基于MLP-ANN与Markov Chain的土地利用变化预测方法——以锡林郭勒盟为例

Future land use simulation based on MLP-ANN and Markov Chain: A case study in Xilingol League

以北方草地典型地区—内蒙古锡林郭勒盟为案例区,在1995年到2000年的土地利用变化与驱动力分析的基础上,利用土地利用转换类型和驱动力模型,采用多层感知人工神经网络模型分析了各种土地利用类型未来的转换潜力;利用马尔可夫链模型,预测了2005和2010年土地利用格局。预测结果显示:高覆盖度草地减少幅度最大,中覆盖度草地减少相对和缓,高、中覆盖度草地的减少造成了未利用地和低覆盖度草地的增加,尤其是前者增加的幅度最大;从空间分布看,高覆盖度草地的减少集中在西北部地区,主要转变为中低覆盖度草地,中覆盖度草地的减少主要集中在西南部地区,其流向主要是以沙化土地为主的未利用地;案例研究表明,多层感知人工神经网络模型与马尔可夫链模型的结合与应用能够在很大程度上预测稳定驱动力作用下的土地利用变化趋势,从而为生态干预提供指导。

Xilingol,a typical steppe region in Inner Mongolia,Northern China,was taken as a case study area to test a land use and cover change（LUCC） simulation method.Based on previous analyses identifying the principal drivers of land use change in the region,a Multi-Layer Perception Artificial Neural Network model（MLP-ANN） was used to analyze the potential change of each land use type.Land use change between the 1995 baseline and 2005 and 2010 was assessed using Markov Chain methods to compare the main flows of predicted land use change.Results show than the main changes are from high/moderate coverage grassland to vacant land and low-coverage grassland.During the process,high-coverage grassland will broadly convert to moderate-coverage grassland in the northwest,while the vacant land will dramatically increase in the southwest from moderate-coverage grassland.These results help to understand the present negative LUCC tendency in the region,and could therefore be useful to relevant institutions in formulating some feasible countermeasures in combating this undesirable LUCC trend.