Typical grassland classification and precision evaluation based on remote sensing data in the northern slope of Tianshan Mountain

Vegetation classification based on remote sensing data can quickly get vegetation change information at large scale. This paper chose the Landsat TM data in 2008 and Landsat ETM+ Pan data in 1999 of the typical grassland in the northern slope of Tianshan Mountain. The data used in this paper were preprocessed firstly, then the two temporal remote sensing data were fused by means of remote sensing and GIS (Geographic Information System) technologies. The vegetation in the study area was classified into eight types by visual interpretation based on the texture feature of the fused remote sensing image. Besides, we established an expert knowledge decision tree system to make a further classification combining the preliminary classification results of the visual interpretation. At last the grasslands in the study area were classified into five types, including plain desert, desert plants, lowmountain desert, temperate meadow and alpine meadow. The results of accuracy evaluation indicated that the overall classification accuracy reaches to 95%, and the overall Kappa coefficient is 0.9396, which suggests that the vegetation classification methods based on the fused image generated by nineyear interval TM and ETM+ Pan images and decision tree classification in the study area have higher feasibility. The classification effect is good and consistent with the actual vegetation cover situation.