A machine learning framework for multi-hazards modeling and mapping in a mountainous area.

Area of Science:

• Geosciences
• Environmental Science
• Disaster Risk Management

Background:

• Mountainous regions face recurrent extreme natural events.
• Accurate multi-hazard risk assessment is crucial for disaster mitigation.
• Previous studies often focused on single hazards, limiting comprehensive risk management.

Purpose of the Study:

• To create an accurate multi-hazard risk map for a mountainous region in Iran.
• To model the probabilities of snow avalanches, landslides, wildfires, land subsidence, and floods.
• To evaluate the effectiveness of machine learning models for multi-hazard prediction.

Main Methods:

• Utilized machine learning models: Support Vector Machine (SVM), Boosted Regression Tree (BRT), and Generalized Linear Model (GLM).
• Incorporated climatic, topographic, geological, social, and morphological factors as input variables.
• Validated model performance using Area Under the Curve (AUC), with all best models achieving AUC > 0.8.

Main Results:

• Support Vector Machine (SVM) demonstrated highest accuracy for landslides, land subsidence, and floods.
• Generalized Linear Model (GLM) was optimal for wildfire prediction.
• Functional Discriminant Analysis (FDA) proved most accurate for snow avalanche risk assessment.

Conclusions:

• Machine learning approaches are highly effective for multi-hazard modeling and risk assessment.
• The generated risk maps provide essential data for effective hazard management and disaster mitigation.
• This study offers a valuable baseline for managing future human-environment interactions with natural hazards in the region.