Mapping thermokarst lakes in different physical states on the central Tibetan plateau

Mapping and monitoring the spatial–temporal evolution of thermokarst lakes on the Tibetan Plateau can significantly contribute to the estimation of climate change impacts on permafrost degradation. Remote sensing-based methods have been developed to monitor the interannual changes of thermokarst lakes. However, the seasonal changes, which are vital for understanding thermokarst lake dynamics, are commonly overlooked. This paper developed automatic methods which integrated the normalized difference water index (NDWI), modified normalized difference water index (MNDWI), principal component analysis (PCA), and independent component analysis (ICA) with Markov random field (MRF) for rapid mapping of thermokarst lakes in different physical states characterized by season changes. To validate the effectiveness, the proposed methods were applied to map thermokarst lakes on the central Tibetan Plateau, using four Sentinel-2 images representing different evolutionary states. Specifically, the area-to-point regression kriging (ATPRK) method was employed to downscale the 20-m short-wave infrared (SWIR) band to 10 m. Using manually digitized lakes as a reference, the proposed methods achieved an average Kappa coefficient of 0.79, significantly outperforming the water index thresholding method that merely attained 0.44. The results corroborate that the proposed methods can be robustly applied in mapping thermokarst lake dynamics with both interannual and seasonal trends on the Tibetan Plateau.