The available datasets are snow depth maps with a spatial resolution of 0.5 m derived from images of the survey camera Vexcel Ultracam mounted on a piloted airplane. Image acquisition was carried out during the approximately peak of winter (time when the thickest snowpack is expected) in spring. The snow depth maps are calculated by the subtraction of a summer-DTM from the processed winter- DSM of the corresponding date. The summer-DTM used was derived from a point cloud of an airborne laser scanner from 2020.

Due to the occurrence of inaccuracies of the calculated snow depth values caused by the photogrammetric method, we applied different masks to significantly increase the reliability of the snow depth maps. We masked out settled areas, high-frequented streets and technical constructions, pixels with high vegetation (height > 0.5 m) , outliers and unrealistic snow depth values. In addition, we modified the snow depth values of snow-free pixels to 0. The information on buildings and infrastructure comes from the exactly classified ALS point cloud and the TLM dataset from Swisstopo (https://www.swisstopo.admin.ch/de/geodata/landscape/tlm3d.html#links). High vegetation is also derived from the classification and the calculated object height from the point cloud. Outliers and unrealistic snow depth values are defined as negative snow depth values and snow depths exceeding 10 m. The classification of each pixel of the corresponding orthophoto into snow-covered or snow-free is based on the application of a threshold of the NDSI or manually determined ratios of the RGB values.

An extensive accuracy assessment proves the high accuracy of the snow depth maps with a root mean square error of 0.25 m for the year 2017 and 0.15 m for the following snow depth maps.

The work is published in: