基于多分类Logistic回归模型的张家口市农用地格局模拟  

Simulation of Farmland Use Pattern in Zhangjiakou Based on Multinomial Logistic Regression Model

作　　者：田媛[1] 许月卿[1] 郭洪峰[1] 吴艳芳[1] 

机构地区：[1]中国农业大学资源与环境学院土地资源管理系,北京100193

出　　处：《资源科学》2012年第8期1493-1499,共7页Resources Science

基　　金：国家自然科学基金(编号:41171088);国土资源部土地资源调查评价项目(编号:DCPJ121504-01);中央高校基本科研业务费专项资金项目(编号:2011JS162);国土资源部公益行业科研专项经费项目(编号:201011006-3)

摘　　要：区域土地利用空间格局的模拟与预测是土地利用变化研究的重要内容。本文基于张家口市第二次全国农村土地调查成果,利用地形、降水、道路、人口密度以及城镇和农村居民点等数据,以草地为参照组,构建了张家口市农用地分布格局的多分类logistic回归模型,对其耕地、园地、林地和草地的空间分布格局进行了模拟,并采用ROC曲线对模拟结果进行了检验。结果表明:坡度、降雨量、高程、人口密度以及至村庄距离是影响张家口市农用地空间分布格局的重要因素,耕地、园地、林地、草地分布概率的ROC曲线检验结果分别为0.7732、0.8800、0.7432、0.6750。模型模拟的耕地、园地、林地、草地空间分布格局与2009年相应地类的空间分布格局具有很好的一致性。该模拟很好地揭示了张家口市农用地格局形成的内在影响机制,研究结果为张家口市未来土地利用变化情景分析及其土地利用管理决策提供了科学依据。多分类logistic回归模型充分利用土地利用系统完整信息,实现多种地类的模拟和预测,为土地利用系统研究提供了一种有力工具。Land use/cover change （LUCC） is the main reason as well as the direct manifestation of global change, so the research on it has become the front in global change and a hot issue at present. Spatial simulation of land use pattern is an important part of LUCC research. Multinomial logistic regression model used in this paper is expanded from binary logistic regression model and can predict the distribution probability of multi-types of land use at one time, and simulate land use pattern more accurately. Based on the achievements of the second national rural land survey, we have simulated the land use pattern of cultivated lands, garden plots, forests and grasslands in Zhangjiakou, Hebei province with data about DEM, rainfall, road, population density and urban and rural residential area. The results show that： 1） The main influencing factors of distribution probability of agricultural land in Zhangjiakou are slope, rainfall, elevation, population density and distance to the nearest village. Among which, slope, elevation, rainfall and distance to the nearest village are negatively correlated with the distribution probability of cultivated lands, while positively correlated with forests; 2） Where there is a higher population density, the distribution probability of cultivated lands and garden plots are higher. GDP and total investment have little influence on spatial distribution of agricultural land. Landforms and water conditions are major factors that influence the spatial distribution of agricultural lands as well as constrain agricultural development and ecological stability in Zhangjiakou; 3） ROC inspection results of cultivated lands, garden plots, forests and grasslands are 0.7732, 0.8800, 0.7432 and 0.6750 respectively, and the spatial distribution probability of each land use type fits well with the real corresponding land use pattern. The model has revealed the inner mechanism of the land use pattern, which can be used for land management decisions and future land use scenarios simulation. Besi

关 键 词：土地利用格局 多分类logistic回归 驱动因素 张家口市 

分 类 号：F301[经济管理—产业经济] F224