Abstract:

Land cover plays a significant role in the earth system science, which reflects the influence of human activities and environmental changes. Within the past decades, several global land cover datasets derived from satellite observations, free of charge, have become available to the scientific community in Europe and America. They offer valuable information on the current state of the land surface. However, considerable disagreements among their classification legends pose significant errors and uncertainties in the use of such datasets. This paper proposes a fuzzysettheorybased method to fuse several land cover classification products for China. First of all, a set of (say, n) target classification legends are defined prior to the fusion process. No doubt that they should be compatible to the classification legends of the original products to be fused. Moreover, each of the target legend must correspond to at least one original legend. Secondly, each original legend is assigned to an affinity score for each target legend, which indicates how much the former agrees with the latter. Scoring, regarded as a technique based on fuzzy theory, is usually performed by experts according to semantic rules. Scoring is followed by the voting and decisionmaking procedure. Among the affinity scores of a pixel, the highest one suggests that the pixel falls into the target class linked by itself. In addition, we have also exploited spatial correlation by weighting the affinity scores of the neighboring pixels. When fusion is completed, a synthetic map (SYNMAP) combining the features of all original classification products is created. Overall consistency of class between SYNMAP and each land cover is engaged to evaluate the fusion method. All the datasets, including SYNMAP, are evaluated after being further categorized into a few of simple classes, each of which include several original or target legends. Note that classification accuracy, which offers an absolute index and is commonly seen, is not presented in the paper since we are short of ground truth data. Nevertheless, the goal of the fuzzytheorybased method is to produce a fused map that accommodates all the advantages of different original land cover data sets and reconcile their discrepancy caused by the disagreement of different classification system. Therefore, the index of consistency between two land covers should suffice. In our experiment, ESA, MODIS/IGBP, MODIS/UMD, and MODIS/PFT are employed as the original land covers to be fused. IGBP legends are set as the target. Meanwhile, nine simple classes are used during evaluation. Overall consistencies indicate improved agreement of SYNMAP with all the other land cover products. It means that the proposed fusion method has successfully combined various features of different land cover products. The conclusions can be used for national and regional numerical model and ecological environment evaluation for further research and applications