Abstract: Improving the accuracy of susceptibility prediction for rainfall-induced landslides and establishing suitable rainfall threshold models are of great significance for regional landslide hazard assessment. Taking Fuling District of Chongqing as a case study, the information value model, BP neural network model, random forest model, information value-BP neural network coupled model, and information value-random forest coupled model were used to evaluate regional landslide susceptibility. By comparing the receiver operating characteristic (ROC) curves, area under the curve (AUC), and susceptibility distribution patterns of different models, a critical monthly average rainfall threshold model for landslides is proposed, and critical monthly average rainfall thresholds for different temporal probabilities were inferred. The susceptibility results were coupled with temporal probability levels to produce regional landslide hazard assessment results. The evaluation accuracy is further validated with 30 randomly selected landslide events and 4 typical landslide cases. The results show that coupling the Information Value and machine learning models compensates for the shortcomings of machine learning in early data input and non-sample selection, enhancing the predictive accuracy of single machine learning models. Among these, the information value-random forest coupled model exhibits the highest predictive accuracy; of the 30 randomly selected landslide samples, 20 cases (67%) occurred in areas with a temporal probability of over 50%, validating the accuracy of the critical monthly average rainfall threshold model. The 4 typical landslide samples selected randomly were primarily in the P4 or P5 temporal probability levels and were located in high to very high-risk areas, aligning well with field survey results. This indicates that the regional landslide hazard assessment based on the information value-random forest coupled model and the critical monthly average rainfall threshold is accurate and reliable.