摘要:With the Landsat5 TM images of the source region of the Yarlung Zangbo River as the datum source, according to the different features of spectral combination of the grassland, and in combination with the 1 : 1 000 000 vegetation type map as well as DEM and NDVI data, the authors set up the rules of grass identification and conducted researches on grass recognition based on decision tree classification. Some conclusions have been reached; ① Due to difference in habitat types, good results of identifying grass can only be achieved by using different spectral combination features; ② Compared with traditional supervised classification, the decision tree classification based on spectral combination features has higher precision of identification, the overall classification accuracy has been improved by 15.4% and the Kappa coefficient has been increased by 0. 225; ③ In the region with elevation ranging from 4 400 m to 5 000 m, the grassland area of Orinus thoroldii is the largest, followed by the mixed meadow of Kobresia humilis and Kobresia pygmaea, and then by the bush of Caragana versicolor and Potentilla fruticos, with Kobresia littledalei having the smallest area.%以雅鲁藏布江源区为研究对象,以Landsat5 TM图像为数据源,根据不同草地类型的波段组合特征,结合源区1:100万植被类型图、DEM和NDVI数据,构建草地判别规则,利用决策树分类法对雅鲁藏布江源区草地类型进行遥感识别.研究结果表明:①不同类型草地因其生境不同,利用不同波段组合特征进行草地类型识别能够达到较好的效果;②与传统的监督分类法相比,基于波段组合特征的决策树分类法具有较高的识别精度(总体精度提高了15.4%,Kappa系数提高了0.225);③在海拔4400~5000m区域内,固沙草草原面积最大,其次为矮嵩草和小嵩草混生草甸,再次为变色锦鸡儿和金露梅灌丛,藏北嵩草草甸面积最小.