Temporal Modeling of Rainfall-Triggered Landslides: A Hybrid Approach Combining Physically-Based Modeling and Extreme Value Analysis

Climate change induced the rise of extreme rainfall, resulting in an increase in the frequency and magnitude of landslides. Hence, a novel temporal modeling of rainfall-induced landslides incorporating both the dynamic nature of rainfall patterns and the slope failure mechanism was proposed. The proposed approach consists of three steps: (1) analysis of a critical continuous rainfall (CCR) using a physical-based model, (2) obtaining the cumulative distribution function of generalized extreme value distribution via the annual maximum rainfall series, and (3) analysis of temporal probability map. The result of the CCR map was validated with the 2018 landslide event in a small area of Hiroshima Prefecture, Japan. The result shows that the CCR map is highly reliable, with an AUC of 71.3%. The proportion of temporal probability >0.5 under the nonstationary model is greater than approximately 1.7, 1.9, 2.0, and 2.3 times the stationary model for the periods of 5, 10, 20, and 50 years, respectively. This indicates that the temporal probability increases according to a longer time period due to climate change-induced increased trend of extreme rainfall. The proposed approach can also be utilized to obtain the landslide temporal probability map for areas lacking landslide inventory.