MODIS fraction snow mapping

Abstract

The current MODIS snow-mapping algorithm uses a binary classification, i.e. a pixel will be identified either as snow or non-snow. A common problem with a binary classifier is the snow area changes a lot with the selection of the threshold value of pixel to be classified as snow. In this paper, we present a new way for 500 m MODIS fraction snow mapping based on the high resolution ASTER data. The ASTER and MODIS data are obtained simultaneously and have the similar illumination and sensor viewing geometry because the instruments are on the same platform. As to a series of temporal MODIS data, there is ASTER data capable of being the reference for the particular time period. First the MODIS and ASTER data received for the same date are co-registered with high accuracy, then each MODIS sub-pixel composition can be calculated using ASTER supervised classification results. The local snow, vegetation and rock end-members can be recognized from the MODIS data which is identified as the homogenous area with comparison to the classification result of the ASTER data. A group of end-members can be used for fraction snow mapping for the particular temporal MODIS data. In addition, snow library spectra is used for depressing the terrain effect and the bands selected for ASTER classification to avoid the problem of the sensor saturation.