Validation of Fractional Snow Cover from AVHRR using Landsat TM

The suite of NOAA satellites carrying the AVHRR sensor provides daily coverage of the world's snowpack. While another satellite-borne sensor, MODIS, may provide more accurate estimates of snow cover for operational forecasting, AVHRR provides a retrospective view, gaining a perspective of historical snowpack, which in turn can supplement operational forecasting. Here we validate a fractional snow cover algorithm for AVHRR in use by the Cold Regions Research and Engineering Laboratory. The approach uses a binary decision tree trained from the theoretical reflectance of snow and non-snow spectra convolved to AVHRR bandwidths. The binary decision tree, which estimates fractional snow cover, uses bands 1 and 2 calibrated with an atmosphere optical model 6S, and a derived band 3, which estimates a reflectance component separated from the emitance component by using temperature data from channel 4, and assumptions about the surface emissivity. Using 26 Landsat TM scenes we validate 79 scenes from NOAA 9, 11, 12 and 14. We investigate the absolute differences from the fine resolution data as well as the relative differences between sensors on the two satellites. Errors of commission are eliminated with a temperature and/or elevation mask. Like most moderate resolution satellite data, georegistration errors contribute to the overall error and can be accounted for when comparing images. The AVHRR algorithm demonstrates sensitivity to fractional snow cover and performs well in comparison to TM.