Analysis of floods in the Eastern Black Sea Region between 1950-2020 with machine learning and other methods

Flood forecasting is frequently studied in the literature using various statistical and hydrological modeling methods. However, predicting floods in advance is inherently challenging. The primary reason for this difficulty is the presence of numerous parameters that influence flood events. In this study, the Eastern Black Sea region of Turkey, which experiences the highest number of flood disasters in the country, was selected as the study area. A detailed inventory analysis was conducted for 41 large-scale flood events in this basin, and the collected data were organized to create flood propagation maps. The study thoroughly examines the advantages and disadvantages of traditional and innovative flood forecasting methods. Particular attention was given to machine learning methods, focusing on generalization challenges such as selecting the most suitable method, optimal data quantity, validation techniques, and identifying the most effective flood-triggering parameters. Additionally, the study explored forecasting future flood events based on past occurrences and inter-basin flood prediction. Using machine learning techniques, a total of 11690 validation results were obtained for 2338 basins. Furthermore, the success rates of existing methods in addressing issues such as flood causes and pre-identification of flood-prone areas using machine learning were analyzed. The findings revealed that the M5P Rule Tree model performed significantly better in flood prediction compared to the Analytical Hierarchy Process (AHP) and Frequency Ratio (FR) models. Flood forecasts conducted with nine different machine learning methods demonstrated that the best results could be achieved through appropriate dataset selection, parameter identification, and validation methods. Additionally, the study found that these methods could effectively predict floods in basins with no prior data, providing promising results for future applications.