Tepetidis et al. (2025) Combining Machine Learning Models and Satellite Data of an Extreme Flood Event for Flood Susceptibility Mapping

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2025
• Date: 2025-09-10
• Authors: Nikos Tepetidis, Ioannis Benekos, Theano Iliopoulou, Panayiotis Dimitriadis, Demetris Koutsoyiannis
• DOI: 10.3390/w17182678

Research Groups

Not provided in the text.

Short Summary

This study applies and evaluates four machine learning models for flood susceptibility mapping in Thessaly, Greece, identifying that tree-based models (Random Forest and XGBoost) achieve superior accuracy and reveal approximately 20% of the basin as highly flood-prone.

Objective

• To apply and evaluate machine learning models (Logistic Regression, Support Vector Machine, Random Forest, eXtreme Gradient Boosting) for creating flood susceptibility maps (FSMs) in Thessaly, Greece, a flood-prone region.

Study Configuration

• Spatial Scale: Thessaly region, Greece; basin scale.
• Temporal Scale: Recent years, with Storm Daniel as a primary reference event for training; flood susceptibility map developed for a 1000-year return period rainfall scenario at a 24-hour scale.

Methodology and Data

• Models used: Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost).
• Data sources: 13 explanatory variables derived from topographical, hydrological, hydraulic, environmental, and infrastructure data; satellite data of affected areas for significant variables; Storm Daniel event data for model training.

Main Results

• Tree-based models (Random Forest and XGBoost) outperformed other classifiers in accuracy.
• The Random Forest model achieved an Area Under the Curve (AUC) value of 96.9%.
• XGBoost achieved an AUC of 96.8%, SVM 94.0%, and LR 90.7%.
• A flood susceptibility map was developed for a 1000-year return period rainfall scenario at a 24-hour scale.
• Analysis revealed that approximately 20% of the basin is highly prone to flooding.

Contributions

• Demonstrates the application and comparative performance of four machine learning algorithms for flood susceptibility mapping in the flood-prone Thessaly region of Greece.
• Identifies the superior performance of tree-based models (Random Forest and XGBoost) in this specific regional context.
• Develops a practical flood susceptibility map for a significant return period (1000-year, 24-hour rainfall) to support long-term flood risk assessment and planning.
• Highlights the potential of machine learning to provide accurate and practical flood risk information, enhancing flood management and disaster preparedness in the region.

Funding

Not provided in the text.

Citation

@article{Tepetidis2025Combining,
  author = {Tepetidis, Nikos and Benekos, Ioannis and Iliopoulou, Theano and Dimitriadis, Panayiotis and Koutsoyiannis, Demetris},
  title = {Combining Machine Learning Models and Satellite Data of an Extreme Flood Event for Flood Susceptibility Mapping},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17182678},
  url = {https://doi.org/10.3390/w17182678}
}