Wetlands shrink simulation using Cellular Automata: a case study in Sanjiang Plain, China

Transcription

1 Available online at Procedia Environmental Sciences 2 (2010) International Society for Environmental Information Sciences 2010 Annual Conference (ISEIS) Wetlands shrink simulation using Cellular Automata: a case study in Sanjiang Plain, China Huan Yu a, Zhengwei He a *, Xin Pan b a College of Earth Sciences, Chengdu University of Technology, No.1 Erxian Bridge East 3rd Road, Chengdu , China b Northeast Institute of Geography and Agricultural Ecology, Chinese Academy of Science, No Weishan Road, Changchun , China Abstract Wetlands are extremely valuable natural resources, the simulation of wetland landscape spatial-temporal evolution can reveal the mechanisms and laws of landscape succession, achieve the sustainable landscape use and provide wetland conservation and management decision support. Thesis takes the inland freshwater wetlands in the Sanjiang Plain for experimental region, carries out experiment of wetland landscape changing process simulation, results show that visual effects of simulation and prediction are both good, and the total accuracy of points to points are also above 84%, which verifies the feasibility and effectiveness of wetland landscape spatial-temporal evolution simulation using Cellular Automata, and provides a new technical idea and method for wetland landscape change and protection research Published by Elsevier Ltd. Open access under CC BY-NC-ND license. Keywords: Wetlands; Cellular Automata; Transformation rules; Sanjiang Plain; Land-cover; Simulation 1. Introduction Transformation of the surface of the Earth receives much attention. As a process in itself, land-cover change has huge impacts on ecosystems, biodiversity, and the global climate [1]. Wetlands are defined as areas of marsh, fen, peatland or water, whether natural or artificial, permanent or temporary, with water that is static or flowing, fresh, brackish or salt, including areas of marine water the depth of which at low tide does not exceed six metres by Convention on Wetlands of International Importance Especially as Waterfowl Habitat. Wetlands are integral parts of the global ecosystem as they can prevent or reduce the severity of floods, feed ground water, and provide unique habitats for flora and fauna [2]. More attention should be paid to the change of wetland covers. Wetlands function degrades and area decreases rapidly caused by activities in the wetland watersheds and predominantly by agricultural ones, i.e., crop and livestock production [3]. Wetland s investigation, management and protection need the support of new scientific techniques urgently. Because the basic information on wetland resource status and trends is critical for evaluating the effectiveness of wetland management activities [4], * Corresponding author. Tel.: ; fax: address: hzw_cdut@126.com Published by Elsevier doi: /j.proenv Open access under CC BY-NC-ND license.

2 226 Huan Yu et al. / Procedia Environmental Sciences 2 (2010) government and wetland experts have been paying more and more attention to the loss and gain of the wetland. And the simulation of wetland shrink can realize the establishment of change process, which provide a scientific reference for recognizing the detail and trend of wetland change, and make the management and protection of wetlands more reasonable. Cellular Automata (CA) were originally conceived by Ulam and Neumann in the 1940s to provide a formal framework for investigating the behaviour of complex, self-reproducible systems [5]. CA consist of a simulation environment represented by a gridded space (raster), in which a set of transition rules determine the attribute of each given cell taking into account the attributes of cells in its vicinities [6]. One key feature of CA is that complex global spatial patterns can be generated by a set of simple local rules. This bottom-up approach coincides with complexity theories stating that a complex system comes from the interactions of simple subsystems [7]. CA provides an effective way of simulating and predicting the spatial-temporal evolution of complex geographical phenomena [8][9]. Landscape or land use dynamics [6][10][11][12], urban expansion [13][14][15][16], spatial dynamics of animals and vegetation population [17][18][19], epidemic propagation [20], Wildfire propagation [21][22], all these studies have demonstrated the capability of CA for simulating and predicting complex geographical processes. This paper presents a case study from a typical inland wetland system in fastdecreasing region, Sanjiang Plain in China. The research it presents, examines the potential of using CA for simulating the change process of wetland shrink. Then two question: whether and how the CA can be used to simulate the wetlands cover change process availably will be answered. 2. Study site The Sanjiang Plain, locates in the northeast region of China, is one of the largest freshwater wetland in the country (Figure 1). Since the end of the 1950s, large-scale development in the Sanjiang Plain marsh has occurred [23]. By 2003, about 80% of natural wetlands had been reclaimed for agriculture and wetland loss still happens so far [24]. With a population of 7.8 million, of which 53.4% is engaged in farming, the Sanjiang Plain has become an important commodity grain and bean supplier for China [25].The regional climate is temperate humid to subhumid continental monsoon. Average temperatures range from -18 C in January to C in July, with a frost-free period of days. Annual precipitation is mm, with 80% occurring in May September. Most of the rivers at the area have riparian wetlands supporting meadow and marsh vegetation. Sedge (Carex spp.) is the dominant plants with Phragmites spp. scattered across some portions [26]. Fig. 1. Location of the study area in Sanjiang Plain, China The study area lies in the northeast of Sanjiang Plain at N, E (Figure 1). This region is chosen as experimental area because of several elements, the Sanjiang Plain is one of the largest

3 Huan Yu et al. / Procedia Environmental Sciences 2 (2010) marsh distribution region and is typical and representative in the global temperate wetland ecosystems; Due to the relatively cold weather, deep surface waters, large marsh patches, and sparse population, reclamation of marshes in this region started relatively late, and experimental area contains two national nature reserve: Honghe Reserve and Sanjiang Reserve, three major river systems: Yalv River, Nongjiang River and Bielahong River, which make the study area keep a relative original scenery in Sanjiang Plain; In addition, during the process of development and utilization in recent decades, the conflict between people and land is obvious, wetland degradation process under human disturbance is representative, which is suitable for carrying out simulation of wetland landscape spatialtemporal evolution. 3. Data and methods 3.1. Data To complete the simulation using CA, research collected three periods (1995, 2000, 2006) land use data, in which 1995 and 2000 were used to obtain decision rules of transformation, 2006 was used to verify predicted results, and each land use data contains 5 types of cover such as water, farmlands, resident area, forest and wetlands; Sorted out soil, topography, terrain, location and other thematic data to obtain the transformation probability under the influence of many geographical conditions, in which soil data contains 22 types of soil, landform data contains 14 types of landforms, river distance is a grid file that reflects the distance to rivers and road distance reflects the distance to road. The units of river distance, road distance and DEM (Digital Elevation Model) data are meters, and the slope is degree; In addition, collected planning, feasibility reports, scientific research reports, maps and documentation of Honghe National Nature Reserve and Sanjiang National Nature Reserve, meteorology, hydrology, groundwater observations and other statistical data, which provide sufficient data for reference. The detail of each dataset was listed as table 1, all the data were coregistered and formatted as GRID format under ArcGIS 9.3. Table 1. List of data description Name Content Resolution Time Source Size Soil Spatial distribution of soil types 30m 30m 1985 Digitizing 5.64MB Landform Spatial distribution of geomorphologic types 30m 30m 1985 Digitizing 5.64MB River distance Distance to rivers 30m 30m 1998 Euclidean distance calculation 22.57MB Road distance Distance to roads 30m 30m 1998 Euclidean distance calculation 22.57MB DEM Digital elevation model 30m 30m 1986 Digitizing 22.57MB Slope Spatial distribution of slope 30m 30m 1986 Calculation from DEM 22.57MB Land use Spatial distribution of land cover types 30m 30m 1995 TM image classification 5.64MB Land use Spatial distribution of land cover types 30m 30m 2000 TM image classification 5.64MB Land use Spatial distribution of land cover types 30m 30m 2006 TM image classification 5.64MB 3.2. Methods Considering the simulation data source, the characteristics of each space partition method and purpose of practical application, thesis selects the square grid as the space basis for CA; Taking into account the area of wetlands and farmlands patches are large, extended Moore neighbor model with a radius of 2 cells is chose for CA; Through the involvement of data mining technology: decision tree, neural network, logistic regression algorithm to

4 228 Huan Yu et al. / Procedia Environmental Sciences 2 (2010) automatically obtain the conversion probability of each cellular under specific natural and man-made conditions, and gain the optimal access algorithm through the effects comparison of three methods, then completes the core part of transformation rules establishment; Further through constraint, random, time and other factors analysis, completed the establishment of transformation rules; Then finishes the simulation programming under Microsoft Visual Studio 2008 software; Finally, through visual identification and points to points comparison methods to complete the evaluation of the effectiveness and accuracy of simulation results. 4. Construction of simulation model 4.1. Analysis of change driving force The statistic results of land-cover change cells between 1995 and 2006 years (Figure 2) show that total area of changing from wetland to other cover types are km2, in which from wetlands to farmlands are km2, the farmlands count about 92.22% of total altered area, then cultivation is the main factor that result in the wetland shrink, which conform to the conclusion that were made by former researchers [24][27][28]. Then the simulation will focus on the process of changing from wetlands to farmlands. On the other side, natural reserves within study area also play a key role of preventing wetland shrink through protective and restoring activities, which will be considered as a knowledge rule that constraints the wetland cover change Area (km 2 ) Farmland Resident Forest Water Land-cover types Fig. 2. Statistics of changing from wetlands to other cover types 4.2. Probability of cellular transformation Definition of transformation rules is the core part of CA, state change of every cellular from the moment T to T+1 is determined by transformation rules, so the entire simulation process is entirely controlled by this rules. And the transformation probability obtain is a key problem of transformation rules establishment, thesis through three data mining methods based on multi-periods land-cover data, calculates the transformation probability of each cellular under river distance, road distance, slop, elevation, soil and landform factors together, and assesses the results through lift chart (Figure 3).

5 Huan Yu et al. / Procedia Environmental Sciences 2 (2010) Overall number (%) Fig. 3. The lift chart of transformation probability mining A lift chart compares the accuracy of the predictions of each model, and can be configured to show accuracy for predictions in general, or for predictions of a specific value. The x-axis of the chart represents the percentage of the test dataset that is used to compare the predictions. The y-axis of the chart represents the percentage of predictions that are correct. Therefore, the ideal line is the diagonal line, which shows that at 50 percent of the data, the model correctly predicts 50% of the cases, the maximum that can be expected [29]. A conclusion can be gained from Figure 3: Microsoft decision tree method is the best, neural networks has some advantages than logistic regression method. Then decision tree method will be used to gain the transformation probability of changing from wetlands to farmlands under the influence of many geographical conditions Constraint factors Land-cover changes are not only affected by natural and man-made factors (such as near to the residential region, convenient transportation, or a favorable natural environment), but also affected by various constraints. Constraint factors can be summarized as three main types according to impact scope: local constraints, regional constraints and global constraints [30]. Local constraints are the own constraints of each cellular, which emphasize the differences between cellular and their neighbors; Regional constraints contain some spatial information of collections, which reflect the differences within a large range; Global constraints include nonspatial information, which can reflect a number of macro effects such as policies or trends. Thesis considers three types of constraint effects together to determine the constraint factors. Cell numbers of cover types around specific cell are considered as local constraint factors; Because wetland protection artificial measures play an important role of controlling wetland landscape change process, especially the establishment of wetland reserves, which slow shrink speed greatly within reserves, so districts within reserves boundary are considered as regional constraints; Indicators such as economic growth and land-cover change area are considered as global constraint factors Random factors Land-cover changes are influenced by many factors, and their mutual relationships are very complex. In addition, a number of random factors and uncertainties are not expected to predict. Research shows that, considering the impact of these uncertainties, adding some random items can make simulation results more closer to reality during the calculation of the probability of cellular transformation [31]. This random item is expressed as: Rnd = 1 + (-ln ) (1) In which, is a random number between 0 and 1, is a parameter controlling the size of random variables. Thesis directly imports it to transformation rules during the process of cellular transformation, then as an important factor of controlling cellular state.

6 230 Huan Yu et al. / Procedia Environmental Sciences 2 (2010) Time factorss Cellular automata model represents time evolution through iteration, so how to set up the relationship between cellular iteration time and real time is a practical problem need to be considered in application. In this study, simulation of land changes is based on building models of two periods land-covers (observation interval T), and then simulate and predict the future. Observation interval T is always much longer than model iteration interval ( t), only when the T are equal to t, the transformation rules can be used directly [32]. Applying cellular transformation rules obtained from T interval to each iteration of model, a certain degree of conversion needs to be done. Supposing there is a land-cover change amount Q between two periods, there may be only part of land takes place change during t, the variation can be calculated as: q = Q / K (2) In which, K is the number of cellular automata iterations running within T time. Then there needs to include certain constraint in each iteration, this constraint can be expressed as: q / Q = 1 / K (3) At last, combines this constraints and other transformation rules together, which means only when the various rules and constraints of cellular for certain types of land cover meet the standards simultaneously, cellular status will be take place changes. 5. Results and discussion 5.1. Simulation results Simulation results are gained through a programming under Microsoft Visual Studio 2008 software based on simulation model, which is shown in Figure 4. The simulation results show that visual effects of year 2000 and 2006 are mostly consistent with actual situation. However, compared with the actual situation no matter what time period, wetland patches show concentrated distribution, and scattered distribution of small patches have not been reflected. In addition, for those should be rapidly diminishing wetland area, simulation results appear in larger plaque area that is inconsistent with the actual situation, and for the upper left corner of experimental region, wetlands have not undergone changes due to the blocking role of river. Those phenomenon are caused by the limiting characteristics of CA. origin-95 origin-00 origin-06 simulation-00 simulation-06 Farmland Wetland Water Forest Resident Fig. 4. The results of simulation

7 Huan Yu et al. / Procedia Environmental Sciences 2 (2010) First, a standard Cellular Automata model is homogeneous, every cell changes in the cellular space is subject to the same rules, in another word, transformation rules are stable and local. Roles of macro factors are not excessively considered in CA, the self-organized state change of system comes from local interactions between the system elements. This affects their authenticity, because the behavior of a system element is not defined in certain environment for specific system, but is always manifested as a kind of tendency and possibility. And the element behaviors depend not only on themselves and their local small environment, but also on the environment of system. This homogeneity will leads to cellular state transformation having a certain convergence, although the transformation rules under the influence of a variety of environmental conditions can control the rate of cellular transformation. In addition, individual cellular in model is usually not mobile, whole movement of Cellular Automata is through the state of individual cellular changing to achieve, and is carried out by description of neighbor local spatial relationship, once the relationship is defined, it will be applied to the whole cellular space. This fixity directly leads to changing rate of large patch is higher than the small one, this speed difference partly comes from reasonable ecology principle that smaller patch is more easily affected by external environment, and also comes from large patches that should be changed rapidly slow the speed and small patches that should be changed slowly accelerate the speed because of the fixity characteristics, which affect the simulation results Accuracy assessment First, in order to facilitate direct observation of the difference of simulation results and real situation, centroids of each patch is calculated and the spatial distribution of simulation results and real situation patches are plotted, which is shown in Figure 5. Vertical ordinate(10km) Year 2000 Year Vertical ordinate(10km) Horizontal ordinate(km) Actual Simulat e Horizontal ordinate(km) Fig. 5. The spatial distribution of centroids for simulation and actual situation Accuracy(%) Wetlands Farmlands Others Land-cover types Fig. 6. The accuracy of land-cover types

8 232 Huan Yu et al. / Procedia Environmental Sciences 2 (2010) The distribution plots of centroids show that simulation results of year 2000 have a high degree of consistent with actual situation, and is better than the 2006 s, which is because the present study is based on the land use of 1995 and 2000 to gain transformation rules, those are more suitable for simulation of the actual land-cover changes from 1995 to 2000, so data periods have an important influence on simulation results, which need to be considered in actual application. For quantitative evaluation of simulation accuracy, overlapping the simulation result maps of year 2000, 2006 and the actual land use map together to gain the simulation accuracy of points to points, the overall simulation accuracy of year 2000 reaches 84.44% and 2006 is 85.25%, both them are higher than 84%, thus validate the feasibility and effectiveness of wetlands spatial-temporal shrink simulation using CA. But the points to points accuracy of year 2006 is higher than year 2000 s, which is contradict with former conclusion. So the specific reasons for this phenomenon needs to be cleared, then further calculates the accuracy based on correct cellular number of each land-cover type and corresponding total number, gains the accuracy values of main land-cover types in year 2000 and 2006 (Figure 6). Accuracy of land-cover type statistical results show that types of wetlands and others in 2000 are higher than 2006 s, whereas type of farmlands in 2000 is lower than 2006 s, thus the overall accuracy of simulation is mainly affected by the farmlands. The main reason for this phenomenon is the area or cellular number of farmland landscape are much larger compared with wetlands and other landscapes. Then we can get a conclusion that points to points method only pay attention to each cellular scale consistency while ignoring the consistency of overall layout, so this method can not serve as a comprehensive assessment of the simulation results alone, combing visual judgments can provide more reasonable results. At last, visual effects judgments, accuracy evaluation all show that CA can achieve satisfactory simulation results. 6. Conclusion Thesis takes the inland freshwater wetlands in the Sanjiang Plain, Northeast of China for experimental region, carries out experiment of wetland landscape evolution process simulation based on CA technique, results show that visual effects of two periods (2000, 2006) are both good, and the overall accuracy are all above 84%, thus verifies the feasibility and effectiveness of wetland landscape spatial-temporal shrink simulation using CA, and provides a new technical idea and method for wetland landscape change and protection research. Although the CA achieves satisfactory results in simulation of wetlands spatial-temporal evolution, but due to the influence of its technical characteristics, wetland patches show concentrated distribution, and scattered distribution of small patches have not been reflected. In addition, large patches that should be changed rapidly slow the speed and small patches that should be changed slowly accelerate the speed because of the fixity of CA, and rivers have the blocking affects for enclosure region. The solvent of those problems still need further research work to revise and improve the Cellular Automata model. Acknowledgements This study was supported by the National Natural Science Foundation of China (Grant No ), the National Key Technology R&D Program Application in the 11th Five Year Period (Grant No. 2008BAK49B02). We also thank Honghe and Sanjiang National Nature Reserves for providing social and economic statistical data. References [1] Jepsen MR, Leisz S, Rasmussen K, Jakobsen J, Møller-Jensen L and Christiansen L. Agent-based modelling of shifting cultivation field patterns, Vietnam. International Journal of Geographical Information Science 2006; 20: [2] Mitsch WJ and Gosselink JG. Wetlands. New York: Van Nostrand Reinhold; [3] Zalidis GC, Mantzavelas AL and Gourvelou E. Environment impacts on Greek wetlands. Wetlands 1997; 17: [4] Megan K, Peyre ML, Reams A and Mendelssohn IA. Lingking actions to outcomes in wetland management: an overview of U.S. state wetland management. Wetlands 2001; 21:

9 Huan Yu et al. / Procedia Environmental Sciences 2 (2010) [5] White R and Engelen G. Cellular automata and fractal urban form: a cellular modelling approach to the evolution of urban land-use patterns. Environment and Planning A 1993; 25: [6] Almeida CM, Gleriani JM, Castejon EF and Soares-Filho BS. Using neural networks and cellular automata for modelling intra-urban land-use dynamics. International Journal of Geographical Information Science 2008; 22: [7] Liu XP, Li X, Liu L, He JQ and Ai B. A bottom-up approach to discover transition rules of cellular automata using ant intelligence. International Journal of Geographical Information Science 2008; 22: [8] Batty M and Xie Y. From cells to cities. Environment and Planning B: Planning and Design 1994; 21: [9] Takeyama M and Couclelis H. Map dynamics: integrating cellular automata and GIS through geo-algebra. International Journal Geographical Information Sciences 1997; 11: [10] Wang Y and Zhang X. A dynamic modeling approach to simulating socioeconomic effects on landscape changes. Ecological Modelling 2001; 140: [11] Soares-Filho BS, Cerqueira GC and Pennachin CL. DINAMICA a stochastic cellular automata model designed to simulate the landscape dynamics in an Amazonian colonization frontier. Ecological Modelling 2002; 154: [12] Moreno N, Wang F and Marceau DJ. Implementation of a dynamic neighborhood in a land-use vector-based cellular automata model. Computers, Environment and Urban Systems 2009; 33: [13] Clarke KC, Hoppen S and Gaydos L. A self-modifying cellular automaton model of historical urbanization in the San Francisco Bay area. Environment and Planning B: Planning and Design 1997; 24: [14] Kheder SAL, Wang J and Shan J. Fuzzy inference guided cellular automata urban-growth modelling using multi-temporal satellite images. International Journal of Geographical Information Science 2008; 22: [15] Vliet JV, White R, Dragicevic S. Modeling urban growth using a variable grid cellular automaton. Computers. Environment and Urban Systems 2009; 33: [16] He CY, Okada N, Zhang QF, Shi PJ and Li JG. Modelling dynamic urban expansion processes incorporating a potential model with cellular automata. Landscape and Urban Planning 2008; 86: [17] Couclelis H. Cellular worlds: a framework for modeling micro-macro dynamics. Environment and Planning A 1985; 17: [18] Couclelis H. Of mice and men: what rodent populations can teach us about complex spatial dynamics. Environment and Planning A 1988; 20: [19] Huang HM, Zhang LQ, Guan YJ and Wang DH. A cellular automata model for population expansion of Spartina alterniflora at Jiuduansha Shoals, Shanghai, China. Estuarine, Coastal and Shelf Science 2008; 77: [20] Sirakoulis GC, Karafyllidis I and Thanailakis A. A cellular automaton model for the effects of population movement and vaccination on epidemic propagation. Ecological Modelling 2000; 133: [21] Clarke KC, Brass JA and Riggan PJ. A cellular automata model of wildfire propagation and extinction. Photogrammetric Engineering & Remote Sensing 1994; 60: [22] Hargrove WW, Gardner RH, Turner MG, Romme WH and Despain DG. Simulating fire patterns in heterogeneous landscapes. Ecological Modelling 2000; 135: [23] Liu XT. Marsh resource and its sustainable utility in the Songnen-Sanjiang Plain. Scientia Geographica Sinica (supplement) 1997; 16: [24] Liu HY, Zhang SK and Lv XG. Wetland landscape structure and the spatial-temporal changes in 50 years in the Sanjiang Plain. Acta Geographica Sinica 2004; 59: [25] Chen GQ and Ma XH. A study on the underground and its water balance change after development in the Sanjiang Plain. Scientia Geographica Sinica (supplement) 1997; 16: [26] Liu XT. Wetland and its rational utilization and conservation in the Sanjiang Plain. In: Chen YY, editors. Study of wetlands in china. Changchun: Jilin Science Press; 1995, p [27] Song KS, Liu DW, Wang ZM, Zhang B, Jin C, Li F, et al. Land use change in sanjiang plain and its driving forces analysis since Acta Geographica Sinica 2008; 63: [28] Liu XT, Ma XH. Sanjiang Plain environment change and zoology care. Beijing: Beijing Scientific Publishing Company; [29] Microsoft Corporation. SQL Server 2008 books online (United States). (accessed April 13, 2010) [30] Li X, Yeh AGO. Constrained cellular automata for modelling sustainable urban forms. Acta Geographica Sinica 1999; 54: [31] Li X, Yeh AGO. Study on zoning of agricultural land protection using cellular automata model. China Environmental Science 2000; 20: [32] Li X, Yeh AGO. Knowledge discovery and geographical cellular automata. Science in China Series D:Earth Sciences 2004; 34: