A hybrid machine learning modelling for optimization of flood susceptibility mapping in the eastern Mediterranean

Abdo, Hazem Ghassan; Richi, Sahar Mohammed; Alqadhi, Saeed; Zeng, Taorui; Prasad, Pankaj; Kotaridis, Ioannis; Alharbi, Maged Muteb; Khaddour, Lina A.; Mallick, Javed

Abstract

Floods are considered one of the most destructive natural disasters due to the human and economic losses caused. The Eastern Mediterranean region is subject to devastating annual flood events due to the complex geographical characteristics of this region. Precise and reliable flood susceptibility prediction represents a complex and critical gap in the Eastern Mediterranean that provides a solid basis for developing flood risk management measures. Integrating machine learning (ML) algorithms and geospatial techniques represents a unique tool for reliable flood susceptibility prediction. This evaluation aims to improve flood susceptibility prediction in the Eastern Mediterranean by comparing the performance of four ML algorithms. This evaluation aims to optimize flood susceptibility prediction in the Eastern Mediterranean by comparing the performance of four ML algorithms, i.e. extreme gradient boosting (XGB), random forest (RF), support vector machine (SVM) and artificial neural network (ANN), and hybridizing the strongest-performing algorithm with the other algorithms. In the Hrysoon river basin in western Syria, 2100 flood events with twenty driving factors were precisely identified to achieve the aim of this investigation. The performance of each algorithm was assessed using various error indicators, including the Area Under the Receiver Operating Characteristic (ROC) Curve (AUC). The results showed that the XGB (AUC = 0.995) algorithm achieved the strongest performance compared to the RF (AUC = 0.991), ANN (AUC = 0.983) and SVM (AUC = 0.979). Regarding the hybridization process, the results revealed that the XGB-SVM performance was the strongest (AUC = 0.995), followed by XGB-ANN and XGB-RF with an AUC value of 0.994. The current assessment illustrated that the distance to river is the most influential among conditioning factors followed by aspect, elevation, slope and rainfall. Overall, this study provided objective and constructive outputs that improved the accuracy of improving flood vulnerability in this region. These outputs enable the establishment of sustainable land management procedures in the Eastern Mediterranean and in Syria, especially in the post-war phase.

Citation

Abdo, H. G., Richi, S. M., Alqadhi, S., Zeng, T., Prasad, P., Kotaridis, I., Alharbi, M. M., Khaddour, L. A., & Mallick, J. (online). A hybrid machine learning modelling for optimization of flood susceptibility mapping in the eastern Mediterranean. Natural Hazards, https://doi.org/10.1007/s11069-024-07081-3

• Journal Article Type: Article
• Acceptance Date: Dec 7, 2024
• Online Publication Date: Dec 23, 2024
• Deposit Date: Jan 3, 2025
• Journal: Natural Hazards
• Print ISSN: 0921-030X
• Electronic ISSN: 1573-0840
• Publisher: Springer
• Peer Reviewed: Peer Reviewed
• DOI: https://doi.org/10.1007/s11069-024-07081-3
• Keywords: Flood susceptibility, Hybrid machine learning, Risk assessment, Eastern Mediterranean, Syria