Increasing permafrost degradation in High Arctic Yedoma landscapes of the New Siberian Islands detected using deep learning segmentation of high-resolution satellite imagery

<jats:p>Climate warming has led to substantial Arctic-Boreal permafrost thaw. For the High Arctic islands, observations of permafrost thaw are scarce so far and mostly from Northern Canada. In this study, we used deep learning segmentation of high-resolution Hexagon and SPOT satellite imagery to analyze spatiotemporal permafrost landscape changes between 1979 and 2022 on the New Siberian Islands. The islands represent an ice-rich Yedoma permafrost landscape incised by a dense thermo-erosional gully network, acting as a drainage system, while the high ice content makes it particularly vulnerable to thaw-induced terrain changes. Our multi-class segmentation approach enabled us to delineate different thaw landforms and study their change trajectories and inter-dependencies between landscape components. Through statistical co-occurrence analyses with geomorphological and environmental conditions, we further analyzed potential driving factors.</jats:p> <jats:p>Our results show widespread formation of new thermokarst mounds (baydzherakhs) associated with ice-wedge thaw, with the extent of affected terrain more than doubling at one study site over a 45-year period. At the other two sites, an expansion is indicated but less clear due to uncertainties related to image quality and lower segmentation performance. While baydzherakhs are associated with steeper, north-to-east-facing slopes and extended spring snow cover duration, we observed a tendency of new baydzherakhs forming along less steep and west-to-south-facing slopes leveling previous topographical asymmetries. Thaw pond reconfiguration indicated progressing upland drying, though gully floors remained relatively stable. These findings highlight the sensitivity of ice-rich permafrost to temperature, snow regime, and regional morphology, implying high permafrost vulnerability in the Siberian High Arctic.</jats:p>