[Coverage extraction and up-scaling of sparse desert vegetation in arid area]

Five kinds of remote sensing inversion models, i.e., linear spectral un-mixing model, sub-pixel un-mixing model, maximal gradient difference model, and two modified maximal gradient difference models, were used to derive f(c) from remote sensing data, and the results were compared with those measured in field, aimed to select appropriate model for deriving the data of the coverage of sparse desert vegetation in arid area. The virtual multi-scale coverage images were generated by using the simple mean scale extending method to verify the inversion information from MODIS data. It was shown that linear un-mixing spectral model had a higher precision than the other models, being applicable for deriving the data of the coverage of sparse desert vegetation, but the selection of end member was rather difficult and affected the application of the model. Sub-pixel un-mixing model was universal, high precision could be obtained based on finely detailed vegetation map, but needed to measure lots of parameters. Maximal gradient difference model was simple and easy to perform, by which, the values of the coverage of crops and bare land predicted with the original model were close to the field-measured results, but the values of the coverage of sparse vegetation were underestimated. The results predicted by the modified three-band maximal gradient difference models were close to the field-measured values, and the inversed results of vegetation coverage under different scales were ideal, indicating that these models were reliable to effectively extract the information of the coverage of sparse vegetation in arid area.