Hybrid CNN-GRU Approach for Flood Prediction in Rushikulya River Basin, India

One of nature’s most destructive hazards, floods cause both human fatalities and property destruction. Numerous cities are influenced by the monsoon, therefore they frequently experience calamity. Early warning of a flood incident could help the public and the authorities plan both short-term and long-term preventive actions, prepare for evacuation and rescue efforts, and provide relief for flood victims. Without taking into account physical processes, data-driven models provide effective alternatives, but the nonstationarity that exists in observations restricts the uses of these models. As a result, a hybrid DL model for forecasting flood discharge based on the convolutional neural network–gated recurrent unit (CNN-GRU) is proposed in this paper. The CNN-GRU model performed better than CNN model, according to the results monthly flood discharge prediction when applied to the Rushikulya River of Odisha, India, with an RMSE of 5.2214 m3/s, R2 of 0.9651, and an NSE of 0.9617 m3/s for all the flood episodes throughout the testing period. Additionally, we draw the conclusion that there is a need to not only enhance technical aspects of flood forecasting but also to close the gap between hydro-meteorological model development, scientific research, and probabilistic ensemble forecasts used in real-world flood management, particularly through effective communication.