An Evaluation of the Spectral Mixture Analysis Applied to MODIS Data

Satellite images of snow-covered area (SCA) have become increasingly useful tools in hydrologic and climatological studies. SCA images are used to help estimate snow water equivalence, a critical input variable for modeling in these fields. Two techniques have been used to map SCA from satellite data: 1) binary, which applies a snow/no-snow threshold to a normalized difference snow index and 2) fractional, which uses spectral mixture modeling. The objective of this work is to assess the differences between the MODIS SCA product and fractional SCA images computed using MODIS surface reflectance data, and identify the importance of vegetation correction. We compared SCA images at peak and mid-melt snowpack in two meso-scale watersheds with mixed land cover in the Pacific Northwest. The spectral mixture model estimated a much lower percent SCA at peak snowpack in comparison to the MODIS SCA images. During mid-melt, snow cover existed only in the forested upper portions of the basin. In this case, the mixture model and the standard MODIS image predicted similar amounts of SCA. At peak snowpack, the spectral mixture model calculated low SCA in forested areas above the snowline in both study basins. These results suggest that the spectral mixture modeling method, as applied here, is inadequate for identifying snow cover beneath a forest canopy where adjoining open areas are also snow-covered. For areas with a more homogeneous snow reflectance conditions, the current MODIS algorithm may estimate SCA as accurately as the spectral mixture method.