Land Cover Mapping And Data Restructuring With The Integration Of Multisource Remote Sensing Information

In view of the problems of the existing global land cover products and the deficiency ofcurrent data fusion methods, this study aims to develop a general framework for building a hybridland cover map by the synergistic combination of a number of land-cover classifications withdifferent legends and spatial resolutions based on Dempster-Shafer and Bayesian MaximumEntropy theory. This paper first evaluated the category accuracy and confusion characteristic offour kinds of land cover products on national and regional scales, then constructed the spatialagreement and spatial homogeneity map and their quantitative relationships with categoryaccuracy. The computational methods of the land cover products and MODIS LAI productscorrespond to the target land cover legend was derived through the LCCS and the minimumdistance vector space; and a Multi-source integrated Land Cover map was generated based on theDempster–Shafer evidence theory and the map’s precision was evaluated from different aspects.Then we compared and analyzed the data restructure result on both macro and micro scalebetween Indicator Kriging and Bayesian Maximum Entropy. We also evaluated the influence ofthe number of data locations in the neighborhood on the predicted accuracy and consume time,and assessed the scale of information yielded by the adjunction of an additional data location inthe prediction process. At last, we compared the results of data restructure between the way ofusing the hard data and soft data together and using the hard data alone,when the accuracy of thesoft data is lower.