Hydrological reorganization of reconstructed land surfaces in an open pit coal mine assessed through remote sensing SWAT simulation and interpretable machine learning

Open-pit coal mining and subsequent restoration can reshape land-surface morphology, land-use composition, and hydrological connectivity, but hydrological responses within reconstructed mining landscapes remain insufficiently understood at the hydrological response unit (HRU) scale. Here, a nested HRU-scale framework was developed for the Juhugeng mining area of the Muli coalfield by integrating remote sensing, SWAT simulation, cross-period HRU matching, machine-learning classification, and SHAP interpretation. Three scenarios representing pre-mining conditions in 2006, post-mining conditions in 2014, and post-restoration conditions in 2024 were constructed, with 2014 HRUs used as baseline units for stage comparison. Four groups of remote-sensing/GIS-derived drivers were considered, including topographic reconfiguration, land-use composition change, landscape pattern, and structural reorganization/connectivity. Repeated grouped validation under the native-resolution SWAT-HRU configuration yielded accuracies of 0.89 ± 0.04 and 0.86 ± 0.03 for LATQ during the mining and restoration stages, respectively, compared with 0.80 ± 0.05 and 0.76 ± 0.07 for SURQ. However, after the three DEMs were harmonized to a common approximately 12.49-m grid and the topographic predictors were recalculated, LATQ accuracies decreased to 0.74 ± 0.05 and 0.73 ± 0.04, whereas SURQ accuracies remained at 0.79 ± 0.05 and 0.74 ± 0.07. The common-resolution analysis retained a dominant topographic contribution to LATQ and the first-ranked position of mean-slope change, but the rankings of slope range and slope variability were resolution-sensitive. Accordingly, the LATQ–slope relationship is interpreted as a broader, resolution-dependent model association rather than as evidence of three independent topographic controls. Adding nested-context variables did not produce a consistent improvement across repeated grouped validation, indicating that the HRU-only versus nested-context comparison should be interpreted as an exploratory test of scale-related information gain rather than definitive evidence of hydrological scale dependence. These findings provide a reproducible, model-based framework for diagnosing process-specific hydrological reorganization in disturbed and reconstructed mining landscapes, while highlighting the need for cautious interpretation where field validation is unavailable.