GIS-based spatial modeling of snow avalanches using four novel ensemble models

Area of Science:

• Geosciences and Remote Sensing
• Natural Hazard Assessment
• Spatial Analysis

Background:

• Snow avalanches pose significant risks to lives and infrastructure in vulnerable regions.
• Accurate susceptibility mapping is crucial for disaster risk reduction and land-use planning.
• Existing methods require enhancement for improved prediction accuracy.

Purpose of the Study:

• To map snow avalanche susceptibility in Sirvan Watershed, Iran, using an innovative hybrid modeling approach.
• To compare the performance of different statistical and machine learning models for avalanche prediction.
• To provide a reliable tool for decision-making in avalanche risk management.

Main Methods:

• Integration of statistical models (belief function and probability density) with machine learning models (multi-layer perceptron and logistic regression).
• Utilized remote sensing data and a geographic information system (GIS) for spatial analysis.
• Developed a snow avalanche inventory map from satellite imagery, documentation, and field surveys.

Main Results:

• The hybrid probability density-logistic regression (PD-LR) model achieved the highest accuracy (AUC = 0.941).
• All tested hybrid models (PD-LR, Bel-LR, Bel-MLP, PD-MLP) showed high predictive performance.
• The study successfully identified and validated snow avalanche-prone areas.

Conclusions:

• The proposed hybrid modeling approach offers accurate and reliable snow avalanche susceptibility mapping.
• The PD-LR model is recommended for its superior performance in identifying high-risk zones.
• This methodology supports effective management and decision-making for mitigating avalanche hazards.