基于SVM遥感分类技术的艾比湖地区20a土地利用与覆被变化及其驱动力分析

Analysis on Land Use and Land Cover Change and Driving Force of Ebinur Lake Region during the Past 20 Years Based on Support Vector Machine Classification

【目的】土地利用与土地覆被变化是影响干旱区生态环境变化的一个重要因子,利用遥感技术可以从大尺度上主动快速地发现变化区域,具有实时性、动态性、宏观性等优势。研究针对传统的遥感图像分类方法在精度和分类效率等方面存在不足,不确定性较高等问题,采用支持向量机(SVM,Support Vector Machine)分类技术,监测了艾比湖地区1990、2001和2011年的土地利用与覆被变化情况,分析近20 a艾比湖地区各类用地在数量和结构上的变化,探讨影响其变化的驱动力因子。【方法】选择生态环境脆弱的新疆艾比湖地区为研究区,采用SVM分类技术对艾比湖地区1990、2001和2011年三期遥感影像进行分类,并与传统的最大似然分类法和决策树分类法进行对比实验,选取最优分类方法结合GIS、GPS技术,在综合野外调查的基础上对新疆艾比湖地区20 a的土地利用与覆被变化进行监测。【结果】SVM分类方法的分类精度和Kappa系数最高,基于该技术分析出1990～2011年间艾比湖地区各类土地利用类型在结构和数量上都发生了较大变化。【结论】SVM分类方法为最优分类方法;在近20 a的时间里,艾比湖地区各类用地均发生了较大的变化;影响艾比湖地区土地利用与覆被变化的驱动力因素主要包括自然因素和人文因素,但由于短时期内自然因素对艾比湖地区土地利用变化的影响因素较小,因而人文因素被认为是影响当地土地利用与土地覆被变化的主要驱动力因素。

[Objective] Land use and land cover change are important factors that affect the changes in the ecological environment of the arid region,using remote sensing technology can find the change area from the large-scale with many superiorities such as providing up-to-date information,offering large archive of dynamic data and wide regional coverage,while the traditional extraction methods of the remote sensing image information have shortcomings including the low classification accuracy and classification efficiency,high indeterminacy.This paper aims mainly to monitor and analyze the land use and land cover change between 1990,2001 and 2011 in Ebinur Lake region based on support vector machine （SVM）,and the change of various land use types in number and structure and driving force in 20 years were analyzed as well.[Method]This paper chose Ebinur Lake region as the study area.Classification of remote sensing images between 1990,2001 and 2011 was used based on support vector machine （SVM）,and compared with traditional classification techniques such as maximum likelihood and Decision Tree,and the best one was applied to combine the technique of GIS and GPS to monitor the land use and land cover change between these years.[Conclusion] SVM has a higher classification accuracy than maximum likelihood and Decision Tree when using different parameter combinations; The various land use types have undergone great changes in number and structure; The influence of land use and land cover changes in nature was rather weak and with the support for the implementation of the policy in Xinjiang,the human social driving factors will become the main driving ones to influence the land use and land cover changes.