Retrieval and Validation of MODIS Subpixel Snow Fraction in Northeast China
东北地区MODIS亚像元积雪覆盖率反演及验证

Using CBERS\|2 CCD data as "ground truth" images and based on the characters of geographic environment and climate in Northeast China, this paper modified the parameters of Salomoson's subpixel snow cover fraction model, retrieved the subpixel snow cover fraction from MODIS data and analyzed the stability and accuracy of the revised model with different ways. The result shows that the modified Salomoson model has stability in different geomorphology\|landscape unit, and the little fluctuation comes from the difference of snow physical character, atmospheric effects, and the errors of snow classification images and image registration. In the Northeast Plain, when NDSI ranges from 0.52 to 0.65, the accuracy of modified model is high. However, because NDSI is based on the non\|linear transformation of band reflectance, snow cover fraction is little underestimated. The pixels, snow cover fraction of which is overestimated, are located in suburb and rural residential area. In contrast, the pixels, covering cities, towns and roads have a higher accuracy. The main reason is that human activity frequency influences the degree of spectral character difference between snow part and non\|snow part in one MODIS pixel. In comparison with MODIS snow products (MOD10A2), snow cover fraction provides much richer information than the traditional snow cover mapping methods, but both of them don't give the exact estimation of snow cover fraction in forest zone.