Snowline Altitude Estimation in Sentinel-1 Satellite Images Using an Extended Chan–Vese Method: Proof of Concept and Prototype of Fusion With Optical Data

Snow cover varies at the scale of a mountain range depending on topography (altitude, orientation and slope) and weather conditions. We provide a simple SAR-based remote sensing tool to calculate snowline altitude estimates (SAEs), i.e., the average elevation above which snow is continuously present. This tool allows to track areas of wet snow and their changes over the seasons, and to monitor the onset and localization of snow melt. Unlike existing approaches, we detect wet snow areas by performing a segmentation on synthesized images in the form of orientation-altitude diagrams. The segmentation is done by means of an enhanced Chan-Vese curve evolution method. Our SAEs are calculated directly from satellite observations, without resorting to existing snow products, and do not fundamentally rely on the choice of thresholds. The algorithm has been tested on Sentinel-1 images during the melting period of 2018 over a large part of the Northern Alps in France, for which various validation data sets are available. A comparison of the SAEs to the model-based Bulletin d’Estimation du Risque d’Avalanche (BERA) and the optical Fractional Snow Cover Copernicus product shows good agreement, with correlations close to 0.86 for the Belledonne massif and above 0.95 for the Vanoise and the Grandes Rousses massifs when comparing SAR-based estimates and BERA. Root Mean Square Errors were found higher for north-facing slopes than for south-facing slopes. Furthermore, we demonstrate that our algorithm can be directly applied to other types of satellite data, as long as the observations are highly sensitive to snow, by providing an example based on the Sentinel-2 Normalized Difference Snow Index. This shows that in the optical case, SAEs can be computed with minimal data, for instance under heavily clouded conditions (depending on the cloud positioning). Finally, we present a proof of concept for a hybrid SAE using a combination of optical and radar data, which represents a crucial progress in the harmonious integration of various types of measurements for snow monitoring.