Susceptibility assessment for rainfall-induced landslides using a revised logistic regression method

Landslide susceptibility is the likelihood of a landslide occurring in an area. The logistic regression (LR) method is one of the most popular methods for landslide susceptibility assessment. For rainfall-induced landslides, yearly or monthly rainfall is commonly used to establish a landslide susceptibility model by the LR method. It is a static susceptibility model, which limits the application to predict future landslide probability under potential rainfall event. This study presents a revised logistic regression method to achieve dynamic landslide susceptibility prediction under cumulative daily rainfall. Five kinds of cumulative daily rainfall are used in the landslide susceptibility assessment. The latest landslide events are used to update the landslide susceptibility model. The receiver operation characteristic curve and area under curve are utilized to evaluate the prediction reliability. The landslide susceptibility assessment in Shenzhen is taken as an illustration of the proposed method. The result indicates the method is capable to achieve a high accuracy of 91.9% when the landslide susceptibility model is updated using seven extreme rainfall events in the past 10 years. This method provides an advance prediction of the potential geo-hazards for a large area using the future rainfall forecast.