Mitigating systematic overestimation of soil organic carbon in underground mining areas by integrating DS-InSAR-derived subsidence factors

Surface subsidence induced by underground mining results in topsoil disruption and loss of soil organic carbon (SOC). However, conventional optical remote sensing often fails to detect the full extent and intensity of such subsidence, leading current digital soil mapping techniques to inevitably overestimate SOC in underground mining areas. This study integrated multi-temporal surface subsidence factors derived from Sentinel-1 imagery and distributed scatterer interferometric synthetic aperture radar (DS-InSAR) technology into an SOC spatial prediction framework for an underground coal mining area in northern Pei County, China. By employing ensemble learning and SHAP analysis, we quantified the spatial patterns and drivers of SOC. Results showed that surface subsidence from underground coal mining affected 40.4% of farmland during 2015–2023, and the mean SOC content in subsided farmland (12.277 g/kg) was significantly lower than that in undisturbed farmland (14.105 g/kg). After the incorporation of subsidence factors, including surface subsidence rate (SSR), distance to subsidence waterlogged areas, and topographic factors corrected with subsidence displacement, the prediction R 2 of the five ensemble learning models for SOC increased by 6.7%–13.3%. Crucially, the mean predicted SOC content decreased by 6.6% to 9.9% across the models, quantitatively confirming that neglecting subsidence factors results in a systematic overestimation of SOC. We further identified a threshold-driven interplay between subsidence and reclamation: in severely subsided areas (SSR < −8 mm/a), the SOC remained low even after reclamation; in moderately subsided areas (SSR between −8 and −1 mm/a), prioritized reclamation maintained high SOC through soil reconstruction and nutrient inputs; whereas in slightly subsided but unreclaimed areas (SSR > −1 mm/a), the SOC continued to show a declining trend due to subsidence. These findings demonstrate that integrating DS-InSAR-derived subsidence factors is essential for mitigating systematic prediction bias in undermined landscapes.