Abstract: In midto highlatitudes and alpine regions snow cover plays a vital role in regional climate. Area and spatial distribution of snow cover in alpine regions varies significantly over time, due to seasonal and interannual variations in climate. Therefore, there is a need for monitoring the area and spatial distribution of snow cover. Recently, remote sensing data become the most popular source for acquiring the snow cover information. There are many optical remote sensing data sources are used for extracting snow cover information, such as NOAA/AVHRR, EOS/MODIS, LandsatTM/ETM+ and so on. Compared to these data sources, HJ-1A and HJ-1B satellites both have comparatively higher temporal and spatial resolution and it is more conducive to monitor the variations of snow cover at small watershed. At present, the study on the methods of extracting snow cover information based on HJ-1A and HJ-1B data is less. In this paper we exploited the methods for extraction of snow cover information in two cases, both HJ-1B/CCD and HJ-1B/ IRS data and just HJ-1B/CCD data. The reason we chose the two cases is that, the two optical satellites HJ-1A and HJ-1B, operating in constellation now, are capable of providing a wholeterritory coverage period in visible light spectrum in two days, infrared in four days. So sometimes we can only obtain CCD image, which can not use the method of normalized snow index to extract snow cover information. Since a large area of forest distribute in the south of the study area, the snow pixels are difficult to identify, so for the first case, choose NDSI and S3 normalized snow indexes and assisted with the NDVI or forest area which retrieved from TM image to extract snow cover. For NDSI, which uses reflectance values of red and SWIR spectral bands of HJ-1B. And S3 index uses reflectance values of NIR, red and SWIR spectral bands. As it showed that, with the aid of vegetation information, the snow cover can be well extracted by two types of normalized snow index. Meanwhile, the results are quite similar to each other and of high accuracy. For the second case, normalized snow index can not be calculated, so we use SVM method with spectrum and texture information to extract snow cover. Compared to the methods of Maximum Likelihood and SVM only with spectrum, this method is the best. With this method, the snow cover area and extraction accuracy are slightly lower than the other methods mentioned above in the first case, but quite close to that. It showed that without IRS data, comparatively accurate snow cover can also be extracted only based on CCD.

Key words: HJ-1B, snow cover, NDSI, S3, SVM