Sub-pixel lake mapping in Tibetan Plateau

Lakes are valuable watersystems, used for production of drinking water, for fisheries and recreation, and can be important indirect or proxy indicator of climatic change. An important aspect to consider is the coverage of lakes, both temporal and spatial, which remote sensing satellites can provide. Modis data is suitable for regional to global operational observations, costing little and having high temporal coverage. But the mixed pixel is a common problem. The Tibetan Plateau is the most sensitive about climatic change and has hundreds of medium and small lakes whose cover is sensitive to the mixed pixels in image. The objective in our study is to develop high accuracy sub-pixel mapping algorithm for lakes cover monitoring in Tibetan Plateau. Firstly, we use the linear spectral unmixing technique to estimate the lake fraction in the mixed pixels. Secondly, we developed an algorithm, based on the concept of spatial dependence, to locate spatially the water within mixed pixel corresponding to the water proportion in the pixel. Finally, we used Aster data as "ground truth" to validate our algorithm. The algorithm produces a finer sub-pixel lake map comparative to the source Modis image.