Data on snow cover distribution is one of the primary inputs to numerical weather prediction (NWP) and hydrological models. Currently, regional NWP models use snow cover data with a yes/no indicator for each model grid cell. Such representation of snow is crude, and does not account for possible subgrid heterogeneity of the snow cover. To realistically represent snow within models used to simulate weather, climate and hydrological processes, the fractional snow-covered area within each grid cell should be known.

Three years ago we developed and launched an automated system for mapping snow over North America. The system is based on combined visible and infrared measurements from the Imager instrument of the Geostationary Operational Environmental Satellite (GOES) and AVHRR NOAA and microwave measurements from the Special Sensor Microwave/Imager (SSM/I) onboard the DMSP satellite. Automated snow maps are produced on a daily basis at a spatial resolution of 4 km. The purpose of this work was to extend the automated technique to produce daily maps of fractional snow cover. Two approaches to determine the snow cover fraction were employed: 1) a map-cell count approach, in which the snow fraction is calculated from the number of 4-km snow-covered pixels in a map cell; and 2) a pixel-level approach, in which the snow fraction is estimated for each individual pixel falling within the map cell and subsequently averaged over a map cell.

In the presentation we describe the technique developed to derive snow fraction and discuss the results of snow fraction retrieval during three winter seasons from 1999-2000 to 2001-2002. Snow fraction maps based on a binary (or map-cell count) approach were produced on a grid of the NOAA 32-km resolution regional numerical weather prediction (Eta) model. To estimate a sub-pixel snow fraction we used GOES Imager data in the visible spectral band and applied a linear mixture technique.

The results of a sub-pixel snow fraction retrievals clearly show the effect of forest, which masks snow cover and thus significantly reduces the observed snow fraction. Maximum value of snow fraction in the areas affected by a substantial seasonal snow was found to be a good indicator of the tree density. In particular, in the presence of snow cover over Great Plains, the snow fraction often reaches 90% whereas in the densest boreal forests in Canada it varies within 30% to 40%. Analysis of matched snow fraction retrievals and matched ground based observations has shown that over plain areas without tree vegetation a strong correlation exists between snow fraction and snow depth for snow depth values of up to 15-20 cm. An attempt was made to apply this relationship to satellite data to roughly estimate snow depth over Great Plains.