Synthesizing Daily Snow Cover Maps Using Satellite Images and Climate Information

Daily snow cover is an essential parameter in hydrology, climate, and environmental studies. Although remote sensing images provide valuable information on snow, they are restricted by clouds, clouds’ shadows, and temporal and spatial coverage. This study synthesizes daily snow cover maps based on climate and near clear sky Sentinel-2 and Landsat images. The motivation of this study is that snow patterns are repeatable between years with similar meteorological characteristics. Accordingly, a distance metric based on climate information is computed, including temperature and precipitation (1km resolution) as well as auxiliary data such as daily MODIS snow cover. This distance quantifies the mismatch between the days when clear sky Landsat or Sentinel-2 data is available, learning days, and days when there is no clear sky satellite data or test days. The proposed methodology is applied on a subset of the Alpine belt called the Western Swiss Alps and on the Jonschwil sub-basin, both located in Switzerland. We have synthesized daily snow cover maps for each of our regions of interest for 20 years since 2000.

To evaluate synthesized snow cover maps, we use leave-one-out cross-validation, comparison with a random selection process, and a degree-day snow model. The leave-one-out assessment shows a good agreement between the actual Landsat and the synthesized one. The synthesized snow cover maps also show that the proposed method output agrees with physical concepts as the physical features have been used along with satellite data in the proposed model. Considering physical features in synthesizing Landsat images is an innovation that allows us to use the methodology to synthesize images for the pre-satellite period. Moreover, random selection assessment shows that considering a metric based on climate and auxiliary data can synthesize snow cover as repeatable patterns depending on meteorological data.