基于MAS-LCM的沙漠化空间模拟方法研究

Spatial Simulation Method of Desertification Based on MAS-LCM Model

以干旱区典型城市磴口县为研究区,利用1995—2015年每隔5年的Landsat TM影像通过遥感解译获取研究区20年的各等级沙漠化空间分布,利用GIS空间分析和重心迁移模型分析沙漠化景观时空变化趋势。并以2010年沙漠化分类数据为基期年数据,利用Logistic元胞自动机(Cellular automata-Markov,CA-Markov)模型(简称LCM)并引入多智能体系统(Multi-agent system,MAS)模型修正转移规则,预测2015年沙漠化分类情况及其空间分布格局。研究结果表明:磴口县20年间重度及极重度沙漠化面积减小,轻度沙漠化景观面积逐渐增大,其中2015年的非沙漠化景观达到37.09%,各类型沙漠化重心远离磴口县城,呈现良好态势。引入MAS模型的CA-Markov预测模型能够显著提升模型的模拟精度,所预测的2015年数据结果 Kappa系数达到0.62,高于CA-Markov模型模拟结果,能较好预测干旱区沙漠化分布情况,为沙漠化监管与治理提供了技术支持。

Dengkou County, a typical city in the arid area, was taken as study area, and the spatial distribution of desertification for every five years from 1995 to 2015 in the study area was obtained by Landsat TM images remote sensing interpretation. Spatial and temporal variation trend of desertification landscape was analyzed by using GIS spatial analysis and gravity center migration model. Based on the 2010 desertification classification data, the 2005--2010 desertification classification area transfer matrix table was used as Markov transfer matrix file. Using the Logistic CA - Markov model （LCM） and introducing the multi-agent system （MAS） model to correct the transfer rule, the desertification classification and its spatial distribution pattern were forecasted and compared to analyze the advantages and disadvantages of the two simulation methods. The results showed that the desertification area of Dengkou County had a significant reduction in severe desertification and very severe desertification over the past 20 years. Mild desertification landscape area and non-desertification area were gradually increased, of which non-desertification landscape reached 37.09% in 2015. Various types of desertification center of gravity left away from Dengkou County, showing a good momentum. The CA - Markov prediction model with MAS model can significantly improve the simulation accuracy of the model. The predicted Kappa coefficient reached 0.62, which was higher than that of CA - Markov model. It can better predict the distribution of desertification in arid areas, and provide technical support for the current and future desertification regulation and governance.