Forest Snow Patterns Derived Using ClustSnow Are Temporally Persistent Under Variable Environmental Conditions

Abstract Snow distribution affects water availability locally and downstream, especially in forests where snow amounts vary across small spatial scales (<3 m). Uncrewed Aerial Vehicle (UAV)‐based Light Detection and Ranging (LiDAR) measurements revealed that the snow distribution follows persistent patterns which, once identified, can be used to extrapolate point observations across space. Existing methods have derived such patterns from individual snow depth surveys, producing extrapolation bases that are limited to the specific timing of the underlying data. The new ClustSnow workflow addresses this limitation by applying clustering algorithms to multi‐temporal UAV‐based LiDAR snow depth maps, producing patterns that serve as an extrapolation basis for snow measurements across entire snow seasons. Here, we evaluate ClustSnow's transferability among sites and patterns' temporal persistence across three snow seasons with different snow dynamics. To support this assessment, we present a novel data set consisting of 19 UAV‐based LiDAR surveys, a network of snow depth sensors, and manual snow measurements, collected in Alptal (Swiss Prealps) and at the Schauinsland summit (German Black Forest). The derived clusters show similarities with forest structure classes at Alptal (up to 91%), but less at Schauinsland (up to 71%). At both sites, snow depth and snow water equivalent (SWE) maps obtained by ClustSnow are consistent with manual measurements and physics‐based snow model simulations, with mean root mean square errors of 8 cm for snow depth and 30 mm for SWE maps. ClustSnow therefore provides a robust framework for determining temporally persistent snow patterns and generating spatiotemporally continuous snow data sets from observations.