Abstract: Groundwater level variations are one of the critical triggering factors for geological disasters such as collapses, landslides, and debris flows.To more accurately predict the evolution of groundwater levels in rainfall-induced landslides, an improved TANK model was proposed based on the influence of rainfall intensity, topography, and geological conditions on groundwater fluctuations.The proposed model introduced three key improvements: first, the traditional rectangular structure was replaced with a trapezoidal structure, enabling a nonlinear instead of linear variation in the water level cross-section; second, the rainfall convergence mechanism was restructured to enhance the sensitivity of water level detection; and third, the model incorporated effective features from monitoring data to calculate and feedback accuracy.Validation using monitoring data from the Taziping landslide demonstrated that the proposed method exhibited high reliability and stability.Compared to the traditional TANK model, the improved TANK model performed better in simulating groundwater level dynamics, with advantages such as shorter training times, higher sensitivity to small and medium-intensity rainfall-induced groundwater changes, and closer alignment with trends observed in monitoring data.In short-, medium-, and long-term predictions, the prediction accuracy (NSE) of the improved model increased by 6.8% to 20.3%, with the highest accuracy observed in medium-term predictions.These findings suggested that the proposed model can effectively simulate the variation trends of groundwater levels under rainfall conditions.The research outcomes can provide a basis for predicting and assessing groundwater status changes in rainfall-induced landslides.