Mapping land cover of the Yellow River source using multi-temporal Landsat images

Abstract

Land cover is a crucial product required to be calibrated, validated and used in various land surface models that provide the boundary conditions for the simulation of climate, carbon cycle and ecosystem change. This paper presented a method to map land cover from multitemporal landsat images using Dempster-Shafer theory of evidence. The method firstly resolved in Gaussian probability density function calculate the basic probability assignment of each single satellite image, then multitemporal landsat images were combined using Dempster's Rule of combination. Finally, a decision rule based on ancillary information is used to make classification decisions. This method had 87.91% overall accuracy for the land cover types compared with the result of the Aerial hyperspectral image classification. The results of this study showed that Dempster-Shafer theory of evidence is an effective tool to map land cover using multitemporal landsat image.