Flood hazard potential evaluation using decision tree state-of-the-art models

Area of Science:

• Environmental Science
• Geospatial Analysis
• Hydrology

Background:

• Floods are frequent and devastating natural disasters globally.
• Accurate flood susceptibility identification is crucial for damage reduction.
• Quantitative flood susceptibility calculation is a growing research area.

Purpose of the Study:

• To accurately map flood susceptibility zones in the Trotus River basin.
• To evaluate the performance of various machine learning algorithms for flood zonation.
• To identify areas with high flood susceptibility in the study region.

Main Methods:

• Application of machine learning algorithms: forest-PA-WOE, best first decision tree-WOE, alternating decision tree-WOE, and logistic regression-WOE.
• Utilizing geographic information system (GIS) for spatial analysis.
• Quantitative calculation and comparison of model performances.

Main Results:

• The forest-PA-WOE model achieved the highest accuracy (0.981).
• Over 16.22% of the Trotus basin is exposed to high and very high flood susceptibility.
• Model performances surpassed previous studies in the same area.

Conclusions:

• Machine learning models, particularly forest-PA-WOE, are highly effective for flood susceptibility zonation.
• The developed models can be reliably applied in data-scarce regions.
• Findings support watershed managers and hazard authorities in flood risk mitigation and planning.