Abstract: The rainfall threshold is one of the most commonly used landslide warning methods currently. However, existing empirical rainfall thresholds are mainly aimed at regional warning of landslides, lacking discussion on the rainfall thresholds for individual landslides within the region that vary spatially. Based on historical rainfall-induced landslide data and hourly rainfall data in Bazhong City from 2014 to 2021, this study employs Kriging interpolation methods to extract four types of short-term rainfall (1 hour, 12 hours, 24 hours, 72 hours) and their corresponding long-term rainfall (7 days before the landslide occurrence). In these four threshold models, we calculate the distribution of long-term and short-term rainfall thresholds in each group, and then validate them by landslide disaster data from 2021. The research results indicate that the prediction accuracy of the four types of threshold models ranges from 40% to 65%, suggesting they have good potential for practical application. Additionally, the prediction accuracy improves with the increase in the duration of short-term rainfall. The prediction accuracy for rainfall thresholds calculated from 72-hour-7-day model is highest, which can reach 62%; while the prediction accuracy for the 1-hour-7-day model is 46%. Based on the highest prediction accuracy of these models, this study calculates the optimal ratios for short-term and long-term disaster-causing rainfall for four types of models, which lead to a quantitative division between short-term rainfall-induced landslides and long-term rainfall-induced landslides. Through the calculation of the spatial distribution of disaster-causing rainfall, this study was able to extract rainfall thresholds at potential landslide locations, achieving the goal of one threshold per site in the region, and enhancing the existing models for calculating rainfall thresholds.