Abstract: Accurate evaluation of the water-bearing capacity of weathered bedrock aquifers is essential for water hazard prevention in shallow coal seam mining in northern Shaanxi. To address the limitations of conventional smoothness-constrained inversion in Magnetic Resonance Sounding (MRS), including insufficient resolution of aquifer boundaries, inaccurate estimation of volumetric water content (w) and mean relaxation time ( T_2^* ), and the consequent large deviations in hydraulic conductivity estimation, a full-envelope signal focusing inversion method based on Minimum Gradient Support (MGS) regularization is proposed. Within the framework of Iteratively Reweighted Least Squares (IRLS), MGS regularization is introduced to impose focusing constraints on w and T_2^* , forming a hybrid objective function with optional prior weighting. A dynamic weight updating mechanism is employed, in which a focusing weight matrix is constructed from the gradient information of model parameters at each iteration to control the layering capability of w and T_2^* . Linear subproblems are efficiently solved using the preconditioned conjugate gradient method, and a projection operator is applied to impose physically feasible bounds. The focusing factor β governs the degree of stratification. A smaller β enhances interface focusing, while a larger β reduces to smoothness constraints. The optimal β is determined by comparing the results of MGS inversion with those of smoothness-constrained inversion. Numerical simulations demonstrate that MGS inversion significantly improves the stratigraphic characterization of w and T_2^* . For the weathered bedrock aquifer model, the relative error of w is approximately 6% compared with about 20% for smoothness constraints, and the upper and lower aquifer boundaries can be clearly resolved. Field data from the Hongliulin coal mine indicate that, when aquifer boundaries are determined by water content gradients, the positioning error of MGS is about 14.1%, much lower than 43.6% for smoothness constraints and 34.4% for the commercial Samovar software. Based on the Seevers model, the hydraulic conductivity calculated from MGS inversion results deviates by about 31.35% from pumping test values, representing improvements of approximately 15% and 47% compared with smoothness-constrained inversion and Samovar, respectively. MGS-based focusing inversion markedly enhances the resolution of aquifer boundaries and improves the inversion accuracy of w and T_2^* , thereby indirectly increasing the reliability of hydraulic conductivity estimation using the Seevers model. The focusing factor β is a key parameter controlling the balance between stratification capability and smoothness, and more intelligent and robust strategies for selecting β will be the focus of future research. At present, the application of MRS in coal mines remains relatively limited. This study presents measured MRS data and inversion results for the weathered bedrock aquifer of the Hongliulin coal mine. The measured signals clearly capture the decay characteristics of NMR responses generated by groundwater, and the inversion results show good consistency with pumping test data from the same location, suggesting that the application potential of MRS in coal mine water hazard investigation warrants further development.