Quick Northward Retreat of the Permafrost Boundary and Controlling Mechanisms in Northeastern China

The southern permafrost margin of Eurasia forms a highly climate-sensitive transition belt where modest warming can rapidly reorganize snow–soil–vegetation coupling and trigger nonlinear retreat. However, long-term, spatially continuous indicators of cold-season thermal penetration remain scarce, limiting the attribution of freeze-depth changes to coupled hydrothermal and land-surface controls. In this study, an integrated framework was developed by combining a surface frost number model, multisource environmental predictors and machine-learning-based SFD (soil freezing depth) inversion, and piecewise structural equation modeling (SEM) to explore the spatiotemporal evolution of permafrost in northeastern China, which is located at the southernmost tip of the Eurasian continent. By coupling an ensemble-mean gradient-boosted tree mapping framework, a spatially continuous SFD for 1980–2020 was reconstructed with high accuracy (R2 = 0.70, RMSE = 34.24 cm) under the station-wise chronological split. The regional average SFD decreased from 171.20 cm in 1980 to 154.25 cm in 2020, corresponding to an overall trend of −0.25 cm/y. A quick northward retreat of the permafrost southern boundary was identified, with the median latitude shifting from 48.37°N to 49.92°N at a rate of 4.31 km/y, producing a degrading permafrost belt of 1.75 × 105 km2 over the past four decades. Three permafrost types, i.e., consistent permafrost, degradation belt and seasonally frozen ground, were identified with contrasting changing characteristics. SEM analysis revealed a contrasting dominant type and corresponding controlling mechanism on the SFD across the three regions. The land surface thermal state regulates the SFD in the consistent permafrost region, whereas vegetation strongly indirectly controls the SFD through modulation of thermal state and soil moisture in the degradation belt. In contrast, hydroclimatic and snow processes are increasingly influential in seasonally frozen ground. Our findings provide a transferable framework for diagnosing changes in SFD and its drivers in warming permafrost transition zones beyond the eastern Eurasian permafrost margin and may be helpful in guiding the modeling of its evolution against the background of climate change.