Improving index-based coastal vulnerability assessment using machine learning in Oman.

Area of Science:

• Environmental Science
• Geospatial Analysis
• Machine Learning Applications

Background:

• Coastal vulnerability assessments are vital for understanding environmental hazard impacts.
• Traditional index-based methods often fail to weigh parameters effectively.
• Oman's coastline requires robust vulnerability mapping due to its environmental significance.

Purpose of the Study:

• To integrate machine learning models with index-based approaches for improved Coastal Vulnerability Index (CVI) calculation.
• To compare the performance of machine learning (Random Forest, XGBoost) with traditional methods (AHP, Shannon's Entropy) for CVI mapping.
• To identify key vulnerability parameters and their distribution along Oman's coastline.

Main Methods:

• Utilized Particle Swarm Optimization to tune Random Forest and XGBoost models.
• Employed feature importance analysis to determine parameter weights for CVI calculation.
• Compared machine learning-derived CVI maps with those from Analytical Hierarchy Process (AHP) and Shannon's Entropy.

Main Results:

• Geomorphology was the most influential parameter, indicating moderate to very high vulnerability in many areas.
• Elevation and slope showed very low vulnerability across most of Oman's coastline.
• Significant differences in CVI results were observed across methods, with machine learning providing a more flexible approach.

Conclusions:

• Integrating machine learning with index-based methods offers a more nuanced and flexible approach to coastal vulnerability assessment.
• Different methodologies prioritize coastal factors distinctively, impacting CVI outcomes.
• The study provides comprehensive CVI maps for Oman's coastline, highlighting areas of concern.