Unraveling fire-permafrost interactions in Northeastern Siberian tundra using InSAR and machine learning

Permafrost soils in boreal forests and tundra store more than two atmospheres worth of carbon, yet the vigorous permafrost-carbon-climate feedback loop remains poorly understood. In addition to ongoing strong warming, fires can further accelerate permafrost degradation and trigger the release of ancient carbon into the atmosphere. Despite the urgency after the recent Arctic fire seasons of 2019, 2020 and 2021, fire-permafrost interactions are currently not included in Earth system models from the sixth assessment of the Intergovernmental Panel on Climate Change (IPCC). This is because large-scale observations of fire-induced permafrost degradation are lacking. Therefore, we studied fire-induced permafrost degradation using the Interferometric Synthetic Aperture Radar (InSAR) technique and time series of Sentinel-1 (S-1) imagery. In this pilot study, we tested our approach on fires from 2019 and 2020 in the Chokurdakh area, Northeastern Siberia. We processed time series S-1 SAR data from a snow-free season (June-October) where S-1 SAR image selection was automated by using the Moderate Resolution Imaging Spectroradiometer snow cover products. To understand the drivers of InSAR-derived subsidence, we applied the XGBoost regression algorithm using subsidence as a response variable and ten other environmental variables as predictor variables. First, we found that the time series InSAR technique is suitable for deriving subsidence over fire-affected permafrost terrain. Second, the fire-affected permafrost terrain exhibited four to five times greater subsidence compared to the surrounding unburned area. Third, the XGBoost regression model revealed land surface temperature (LST) and albedo (derived from Landsat data) as the primary predictor variables  of surface subsidence, accounting for more than 50% of the predictive power. The permafrost degradation in many tundra areas is likely dominated by fire-induced changes in the surface energy balance.  From this pilot study, we conclude that our approach has the potential to study fire-permafrost interaction and environmental drivers of surface subsidence at the northern circumpolar scale. Models can also use our data to parameterize subsidence and thermokarst processes associated with permafrost degradation due to fire.