Towards a snow-depth distribution model in a heterogeneous subalpine forest using a Landsat TM image and an aerial photograph

For landscapes with a complex topography and a heterogeneous forest mosaic it is not feasible to map the snow depth directly from optical satellite images. In this paper, an indirect method to predict the snow-depth distribution is presented and applied to a 0.7 km(2) subalpine catchment in central Switzerland. The method consists of (a) a parsimonious linear regression model which includes the attributes of topography and vegetation indices (derived from a Landsat Thematic Mapper (TM) image) as explanatory variables, and (b) geostatistical interpolation techniques. A previous analysis of the forest mosaic revealed two main scales showing up in the LandsatTM image and an aerial photograph. This discrepancy in scale was assumed to be the major reason why the vegetation indices derived from the LandsatTM image were only weak explanators of the snow-depth variation measured at 100-200 locations within the catchment. Surprisingly, the geostatistical interpolation (universal kriging) was not able to improve the prediction of the snow-depth distribution significantly. The residuals of the regression model showed hardly any spatial dependence for single snow-measurement dates.