基于NDSI-NDVI特征空间的积雪面积反演研究  

Retrieval of snow cover area based on NDSI-NDVI feature space

作　　者：陈文倩[1] 丁建丽[1] 孙永猛[1] 王瑾杰[1] 张喆[1] 

机构地区：[1]新疆大学资源与环境科学学院绿洲生态教育部重点实验室,新疆乌鲁木齐830046

出　　处：《冰川冻土》2015年第4期1059-1066,共8页Journal of Glaciology and Geocryology

基　　金：国家自然科学基金项目(41161059);自治区科技支疆项目(201504051064);2014新疆研究生科研创新项目(XJGRI2014022)资助

摘　　要：积雪是新疆高海拔地区大多数河流的重要补给来源之一,不仅是春汛期间融雪性洪水灾害和冰冻灾害的直接原因,在水资源管理、灾害防治和融雪模拟预报中也扮演着重要角色.针对目前积雪信息提取方法的优势与不足,结合二维特征空间理论,构建积雪信息反演模型,并与支持向量机提取积雪信息进行精度对比分析.结果表明:相比其他积雪信息提取方法,利用归一化积雪指数(NDSI)和归一化植被指数(NDVI)建立二维特征空间,且在特征空间基础之上构建的NN模型,反演新疆北部积雪信息精度较高,相关系数达0.837,提取精度优于支持向量机(SVM)方法,对当地防洪灾害模拟预测、生态环境保护、社会经济发展等方面具有一定参考意义.Snowis one of the important water resources of most rivers in high altitudes of Xinjiang region. It is not only the direct cause of spring flood due to snowand ice melting,but also plays an important role in local water resource management,prevention disasters,forecast and simulation of snowmelt. Therefore,the extraction of snowcover information seems to have become one of the most important basic works at locally. In this paper,in allusion to the advantages and disadvantages of other extraction methods of snowinformation at present,a retrieval model of snowinformation is built,and compared the accuracy of extracting snowinformation with that from the support vector machines. The comparison shows that,relative to other snowinformation extraction methods,the NN model is better,which is constructed on the basis of the two-dimensional feature space by using of normalized difference snowindex( NDSI) and normalized difference vegetation index( NDVI). It is confirmed that NN model is able to extract snowinformation in northern Xinjiang region with correlation coefficient of up to 0. 837,better than support vector machine( SVM) method. It is useful to flood control,disaster simulation,and prediction of the local ecological and environmental protection effects and other aspects of social and economic development.

关 键 词：特征空间 MODIS NN模型 积雪面积 

分 类 号：P426.635[天文地球—大气科学及气象学] P407