Abstract: Considering the rainfall characteristics and geological environment of historical landslides in the region, it is important to construct a reasonable rainfall threshold model for the prediction and prediction of regional landslide disasters. In this paper, 1368 rain-induced landslides in Chongqing from 2013 to 2021 were selected as the research objects. Based on detailed landslide cataloging data and regional geological environment data, rainfall events were divided and landslide sample database was constructed. Based on Random forest (RF) algorithm, a rainfall threshold model of the day's triggered rainfall - previous effective rainfall (Rd-Rp) was established, and different warning levels were divided based on probability classification. Based on the medium early warning probability, 19 threshold criterions of critical rainfall were constructed according to four geological environments, namely disaster pregnant environment, engineering rock group, land use and soil type. The results show that the Rd Rp threshold model can well reflect the impact of the daily excitation rainfall and the early effective rainfall on the landslide development, and the early effective rainfall plays a leading role in the landslide induction. Different geological environment types have different sensitivity to rainfall factors, and tectonics, engineering properties of rock and soil mass, vegetation coverage and soil properties have important effects on landslide development. The threshold of rock formations mainly composed of hard rocks is higher than that of rock formations mainly composed of soft rocks, and the threshold of forest land is higher than that of cultivated land. The threshold of areas severely damaged by human engineering activities will decrease. The study can provide new ideas and reference for the meteorological early warning management of rainfall landslides in Chongqing.