Evaluation of coseismic landslide susceptibility by combining Newmark model and XGBoost algorithm

Area of Science:

• Geosciences
• Earthquake Engineering
• Machine Learning Applications

Background:

• Coseismic landslides pose significant threats in hilly regions following earthquakes.
• Accurate landslide susceptibility assessment is vital for disaster mitigation and forecasting.
• Existing models often lack comprehensive integration of physical principles and data-driven approaches.

Purpose of the Study:

• To develop and validate a novel landslide hazard assessment model by combining the Newmark physical model with machine learning techniques.
• To evaluate the performance of hybrid models against traditional physical and standalone machine learning models.
• To enhance the reliability and generalizability of seismic landslide susceptibility assessments.

Main Methods:

• Integration of the Newmark physical model with machine learning algorithms (XGBoost and Random Forest).
• Utilization of a comprehensive indicator ([Formula: see text]) incorporating rock strata, moisture content, and slope gradient as a key feature.
• Application of Monte Carlo simulations to address uncertainty in geotechnical parameters.
• Development and comparison of six landslide susceptibility models, including hybrid and independent approaches.

Main Results:

• Hybrid models (N_XGB, N_RF) integrating Newmark model outputs with machine learning showed enhanced performance.
• The N_XGB model achieved the highest accuracy (AUC of 0.96) in the Jiuzhaigou region.
• The [Formula: see text] indicator demonstrated superior predictive accuracy compared to Newmark displacement alone.
• Models exhibited robust generalizability with validation accuracies of 0.88 (Ludian) and 0.86 (Luding).

Conclusions:

• Integrating physical principles with data-driven machine learning significantly improves seismic landslide susceptibility assessment.
• The developed hybrid approach offers a reliable framework for regional landslide risk assessment and management.
• The study underscores the effectiveness of combining physical model outputs with advanced machine learning for geological hazard prediction.