of China (Grant No. 40801069). 90% threshold probability of nonurban grid cell to urban grid cell were given in this application, which obviously work REFERENCES well for the edge growth around the urban core and limit the growth of new individual urbanization size [3]. It underestimates the urban growth area, diffused urban growth [1] N. B. Grimm, S. H. Faeth, N. E. Golubiewski, C. L. Redman, J. Wu, X. and other landscape metrics such as Contagion and Number of Bai, and J. M. Briggs, “Global Change and the Ecology of Cities,” Science, Vol. 319, pp. 756–760, February 2008. Patches. Even those, the urban land growth show the diffused [2] J. Wu and R. Hobbs, "Key issues and research priorities in landscape pattern under AUD and CT scenarios. The stringent protective ecology: an idiosyncratic synthesis," Landscape Ecology, Vol. 17, pp. level of land resources is important for controlling urban 355–365, December 2002. sprawl but egregious protection limit urban growth and [3] M. Herold, N. C. Goldstein, and K. C. Clarke, "The spatiotemporal form economy development such as the LUD scenario. of urban growth: measurement, analysis and modeling," Remote Sensing of Environment, Vol. 86, pp. 286–302, August 2003. [4] I. Masser, "Managing our urban future: the role of remote sensing and geographic information systems," Habitat international,Vol. 25, pp. 503- 512, December 2001. [5] J. Cheng and I. Masser, "Urban growth pattern modeling: a case study of Wuhan city, PR China," Landscape and Urban Planning Vol. 62, pp. 199–217, February 2003. [6] T. Zhang, "Land market forces and government’s role in sprawl the case of China," Cities, Vol. 17, pp. 123–135, April 2000. [7] F. Wu and A. G. Yeh, "Changing spatial distribution and determinants of land development in Chinese cities in the transition from a centrally planned economy to a socialist market economy: a case study of Guangzhou," Urban Study, Vol. 34 pp. 1851–1879, November 1997. [8] A. G. Yeh and F.Wu, "The transformation of the urban planning system in China from a centrally-planned to transitional economy," Prog. Plann. Vol.51, pp. 167–249, April 1999. [9] A. Schneider, K. C. Seto, D. R. Webster, "Urban growth in Chengdu, Figure 10. Contagion under each scenario in 2004, 2030, and 2050 Western China: application of remote sensing to assess planning and policy outcomes," Environment and Planning B, Vol. 32, pp. 323-345, Scenario design is important for application of SLEUTH May 2005. model [30, 49, 50]. Scientific, divisive, contrastive, and [10] H Tang, Science of City. 2nd ed., Harbin: Harbin Institute of Technology Press, 2004, pp. 8-35. (In Chinese) detailed scenario design affect the output. Scenario design is an [11] C. W. Pannell, “China's continuing urban transition”, Environment and important component of application of dynamic predictive Planning A Vol. 34, pp.1571- 1589, September 2002. urban growth models such as SLEUTH, which decides whether [12] M. Bürgi, A. M. Hersperger, and N. Schneeberger, "Driving forces of we can environ the alternative future urban growth pattern and landscape change-current and new directions," Landscape Ecology, Vol. its potential environmental impacts. Data attainability and 19, pp. 857–868, December 2004. spatialization are important for integrating different policies [13] J. R. Jensen and D. C. Cowen, "Remote sensing of urban/suburban [55]. We integrated industrial zones data, urban development infrastructure and socio-economic attributes," Photogrammetric policies, environmental protection policies, and different points engineering and remote sensing, Vol. 65, pp. 611–622, April 1999. of urban development into scenarios design. More detailed data [14] J. G. Masek, F. E. Lindsay, S. N. Goward, "Dynamics of urban growth and policies may improve the results. in the Washington DC metropolitan area, 1973–1996, from Landsat observations," Intl J Remote Sens, Vol. 21, pp. 3473–3486, December SLEUTH performed well in the way of the quantitative 2000. simulation of urban growth, however, the simulation accuracy [15] K. Seto and W. Liu, "Comparing ARTMAP neural network with the maximum-likelihood classifier for detecting urban change," Photogram for spatial location of new development at the pixel scale and Eng Remote Sens, Vol. 69, pp. 981–990, 2003. urban spatial pattern still needs to be improved through our [16] X. Yang and Z. Liu, "Use of satellite-derived landscape imperviousness previous study [56]. So it is not feasible for SLEUTH model index to characterize urban spatial growth," Comput Environ Urban using in the site planning and detailed planning directly. But Syst, Vol. 29, pp. 524–540, May 2005. SLEUTH may perform well in spatial planning, master [17] C. Xu, M. Liu, C. Zhang, S. An, W. Yu, J M.Chen," The spatiotemporal planning, and concept planning, and be useful for dynamics of rapid urban growth in the Nanjing metropolitan region of understanding the alternative future planning scenarios. A China," Landscape Ecology, Vol. 22, pp. 925–937, July 2007.

[18] M. Batty and Y. Xie, "From cells to cities," Environment and Planning [37] E. A. Silva and K. C. Clarke, “Calibration of the SLEUTH urban growth B, Vol. 21, pp. 531–548, December 1994. model for Lisbon and Porto, Spain,” Computers, Environment and [19] M. Batty and Y. Xie, "Modelling inside GIS: Part 2. Selecting and Urban Systems,Vol. 26, pp. 525-552, December 2002. calibrating urban models using ARC-INFO," International Journal of [38] K. Civerolo, C. Hogrefe, B. Lynn, et al, “Estimating the effects of Geographical Information Systems, Vol. 8, pp. 451–470, September increased urbanization on surface meteorology and ozone concentrations 1994. in the New York City metropolitan region,” Atmospheric Environment, [20] J. Landis, "Imaging land use futures: applying the California Urban Vol. 41, pp. 1803–1818, March 2007. Futures Model," Journal of the American Planning Association, Vol. 61, [39] E. J. Gustafson, “Quantifying landscape spatial pattern: what is the state pp. 438–457, July 1995. of the art?” Ecosystems, Vol. 1, pp. 43–156, March 1998. [21] R. White and G. Engelen, "Cellular automata as the basis of integrated [40] M. Herold, H. Couclelis, and K. C. Clarke, "The role of spatial metrics dynamic regional modelling," Environment and Planning B, Vol. 24, pp. in the analysis and modeling of urban land use change," Computers, 235–246, March 1997. Environment and Urban Systems, Vol. 29, pp. 369-399, July 2005. [22] K. C. Clarke and J. Gaydos, "Loose-coupling a cellular automaton [41] C. Dietzel, M. Herold, J. J. Hemphill, and K. C. Clarke, “Spatio- model and GIS: long-term urban growth prediction for San Francisco temporal dynamics in California’s Central Valley: Empirical links to and Washington/Baltimore," International Journal of Geographic urban theory,” International Journal of Geographical Information Information Science, Vol. 12, pp. 699–714, October-November 1998. Science, Vol. 19, pp. 175–195, February 2005. [23] F. Wu and C. J. Webster, "Simulation of land development through the [42] K. Seto and M. Fragkias, “Quantifying spatiotemporal patterns of urban integration of cellular automata and multi-criteria evaluation," land-use change in four cities of China with time series landscape Environment and Planning B, Vol. 25, pp. 103–126, January 1998. metrics,” Landscape Ecology, Vol. 20, pp. 871–888, November 2005. [24] F. Wu and C. J. Webster, "Simulating artificial cities in a GIS [43] R. C. Corry, J. I. Nassauer, “Limitations of using landscape pattern environment: urban growth under alternative regulation regimes," indices to evaluate the ecological consequences of alternative plans and International Journal of Geographical Information Science, Vol. 14, pp. designs,” Landscape and Urban Planning, Vol. 72, pp. 265–280, May 625–648, October 2000. 2005. [25] X. Li and A. G. Yeh, "Modelling sustainable urban development by the [44] Y. Xie, M. Yu, Y. Bai, and X. Xing,“Ecological analysis of an emerging integration of constrained cellular automata and GIS," International urban landscape pattern-desakota: a case study in Suzhou, China,” Journal of Geographical Information Science, Vol. 14, pp. 131–152, Landscape Ecology, Vol. 21, pp. 1297–1309, November 2006. March 2000. [45] W. Ji, J. Ma, R. W. Twibell, and K. Underhill, “Characterizing urban [26] Y. Wang and X. Zhang, "A dynamic modelling approach to simulating sprawl using multi-stage remote sensing images and landscape metrics,” socioeconomic effects on landscape changes," Ecological Modelling, Computers, Environment and Urban Systems, Vol. 30, pp. 861-879, Vol.140, 141–162, May 2001. November 2006. [27] X. Yang and C. P. Lo, “Modelling urban growth and landscape changes [46] F. Kong and N. Nakagoshi, “Spatial-temporal gradient analysis of urban in the Atlanta metropolitan area” International Journal of Geographical green spaces in Jinan, China,” Landscape and Urban Planning, Vol. 78, Information Science, Vol. 17, pp. 463–488, July 2003. pp. 147–164, November 2006. [28] D. Z. Sui and H. Zeng, "Modelling the dynamics of landscape structure [47] Y. Lin, N. Hong, P. Wu, C. Wu, P. H.Verburg, “Impacts of land use in Asia’s emerging desakota regions: a case study in Shenzhen," change scenarios on hydrology and land use patterns in the Wu-Tu Landscape and Urban Planning, Vol. 53, pp. 37–52, January 2001. watershed in Northern Taiwan,” Landscape and Urban Planning, Vol. [29] C. A. Jantz, S. J. Goetz, M. K. Shelley, “Using the SLEUTH urban 80, pp. 111–126, March 2007. growth model to simulate the impacts of future policy scenarios on [48] K. McGarigal, E. En, and C. Holmes, (2002, April),“FRAGSTATS urban land use in the Baltimore–Washington metropolitan area” (Version 3):FRAGSTATSMetrics. University of Massachusetts- Environment and Planning B: Planning and Design, vol. 31, pp. 1–15, Produced Program” [Online]. Available: March 2003. http://www.umass.edu/landeco/research/fragstats/documents/fragstats [30] F. Xi, H. S. He, Y. Hu, et al, "Simulation and prediction of urban and documents.html. rural settlements growth and land use change in Yingkou city," Chinese [49] W. Xiang and K. C. Clarke, “The use of scenarios in land-use planning,” Journal of Applied Ecology, Vol. 19, pp. 1529-1536, July 2008. (In Environment and Planning B: Planning and Design, Vol. 30, pp. 885- Chinese) 909, November 2003. [31] Shenyang Evening Paper, "Liaoning central city clusters passed the [50] J. I. Nassauer, R. C. Corry, “Using normative scenarios in landscape examination and approviation and Construction of three metropolitan ecology,” Landscape Ecology, Vol. 19, pp. 343–356, May 2004. areas," Shenyang Evening Paper Express, Shenyang China, 2008. [51] Y. Li, Z. Hu, Q. Gao, et al, “Ecological safety pattern of spatial [32] K. C.Clarke, L. Gaydos, S. Hoppen, "A self-modifying cellular extension in Shenyang China,” Chinese Journal of Ecology, Vol. 26, pp. automaton model of historical urbanization in the San Francisco Bay 875-881, June 2007.(In Chinese) area," Environment and Planning B, Vol. 24, pp. 247–261, March 1997. [52] N. C.Goldstein, “Brains vs. brawn-Comparative strategies for the [33] USGS. (2003, January) Project Gigalopolis: Urban and Land Cover calibration of a cellular automata–based urban growth model,” In Modeling. [Online]. Available: http://www.ncgia.ucsb.edu/projects/gig/. GeoDynamics, ed. P. Atkinson, G. Foody, S. Darby, and F. Wu. Boca [34] J. Candau, S. Rasmussen, K. C. Clarke, “A coupled cellular automaton Raton: CRC Press, pp 57–62, 2004. model for land use/land cover dynamics” 4th International Conference [53] C. Dietzel, K. C.Clarke, “Toward optimal calibration of the SLEUTH on Integrating GIS and Environmental Modeling(GIS/EM4): Problems, land use model,” Transactions in GIS, Vol. 11, pp. 29-45, February Prospects and Research Needs. Banff, Alberta,Canada, 2000, 2007. September2-8, (2003-07-10) [2007-09-25], [Online]. Available: [54] J. D. Wickham, R. V. O'Neill, K. H. Riitters, E. R. Smith, T. G. Wade, http://www.colorado.edu/research/cires/banff/pubpapers/94/ and K. B. Jones, “Geographic targeting of increases in nutrient export [35] C. Dietzel and K. C. Clarke, "The effect of disaggregating land use due to future urbanization.” Ecological Applications, Vol. 12, pp. 93- categories in cellular automata during model calibration and 106, December 2002. forecasting," Computers, Environment and Urban Systems, Vol. 30, pp. [55] Y. Song, C. Ding and G. Knaap, "Envisioning Beijing 2020 through 78-101, July 2006. sketches of urban scenarios," Habitat International, Vol. 30, pp. 1018– [36] W. Solecki and C. Oliveri, “Downscaling climate change scenarios in an 1034, December 2006. urban land use change model,” Journal of Environmental Management, [56] X. Wu, Y. Hu, H. S. He, R. Bu, J. Onsted, and F. Xi, "Performance Vol. 72, pp. 105-115, March 2004. Evaluation of the SLEUTH Model in the Shenyang Metropolitan Area of Northeastern China," Environ Model Assess, in press.