Suojia-Qumahe Nature Reserve, which locates in the source region of Three Rivers (Yangtze River, Yellow River and Lancang River), was taken as the research field to discuss the proper method for remote sensing classification of highland wetlands. The TM images, DEM, NDWI (Normalized Difference Water Indices) and the brightness, greenness and humidity after the tasseled cap transformation were used as the indicators to establish the decision tree model to distinguish the different wetlands and other land cover types. The authors compared the results with the traditional maximum likelihood supervised classification, it showed that the decision tree method based on the indices can improve the overall accuracy by 12.05% , and the overall kappa coefficient by 0.140 7. For rivers, lakes, swamps and floodplains, the producer's accuracy and user's accuracy increased by 6.06% , 6.25% ; 0.12% , 3.13% ; 6.99% , 25.00% and 6. 12% , 28. 13% respectively. The results of this study suggest that the decision tree method based on indices is an effective tool for wetlands remote sensing classification in highland area.%以索加-曲麻河区域为例,探讨了三江源区域高寒湿地遥感分类方法.利用TM影像数据和DEM及缨帽变换后的亮度、绿度、湿度,以及归一化水体指数( NDWI)等复合识别指标,构建决策树模型,对研究区不同地类进行区分.然后通过与传统的最大似然法监督分类所得到的结果进行对比,结果表明:利用基于指数的决策树分类方法对高寒湿地类型进行分类,较传统的最大似然法监督分类总体精度提高12.05％；总体kappa系数提高0.140 7；对于河流、湖泊、沼泽、滩地等湿地类型,生产者精度和用户精度分别提高了6.06％,6.25％；0.12％,3.13％；6.99％,25.00％；6.12％,28.13％,比监督分类均有明显的提高.证明基于指数的决策树分类方法是高寒区域湿地遥感分类的一种有效手段.
展开▼
