Mass wasting susceptibility assessment of snow avalanches using machine learning models

Area of Science:

• Geosciences
• Environmental Science
• Natural Hazard Assessment

Background:

• Snow avalanches pose severe risks to mountainous regions, causing socioeconomic and environmental damage.
• Assessing snow avalanche susceptibility is crucial for effective hazard mitigation and land-use planning.

Purpose of the Study:

• To evaluate machine learning methods for modeling snow avalanche-induced mass wasting.
• To identify key factors influencing avalanche susceptibility.

Main Methods:

• Four machine learning models were assessed: Generalized Additive Model (GAM), Multivariate Adaptive Regression Spline (MARS), Boosted Regression Trees (BRT), and Support Vector Machine (SVM).
• Recursive Feature Elimination (RFE) identified crucial features for model calibration.
• Model performance was evaluated using accuracy, Kappa, precision, recall, and Area Under the Curve (AUC).

Main Results:

• All models demonstrated strong performance, with accuracy exceeding 0.88 and AUC surpassing 0.89.
• The Support Vector Machine (SVM) model exhibited superior predictive capability compared to GAM, MARS, and BRT.
• Topographic Position Index (TPI) and Distance to Stream (DTS) were identified as the most influential variables in susceptibility mapping.

Conclusions:

• Machine learning, especially SVM, provides a robust framework for snow avalanche mass wasting susceptibility modeling.
• TPI and DTS are critical predictors for understanding and mapping avalanche-prone areas.
• Accurate susceptibility maps are vital for developing effective mitigation strategies and land-use policies in hazardous regions.