Hazard prediction modeling for typhoon-triggered shallow landslides by integrating physically-based model and extreme rainfall analysis

There is no doubt that extreme typhoon and rainstorm events will be significantly affected in coastal regions due to climate change. However, few efficient methods have been proposed to quantify and predict future changes of landslide hazard at regional scales. This study presents an innovative framework regarding this topic in three significant ways: (i) High-resolution satellite images were applied to prepare a landslide inventory during a typhoon event. (ii) A physically-based probabilistic model, namely Monte Carlo optimized TRIGRS model (MC-TRIGRS), was constructed for a probabilistic scrutiny of rainfall-triggered landslides. This model underwent calibration via orthogonal tests to evaluate the importance of input parameters. (iii) Five distribution functions were examined to capture the evolution trends of extreme rainfall frequency versus time windows (Far History, Middle History, and Near History). Landslide hazard maps and their variations under different rainfall scenarios were scrutinized by integrating the MC-TRIGRS model and extreme rainfall analysis. The Wencheng region located in SE China and the 2016 Typhoon Megi were selected as the case study. The compiled landslide inventory comprised 499 shallow landslides over a 609 km2 area. The MC-TRIGRS simulated these landslide events well, with a best-fit accuracy of 82.8% during the parameter calibration. The Generalized Extreme Value (GEV) distribution yielded the best performance when fitting the return period of extreme rainfall. A closer time window to the present presented higher extreme rainfall and landslide hazard areas. The generated map from the whole history rainfall data underestimated the overall landslide hazard level. The proposed approach demonstrates its potential for landslide prediction under extreme scenarios in similar regions.