An integrated method for assessing vulnerability of buildings caused by debris flows in mountainous areas

Abstract. The vulnerability assessment of buildings in future scenarios is critical to decreasing potential losses caused by debris flows in mountainous areas due to complex topographical conditions that could increase environmental vulnerability to climate change. However, a lack of reliable methods limits the accurate estimation of a physical damage and the associated economic loss. Therefore, an integrated method of physical vulnerability matrix and machine learning model was developed to benefit the estimation of damage degree for buildings caused by a future debris-flow event. By considering the building structures (reinforced-concrete (RC) frame and non-RC frame), spatial positions between buildings and the debris-flow channels (horizontal distance (HD) and vertical distance (VD)), and impact pressure (Pt) on buildings, a physical vulnerability matrix was proposed that links physical damage with the four factors. In order to overcome the difficulty in estimating the possible impact pressure on buildings, an ensemble machine learning (ML) model (XGBoost) was developed with the involvement of geological factors. Additionally, HD and VD were decided based on satellite images. The Longxihe Basin, Sichuan, China, was selected as a case study. The results show that the ML model can achieve a reliable impact pressure prediction because the mean absolute percentage error (MAPE), root mean square error (RMSE), and mean absolute error (MAE) values are 9.53 %, 3.78 kPa, and 2.47 kPa, respectively. Furthermore, 13.9 % of buildings in the Longxihe Basin may suffer severe damage caused by a future debris-flow event, and the highest economic loss is found for a residential building, reaching EUR 5.1×105. Overall, our work can provide scientific support for the site selection of future constructions.