Gap-filled snow cover fraction from Sentinel-1 and Sentinel-2 constellations 

Abstract

<div> <p><span>Copernicus Land Monitoring Service has recently launched a group of high-resolution snow cover products which are derived from Sentinel-1 and Sentinel-2 constellations. High- Resolution Snow and Ice Monitoring (HRSI) products include Fractional Snow Cover (FSC) from the Sentinel-2 constellation and Wet and Dry Snow (WDS) covering Europe and SAR Wet Snow (SWS) products for selected mountain regions derived from the Sentinel-1 constellation. The FSC and WDS products have gaps in the snow cover data due to cloud presence and the SWS product provides only information on the melting snow extent, but dry snow areas and snow-free areas cannot be discriminated by means of SAR data. In the same portfolio, we provide the daily cumulative Gap-filled Fractional Snow Cover (GFSC) product, which is a fusion of those three products. In this product, we gap-fill the FSC product using the wet snow presence detected by the SWS in the spatial domain. In the temporal domain, all recent data in the last 7 days are used for gap-filling by temporal composition. The product aims to have a complete snow cover map of Europe.</span><span> </span></p> </div><div> <p><span>The quality of the product is assessed using in-situ data and </span><em><span>gap simulation, </span></em><span>for the period of 09.2017 - 08.2018 for mountain ranges in the Pyrenees, Alps, Scandinavia , East Turkey and Corsica, covered by 34 Sentinel-2 tiles. In-situ snow depth information is converted to binary snow cover information using a snow depth threshold. For the gap simulation method, as first step, FSC products with observed snow information are selected. Then, an artificial cloud mask is overlaid on these products, and the gap-filling method is run to generate GFSC products. The resulting GFSC products are compared with the corresponding observed FSC products, considering them as reference data. This comparison shows the agreement between the FSC product and the gap-filling methods. For both comparison methods, FSC values in GFSC and FSC products are converted to binary snow cover information using an FSC threshold. Resulting binary snow cover information is used in contingency tables and performance metrics are calculated for the product and for different gap-filling methods.</span><span> </span></p> </div><div> <p><span>We have found that the gap-filling provides 5 times more pixels with snow cover information and the quality is fairly good. The comparison with in-situ data shows an accuracy over 88% in temporal gap-filling and precision over 87% in spatial gap-filling. The comparison of the gap simulated GFSC and the FSC products shows an accuracy over 97% in temporal gap-filling and precision over 83% in spatial gap-filling. Temporal gap-filling performance is consistent throughout the seasons, although it is less accurate in the accumulation season for the spatial gap-filling, which is expected as the wet snow algorithm is developed for the melting season conditions. The assessment shows that the methods are working well and 7 days old FSC, WDS and SWS data are still valid to fill the gaps in the data.</span><span> </span></p> </div>