Prediction of deformation-gradient-exceedance-induced D-InSAR undetectable-zone boundaries and boundary-constrained fusion for full-basin mining subsidence estimation

Mining-induced large deformation gradients can cause phase-unwrapping failures in differential interferometric synthetic aperture radar (D-InSAR) observations, making gradient-exceedance-induced undetectable-zone boundaries difficult to identify and limiting refined full-basin subsidence estimation. This study develops a prediction model-based undetectable-zone boundary prediction model (PM-UBPM) and a boundary-constrained deformation field fusion model (BCDF-PM) to predict these undetectable-zone boundaries and reconstruct refined full-basin mining subsidence. Multi-scenario simulations showed that the unwrapping-failure regions derived by the minimum-cost-flow (MCF) method, especially their outer boundaries, were generally consistent with the simulated line-of-sight (LOS) deformation-gradient-exceedance regions, providing the basis for prior prediction of D-InSAR undetectable-zone boundaries. PM-UBPM uses temporal probability integral method (TPIM)-predicted LOS deformation gradients, a coherence-dependent detectability threshold, and spatial-connectivity correction to predict undetectable-zone boundaries before deformation inversion. In simulations, the predicted regions achieved mean intersection-over-union (IoU) and Dice coefficients of 0.887 and 0.939 against MCF-derived unwrapping-failure regions. In a Yuzhou mining case, D-InSAR-derived subsidence agreed with levelling measurements in the predicted detectable zone, with a root-mean-square error (RMSE) of approximately 16 mm, whereas the predicted undetectable zone showed severe distortion, with an RMSE of approximately 1349 mm. Using the predicted boundary, BCDF-PM fused D-InSAR-observed and TPIM-predicted deformation through a nonlinear weighting function, alleviating discontinuities at heterogeneous-data fusion boundaries and reducing the RMSE in the predicted detectable zone to 14 mm. The proposed framework provides a prior constraint for D-InSAR reliability assessment and boundary-constrained full-basin subsidence reconstruction in large-gradient mining areas.