Real-Time Flood Risk Assessment and Forecasting with Hybrid BiCNN and Deep Learning Models

The catchment viewpoint has historically been the preferred approach for flood estimation and management among water resources engineers, hydrologists, and statisticians. This perspective considers meteorology, geography, and geology of a specific catchment when assessing flood potential. Enhanced methodologies are necessary for improved flood predictions due to the limitations in breadth and accuracy of traditional methods. In response to these challenges, this research employs VMD to enhance the quality of the input data utilised for model training. A novel technique utilising BiCNN is proposed, integrating bilinear interactions into the CNN architecture to capture complex spatial and channel-wise correlations. The model attained an F1 Score of 96.92%, a recall of 92.64%, a precision of 94.23%, and an overall accuracy of 96.84% on previously unseen images, indicating exceptional performance. The flood risk assessment results validate that the proposed strategy is superior than other advanced methodologies. The findings suggest that the development of a reliable flood risk assessment grading system facilitates more precise and timely flood predictions. This advancement may result in improved flood management strategies and more informed decision-making during a crisis.