Revealing the impacts of environmental and anthropogenic factors on permafrost deformation in the central Qinghai-Tibet Plateau using InSAR and interpretable machine learning

• Interpretable ML reveals environmental impacts on permafrost deformation. • Soil hydrothermal effects are nonlinear, with moisture as the key driver. • Long-term vegetation changes regulate deformation in permafrost regions. • Differentiated impacts of infrastructure and human activities on deformation. Permafrost stability on the Qinghai-Tibet Plateau (QTP) is vital amid environmental changes and human activities. Interferometric Synthetic Aperture Radar (InSAR) effectively monitors permafrost deformation, capturing seasonal and long-term trends via active layer hydrothermal shifts. While prior studies emphasized large-scale deformation, finer-scale regulatory effects of environmental factors and human activities remain underexplored. We used Small Baseline Subset (SBAS)-InSAR to derive ground deformation time series and developed an XGBoost-based model to assess impacts of soil hydrothermal conditions, vegetation, terrain, and human activities. Our results show that soil moisture plays a dominant role in permafrost deformation, with its interaction with soil temperature exhibiting nonlinear effects on permafrost deformation. Long-term changes in Normalized Difference Vegetation Index (NDVI) were significantly positively correlated with the seasonal deformation amplitude (R 2 = 0.3546, p < 0.001). Geomorphons exerted significant control, with valleys and lowlands exhibiting reduced permafrost deformation due to distinct hydrothermal conditions, whereas highlands demonstrated greater stability. Human infrastructure further influenced ground deformation. The Wudaoliang (WDL) Train Station (median subsidence rate: −11.55 mm/yr) and Qinghai-Tibet Railway (median subsidence rate: −8.75 mm/yr) exhibited strong regulatory effects, whereas WDL Town (median subsidence rate: −12.45 mm/yr) and the Qinghai-Tibet Highway (median subsidence rate: −10.32 mm/yr) experienced more pronounced deformation, which highlights the importance of engineering design in mitigating permafrost degradation. The results provide new insights into the regulatory mechanisms of environmental factors and human activities on permafrost deformation across the QTP using interpretable machine learning (ML), which is important for environmental conservation and engineering disaster prevention in permafrost regions.