Mapping subsurface combustion cavities in underground coal fires using nonlocal DS-InSAR and physics-based inversion

Underground coal fires threaten environmental sustainability, resource conservation, and infrastructure safety, yet their concealed nature makes subsurface combustion processes difficult to characterize using a single observation technique. In this paper, we present an integrated remote sensing framework combining deformation and thermal observations with physics-based inversion to investigate underground coal-fire dynamics in the Shuixigou coalfield, Xinjiang, China. First, a nonlocal distributed scatterer InSAR approach was applied to 32 Sentinel-1A images spanning January 2018 to December 2019, yielding high-density time-series line-of-sight deformation. Meanwhile, Landsat-8 thermal infrared imagery was processed using a generalized single-channel algorithm to derive time-series land surface temperature. Moreover, we characterized the spatiotemporal evolution of coal-fire activity through a joint analysis of the deformation and thermal time series. Furthermore, the cumulative deformation field was then incorporated into an Okada rectangular dislocation model with a smooth regularization constraint, yielding the geometry of equivalent combustion cavities. In our experimental coal-mining area, two major subsidence basins were identified, with maximum cumulative line-of-sight displacements of approximately −160 mm and −120 mm. Besides, the inferred combustion sources were located at depths of 53.11 m and 80.65 m, consistent with independent geophysical survey estimates of 56.0 m and 85.5 m. Combined analysis further reveals that long-term combustion zones are associated with concentrated subsidence, while thermal anomalies preferentially occur along fracture systems surrounding subsidence centers, indicating distinct thermal and mechanical responses.