A Constrained Spectral Unmixing Approach to Snow- Cover Mapping in Forests using MODIS Data

Abstract� A snow-cover mapping method accounting for forests
(SnowFrac) is presented. SnowFrac uses spectral unmixing and
endmember constraints to estimate the snow-cover fraction of a
pixel. The unmixing is based on a linear spectral mixture model,
which includes endmembers for snow, coniferous trees, branches
of leafless deciduous trees and snow-free ground. Model input
consists of a land-cover fraction map and endmember spectra.
The land-cover fraction map is applied in the unmixing
procedure to identify the number and types of endmembers for
every pixel, but also to set constraints on the area fractions of the
forest endmembers. Results are presented for non-forested areas,
deciduous forests, coniferous forests and mixed
deciduous/coniferous forests. Results are also compared to the
MODIS L2 500 m snow product.