With the availability of Moderate Resolution Imaging Spectroradiometer (MODIS) data following the June 1998 launch of the Earth Observing System (EOS) satellite, daily, global snow-cover mapping will be performed automatically, at a spatial resolution of 500 m. The accuracy of the derived snow-cover maps will be assessed through comparisons with other snow maps and from ground and aircraft measurements. Previous research has shown that the accuracy of snow mapping is dependent upon land-cover type, sensor parameters and the algorithm employed. The greatest uncertainty in snow-mapping accuracy is found within the Earth's forested regions. In forests, the accuracy of snow mapping varies widely, due, at least in part to the type and density of the canopy. Using the EROS Data Center (EDC) land-cover maps of North America and Eurasia, we classified the Earth's land surface into 7 different categories as follows: mixed agriculture and forest, forest, barren/sparsely vegetated, tundra, grassland /shrublands, wetlands and snow/ice, and water. The average monthly snowline position, as determined from the NOAA National Environmental Satellite, Data and Information Service (NESDIS) has been determined. The land-cover types were determined for the areas north of the snowline. Preliminary snow-mapping errors for North America, determined from previous field, aircraft and satellite results, were calculated for each of the 7 categories. Results show that the greatest errors in snow mapping in North America can be expected from November through April (9%) because it is at those times that the snowline is located at its most southerly positions and when snow covers all of the boreal forests and most of the forests located in the mid-continent regions.

Snow-mapping errors in dense forests were derived from Landsat Thematic Mapper satellite and MODIS Airborne Simulator aircraft data and simultaneous field measurements in the boreal forest in southern Saskatchewan and central Alaska when the ground was continuously snow covered. The measured errors are variable, ranging from <1-43% using the at-launch MODIS snow-mapping algorithm. In most forests studied, snow-mapping errors were <10%. A minimum error of 5% is assumed for each of the 7 land-cover types, and 5% is added to the measured errors in the forests, because snow mapping from a satellite will result in errors relating to mixed-pixel effects. (Not discussed or accounted for, is the fact that errors may be greater during the spring snowmelt period because error estimates are more difficult to make when the snow cover is discontinuous.) The assumed errors for the 7 land-cover types are: 15% for the forested areas, 10% for mixed agricultural and forest lands, 5% for barren/sparsely-vegetated lands, 5% for tundra areas, 5% for grassland/shrubland areas, 5% for wetlands and 5% for snow/ice areas. These errors were extrapolated to the North American continent (north of the snowline for each month) to estimate the monthly and annual error in mapping snow using the MODIS at-launch algorithm. The average annual snow-mapping error for North America is approximately 7%. These numbers will be refined as we acquire several years of MODIS global snow-cover data beginning in 1998.

#Code 900, NASA/Goddard Space Flight Center, Greenbelt, MD 20771

@Universities Space Research Corporation, Seabrook, MD 20706

+General Sciences Corporation, Laurel, MD 20707