: Arctic permafrost is experiencing unprecedented degradation, and the resulting thaw settlement hazards pose severe threats to the life and infrastructure of local people. However, a gap remains in our comprehensive understanding of the controlling factors and training sample selection strategies. Combining GeoDetector with machine learning models, we conducted a comprehensive evaluation based on 12 factors, including topographic, edaphic, vegetation cover, and climatic properties. The results revealed that among the factors inducing thaw settlement events, the interaction between ground ice content (GIC) and thawing degree day (TDD) played the most critical role ( q = 0.71), whereas GIC alone demonstrated moderate explanatory power ( q = 0.39). This finding indicated, to a certain extent, that the presence of ground ice was a necessary condition, rather than a deterministic factor, for thaw settlement in permafrost regions. The thaw settlement susceptibility map produced by the optimal machine learning model implied that more than 40% of the circum-Arctic permafrost was in high- and very high-susceptibility regions, with a relatively small proportion (13.81%) in medium-susceptibility regions. The hazard risk assessment results revealed that, 2.71 million people and buildings covering 204 km 2 and 110,312 km of roads are currently under the threat of varying risk levels. Notably, the proportions of high-risk (including very high-risk) regions were 73.34%, 75.74%, and 64.8%, respectively. High-risk regions clearly exhibited a high intensity of human activity. Our results demonstrated a notably increased threat of thaw settlement in the circum-Arctic permafrost region compared with previous research.
NI et al. (Mon,) studied this question.