Mapping Drained Lake Basins in Arctic Lowland Permafrost Regions

Abstract

Lakes and drained lake basins (DLBs) cover 20% to 75% of permafrost lowland regions of the pan-Arctic. Lakes and DLBs of different ages create a heterogeneous and dynamic mosaic of terrain units, provide unique habitats, and substantially impact biogeochemical cycling. The formation of DLBs leads to successional dynamics that include permafrost aggradation, soil formation, and carbon sequestration. A better understanding of widespread lakesuch as DLB systems is needed to understand future dynamics, permafrost hydrology, and improve the representation of Arctic landscapes in climate and Earth system models. Here we present a novel scalable mapping approach to delineate DLB area extent. To ensure applicability across the pan-Arctic, we employ a Random Forest model, utilizing multispectral remote sensing data (Landsat-8 Tasseled Cap index), surface characteristics, and terrain information, going beyond the lake-rich coastal lowland to include boreal forest and discontinuous permafrost zones. The method was tested for a range of study areas across the Arctic, representing different permafrost, climatological, and geological settings. The resulting RF models show high accuracies (0.82–0.86) and perform well in all tested settings. Model evaluation reveals high feature importance of terrain parameters, such as slope and elevation above stream (EAS), with EAS being a novel addition, not having been used by previous DLB mapping activities. A comparison with previously published datasets and approaches reveals reduced uncertainty and an increase in DLB area captured. The availability of pan-Arctic applicable models will enable monitoring of critical landscape change in permafrost lowlands at annual to decadal scales.