Automated satellite cloud analysis: a multispectral approach to the problem of snow/cloud discrimination

Abstract

An algorithm is developed and evaluated for discriminating among clouds, snow cover and clear land. The multispectral technique uses daytime images of AVHRR channels 1 (0.63^m). 3 (3.7jim) and 4 (11.0[im). Reflectance is derived for channel 3 by using the channel 4 emission temperature to estimate and remove the channel 3 thermal emission. Separation of clouds from snow and land is based primarily on this derived channel 3 reflectance. Using this technique, observed reflectance in channel 3 is 2 to 4 percent for snow, 3 to 10 percent for land, 2 to 27 percent for ice clouds and 8 to 36 percent for liquid clouds. These values overlap for thin cirrus and snow, so the routine then attempts analysis of cirrus based on its different transmissive properties between channels 3 and 4. Six images were analyzed and the total cloud cover was verified against a total of 1 10 conventional surface observations using the standard categories of clear, scattered, broken and overcast. The routine was quite successful, with the analyzed sky cover being within category for 55 percent of the stations, one category different for 33 percent, 2 categories different for 9 percent and 3 categories different for 3 percent of the stations. A major remaining problem is discrimination between ice clouds and snow cover due to the great similarity of reflective properties of these two surfaces.