Precise Prediction of Debris Flow Initiation Zones Derived from General Landslide Inventory Using Machine Learning

Abstract

Debris flows are among the most destructive geotechnical hazards in mountainous regions due to their rapid movement and high destructive potential, often causing significant loss of life and property. The main objective of this study is to derive potential debris flow initiation areas from existing landslide inventory and to predict these areas more accurately using machine learning methods. In this study, large-area landslide polygons obtained from the MTA General Landslide Inventory were transformed into potential debris flow initiation points through geometric transformation and geomorphological threshold filtering. This procedure aims to reduce spatial uncertainties associated with the direct use of conventional landslide inventory data and to generate a more representative dataset for susceptibility modeling. The derived initiation points were analyzed using a machine learning model together with morphometric, environmental, and geological variables. Multicollinearity among the predictor variables was evaluated using Variance Inflation Factor (VIF) analysis. Model performance assessed by ROC analysis yielded an AUC value of 0.946, indicating high predictive capability.