Comparing high-resolution snow mapping approaches in palsa mires: UAS lidar vs. modelling

Abstract. Snow cover has an important role in permafrost processes and dynamics, creating cooling and warming systems and impacting the aggradation and degradation of frozen soil. Despite theoretical, experimental and remote-sensing-based research, a comprehensive understanding of small-scale snow distribution at palsas remains limited. In this study, we used unoccupied aerial systems (UASs) equipped with light detection and ranging (lidar) to generate high-resolution digital terrain models (DTMs) and derive spatially continuous snow depth maps over palsa mires in northwestern Finland. For the first time, snow distribution was recorded over a palsa using UAS lidar. The resulting snow depth maps showed sufficient accuracy, with a root-mean-square error (RMSE) of 23.49 cm and an R2 value of 0.691 when compared to in situ measured snow depth validation data. To enhance the interpretation of snow distribution patterns, we applied a random forest (RF) machine learning model trained with in situ snow depth measurements and terrain parameters derived from the UAS lidar DTMs. This approach resulted in improved accuracy, with an RMSE of 18.33 cm and an R2 value of 0.77. RF performs particularly well when modelling snow distribution over thermokarst and vegetated areas, demonstrating the potential of machine learning to capture small-scale patterns based on field observations. The UAS lidar also enables a very detailed analysis of the interactions between snow and permafrost. Both approaches reveal snow accumulation zones, especially at steep palsa margins and within cracks, where insulation limits frost penetration and contributes to degradation processes such as block erosion. In contrast, a thinner snow depth on exposed palsa surfaces allows deeper frost penetration, which initially stabilizes the ice core but then leads to the formation of steep edges and further degradation.