Validation of VIIRS snow cover in Central European Highlands

Abstract. The Moderate Resolution Imaging Spectroradiometer (MODIS) is one of the most attractive remote sensing datasets used for mapping snow cover. The MODIS product is expected to be replaced by the Visible Infrared Imaging Radiometer Suite (VIIRS) snow cover product in the near future. Therefore, a reliable and accurate evaluation of this product is needed for future hydrological applications. This study aims to assess the mapping accuracy of the VIIRS snow product at the regional scale (i.e., at 631 climate stations in Austria) and within a small experimental catchment, the Jalovecký Creek catchment in northern Slovakia, using extensive snow course measurements conducted at both open and forested sites between January 2012 and December 2020. In the VIIRS snow cover product, the Normalized Difference Snow Index (NDSI) is used to detect snow. A threshold of NDSI (T_NDSI) is needed for distinguishing snow from snow-free land. Based on the daily snow depth observations from climate stations/snow course locations, the optimal NDSI threshold (OT_NDSI) is first determined through a detailed sensitivity test (100 different T_NDSI from 1 to 100 with a step of 1). The overall accuracy (OA) of VIIRS data is then evaluated based on the OT_NDSI. The assessment of the OT_NDSI/OA is performed for all climate stations/snow profiles, as well as for different groups of stations/snow profiles representing different physiographic and land cover conditions.

The findings demonstrate that the classification accuracy for 631 Austrian stations using the optimal thresholds ranges from 52.3 % to 99.3 %, with a median of 92.5 %. The NDSI thresholds vary seasonally and decrease with increasing elevation. The NDSI thresholds fitted to different months and elevations show the smallest differences to OT_NDSI in overall accuracy. The NDSI thresholds fitted to different elevations improve regional snow cover mapping by 1.5 % between 900 – 1200 m a.s.l. and by 3.1 % above 1200 m a.s.l. At the catchment scale, the difference is found between open and forested sites, where the mapping accuracy is lower in the forest. VIIRS enables snow mapping accuracy from 71.5 to 95.9 % at the open site (during winter, median OA is up to 98.8 %) and from 58.1 to 93.7 % at the forest site (during winter, median OA is up to 91.5 %). The overall accuracy at the site with the most measurements is 95.9 % (Červenec – open site) and 93.7 % (Červenec – forest site), respectively. The accuracy at the forest is more sensitive to seasonal variation compared to the open area, where the accuracy is more stable and accurate across the year.