[A-TT35] Machine Learning Techniques in Weather, Climate, Ocean, Hydrology and Disease Predictions

[ATT35-09] Application of Machine Learning Models for Rainfall-Induced Landslide Prediction: Evaluating the Impact of Hydrological Parameters from SWAT

*Chih-Mei Lu¹, Li-Chi Chiang¹ (1.NTU)

Keywords:Landslides Prediction, Machine Learning, SWAT Model, Hydrological parameters, Rainfall-induced landslides

Landslides are among the most devastating natural disasters worldwide, often triggered by extreme rainfall events and resulting in significant human and economic losses. In recent years, machine learning (ML) techniques have been widely applied to rainfall-induced landslide (RIL) prediction, enhancing predictive accuracy while reducing the limitations of traditional empirical methods. To evaluate the performance of different ML models in landslide prediction, four models - Generalized Linear Model (GLM), Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) - are compared by incorporating hydrological parameters, specifically percolation (PERC) and the fraction of water content in soil (FWC) simulated using the Soil and Water Assessment Tool (SWAT). Additionally, four landslide-related parameters, including daily and two-day cumulative rainfall, land use, and soil type, are used as input variables for landslide prediction. A total of 11 rainfall events that resulted in landslides and 27 rainfall events without landslides from 2003 to 2022 are analyzed in conjunction with hydrological simulation results. The ML models are trained using both the landslide-related parameters and the two hydrological parameters. To address data imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is applied. Model performance is evaluated using Accuracy, Precision, Recall, F1-score, ROC-AUC, and Matthews Correlation Coefficient (MCC). The results indicate that incorporating hydrological parameters significantly improves landslide prediction accuracy, particularly in the RF and XGBoost models, highlighting the crucial role of hydrological conditions in slope stability. Moreover, the SWAT model provides valuable parameter information for landslide prediction. Therefore, integrating hydrological models with ML techniques for landslide forecasting can contribute to the development of more robust early warning systems.