Random forest and naïve Bayes approaches as tools for flash flood hazard susceptibility prediction, South Ras El-Zait, Gulf of Suez Coast, Egypt

• Published in: Arabian journal of geosciences Vol. 15; no. 3
• Main Author: Abu El-Magd, Sherif Ahmed
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.02.2022; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Bayesian analysis; Earth and Environmental Science; Earth science; Earth Sciences; Elevation; Flash flooding; Flash floods; Flood forecasting; Flood hazards; Flood mapping; Flood predictions; Flooding; Floods; Information processing; Learning algorithms; Machine learning; Original Paper; Performance prediction; Prediction models; Remote sensing; River discharge; Urbanization; Gulf of Suez; Egypt; Wadi El-Dib; Naïve Bayes; Random forest; Machine learning; Egypt; Flash flood model
• ISSN: 1866-7511; 1866-7538
• DOI: 10.1007/s12517-022-09531-3

Machine learning (ML) algorithms are reliable approaches to address incomplete datasets in existing studies. In this study, the ML algorithms naïve Bayes (NB) and random forest (RF) were used to generate a flash flood forecasting model in Wadi El-Dib on the Gulf of Suez Coast at the Eastern Desert of Egypt. A total of 1117 point locations of field data and remote sensing data were mapped to prepare a flood inventory map. The relationships between the flood controlling factors were assessed and evaluated based on the implemented approaches. Slope degree, distance from streams, topographic wetness index, and elevation are the most important controlling factors out of the input seven themes. The proposed prediction model for the identification of flooding and nonflooding areas achieved reliable accuracy for the implemented approaches according to the area under the curve. Results demonstrate that the flash flood model was able to simulate flooding and nonflooding areas with improved accuracy. The NB and RF models achieved predictive performance with an accuracy of 85% to 88%, respectively. The susceptibility map was classified into flooding zones and nonflooding zones, which might be helpful for urbanization planning and management. Our findings indicate that about 83% of the field data were plotted into susceptible flooding zones and that eastern areas with gentle slopes have high potential for flash floods. ML can extract and generate useful information, and related models could be applied in such studies and similar areas.