Abstract
Modeling the runoff generation of high elevation Alpine catchments requires fundamental knowledge of the snow storage and the spatial distribution of snow cover. Since in-situ snow observations are often very scarce and represent only a point information, spatial snow information from satellite data is used since decades. However, the accuracy of snow cover maps through remote sensing products depends strongly on the cloudiness. In order to generate a spatial and temporal highly resolved dataset of snow cover maps, we applied the pixel identification processor (IdePix available in SNAP v7.0) to retrieve diverse cloud layers from Sentinel-2 Level-1C products. This makes it possible to use also high-clouded images for the snow detection, which increases significantly the data availability for the later performed snow model calibration. Cloudy areas, for which snow detection by the NDSI calculation is not possible, are set to no data. Sentinel-2 images that do not have cloud information require an extra correction based on the assumption that the snow cover has a pronounced elevation gradient. The entire NDSI dataset is subdivided into 200 m elevation zones and statistically analyzed. Thereby, the cloud-influenced images clearly stand out as outliers in the elevation zones >3000 m. If an elevation zone is detected as an outlier, the corresponding elevation zone is set to no data as well. After the comprehensive cloud detection, a pixel wise comparison with in-situ snow depth observation of four different sites allows us a first validation of the snow detection quality. In a second step, the generated snow maps are compared with the snow and cloud detection algorithm developed by Eurac Research. The final snow cover maps are used together with the in-situ snow depth observations to calibrate two different snowmelt approaches of the hydrological model WaSiM - the T-index and the energy balance-based approach (including gravitational snow redistribution) - over a mountainous basin in the Eastern Italian Alps.