Travel distance estimation of landslide-induced debris flows by machine learning method in Nepal Himalaya after the Gorkha earthquake

Abstract

Debris flows are more likely to be triggered in the earthquake-strike areas with a widespread presence of unstable slopes, causing severe casualties and changing the surrounding natural topography. In such scenario, estimating the travel distance of debris flows becomes crucial to understand the hazardous areas. Therefore, a hybrid machine learning model (GA-XGBoost) was employed to achieve a reliable estimation of debris-flow travel distance. This model was applied to the Nepal Himalayas, the site of the 2015 Gorkha earthquake. We selected four geomorphological factors for travel distance estimation. They are the volume of failure mass (VL), the height difference between the material source center and end point of movement mass (H), the mean gradient of the travel path (J), and the mean curvature of the travel path (C). Furthermore, to eliminate the noise information and enhance stability of input data, a principal component analysis (PCA) was used to generate three principal components (PC1, PC2, and PC3) from the selected factors to serve as input variables of model development. The performance of this model was evaluated using the assessment indexes, resulting in a mean absolute percentage error (MAPE) of 8.71%, a root mean square error (RMSE) of 144.3 m, and a mean absolute error (MAE) of 86.1 m. Four empirical approaches were also introduced for comparison analysis. Our proposed model has proven to be superior and effective, as the estimated results closely match the actual values. All the results affirm the suitability of our developed model for estimating the travel distance of landslide-induced debris flows following a strong earthquake.