An integrated strategy for evaluating flood susceptibility combining deep neural networks with biologically inspired meta-heuristic algorithms

Predicting the likelihood of future flooding remains a challenging problem due to diverse hydrological conditions and heightened flood vulnerability. Previous flood susceptibility mapping efforts have often been constrained by limited predictive capabilities and a lack of integration with advanced computational methods. This study developed a flood susceptibility map (FSM) for Xinxiang City, Henan Province, China, improving predictive accuracy by incorporating a long short-term memory network (LSTM) with three biologically inspired meta-heuristic algorithms: the Whale Optimization Algorithm (WOA), the Northern Goshawk Algorithm (NGO), and Snake Optimization (SO). A flood list map containing 300 flood locations was established through the construction of a spatial flood database incorporating 12 explanatory factors for flooding. The relationship between these factors and flood occurrences was examined using the variance inflation factor (VIF), random forest (RF), and frequency ratio (FR) methods. The effectiveness and predictive capabilities of these models were compared and validated using statistical techniques, the receiver operating characteristics (ROC) curve, and the area under the ROC curve (AUC). The optimized WOA-LSTM and SO-LSTM models outperformed others, achieving a kappa coefficient of 0.966 and an AUC value close to 1, indicating superior prediction accuracy and stability. The model effectively categorized risk regions into six levels, facilitating flood risk management in geologically similar areas. This research contributes to the field by demonstrating the effectiveness of combining LSTM with meta-heuristic algorithms for enhanced flood susceptibility prediction.