Abstract: Mining activities superimposed on karst terrains are highly susceptible to triggering karst collapse, particularly during mine dewatering and post-closure groundwater recovery. Traditional monitoring methods often fail to adequately capture regional groundwater dynamics due to sparse monitoring networks, thereby limiting the accuracy of karst collapse early warning and susceptibility assessments. Here, this study investigated the Meitanba Mining Area in central Hunan, a region characterized by extensive karst development, integrating geological survey, borehole exploration and historical water level monitoring, and established a groundwater numerical simulation model via GMS software to simulate variations in groundwater levels. The mechanism of karst collapse geological hazards induced by water level fluctuations during the process of mine drainage and pit water level recovery processes was explored. Historical water level monitoring was used to verify the model. Furthermore, the response relationship between groundwater level fluctuations and historical collapse events was revealed. Combining evaluation indicators such as karst development degree, overburden thickness, groundwater level variation amplitude, and historical collapse distribution, susceptibility zoning assessment was conducted using the Analytic Hierarchy Process (AHP) and Fuzzy Comprehensive Evaluation Method. The numerical simulations demonstrated good predictive performance, with groundwater-level prediction errors of less than 0.5 m across 72% of the study area. The synergistic effects of factors such as overburden soil softening, suffosion, air pressure fluctuations, and buoyancy changes induced by groundwater level variations contribute to karst collapse hazards. In the Meitanba area of central Hunan, karst collapse is highly likely to occur when the variation amplitude of groundwater level approaches or exceeds 0.51 m/h during water level rise, or approaches or exceeds 0.96 m/h during water level decline. Given that groundwater level fluctuates frequently and is affected by multiple factors, the lower-limit principle is adopted for risk early warning and prevention. Accordingly, the groundwater level variation amplitude of approaching or exceeding 0.51 m/h is defined as the threshold for inducing karst collapse in the study area. The susceptibility evaluation model integrating this critical threshold with multi-source data achieves an accuracy of 91.4%, providing valuable insights for karst collapse risk prevention in mining areas and disaster mitigation practices in similar regions.