Assessing and mapping multi-hazard risk susceptibility using a machine learning technique

Area of Science:

• Environmental Science
• Geosciences
• Disaster Management

Background:

• Natural hazards pose significant risks to urban and watershed areas.
• Effective disaster management requires accurate susceptibility assessments.
• Understanding multi-hazard interactions is crucial for comprehensive risk evaluation.

Purpose of the Study:

• To conduct a multi-hazard probability assessment for floods, forest fires, and landslides in Fars Province, Shiraz City, and its watersheds.
• To identify and prioritize factors influencing each hazard type.
• To generate susceptibility maps for informed disaster mitigation strategies.

Main Methods:

• Utilized the Boruta algorithm to identify and prioritize key factors for floods, forest fires, and landslides.
• Employed a Random Forest (RF) model for creating susceptibility maps.
• Validated models using Area Under the Curve (AUC) and other accuracy metrics.

Main Results:

• Over 42% of the area is not susceptible to any hazard, while 2.67% faces risk from all three.
• Shiraz City shows high susceptibility to flooding (25%) and landslides (16%).
• Specific watershed risks identified: Dorodzan (landslides, floods), Maharlou (floods), Tashk-Bakhtegan (floods, landslides), and Ghareaghaj (forest fires, floods).

Conclusions:

• The study provides a robust multi-hazard susceptibility assessment for Fars Province.
• Developed susceptibility maps offer valuable insights for researchers and stakeholders in disaster risk reduction.
• Findings support the forecasting of spatial hazard behavior under various mitigation scenarios.