MODIS, the Moderate Resolution Imaging Spectroradiometer, is scheduled for launch in June 1998 aboard the first Earth Observing System (EOS) platform. Global, daily snow-cover maps will be produced at a spatial resolution of 500 m from MODIS data using a fully automated and computationally frugal algorithm. Several modifications to the original snow-mapping algorithm have been made to improve mapping accuracy in forests.

The accuracy of snow maps derived from the original and enhanced MODIS snow-mapping algorithms was assessed through comparison with three other snow-mapping products. Two of these products, one developed at the University of California–Santa Barbara (UCSB) for the Sierra Nevada, and another developed at the Eldgenössische Technische Hochschule Zürich (ETHZ) for the Upper Rhein-Felsberg basin in Switzerland, are area-specific. The third is an operational snow-cover product produced for portions of the United States and Canada by the National Operational Hydrologic Remote Sensing Center (NOHRSC). In all three comparisons, the products from both MODIS algorithms compare well with products produced from the other algorithms. The enhanced MODIS algorithm consistently mapped a higher proportion of pixels identified than the original, suggesting it offers improved snow detection.

Unlike the MODIS binary (snow/no snow) algorithms, the UCSB algorithm provides quantitative estimates of the snow-covered area (SCA) which is the fraction of each pixel covered by snow. Comparison of the UCSB and MODIS algorithms for a May 10, 1991 Landsat Thematic Mapper (TM) image indicates that for both the original and enhanced MODIS algorithms, pixels with approximately 50% or higher SCA are mapped as snow. The MODIS algorithms correctly identified 93% or greater of the pixels with SCA estimates of 50% or more. However, the enhanced MODIS algorithm classified more pixels with <50% SCA as snow than did the original MODIS algorithm (24% compared to 6%) thus more closely matching the results of the SCA algorithm.

The accuracies of the MODIS algorithms compared to the algorithm developed at ETHZ for May 25 and July 12, 1994 Landsat TM scenes were mixed. The ETHZ algorithm identifies both snow and transitional (50% snow) pixels. For the May 25 scene, both MODIS algorithms correctly identified the majority of snow pixels as snow (70% and 85% for the original and enhanced algorithms, respectively). Lower percentages of the transitional snow areas (32% and 48%, respectively) were correctly mapped. However, accuracies for the July 12 scene were lower - 45% and 54% for the original and enhanced algorithms, respectively. Less than 20% of the transitional areas were correctly mapped.

Snow cover products produced using the MODIS algorithm with Landsat TM images were acquired over portions of New Hampshire and New York in January and February 1997. These were compared to the operationally-produced NORHSC snow maps for the same time period. The enhanced MODIS algorithm mapped a significantly higher proportion of each TM scene as snow than did the original MODIS algorithm (19% and 36% for New Hampshire and New York, respectively). Visually, the NORHSC snow-cover maps are in much closer agreement with the enhanced MODIS algorithm while the original algorithm is more affected by land-cover variations.

# ETHZ, Zürich, Swizerland