Sequential Optimization of XGBoost through Nature-Inspired Algorithms for Flash Flood Modelling in Beas Sub-Basin, Western Himalayas

Indian Himalayan Region (IHR) is highly vulnerable to flash floods; therefore, this study aimed to predict flash flood susceptibility in the Beas River Sub-Basin (BRSB) of IHR through sequential optimization of XGBoost using four metaheuristic algorithms. Among these, XGBoost-AAEO achieved the highest predictive performance with an accuracy of 86.2% and an AUC of 0.93, outperforming the standalone XGBoost (accuracy=76.7%, AUC=0.849), thereby confirming the effectiveness of the proposed optimization framework. Results show that approximately 2061 km2 (16.37%) of BRSB is high and very high susceptible. In the Western BRSB, high flash flood-prone zones include Dharamshala, McLeodganj and Kangra; in middle sub-basin, Palampur, Baijnath, and Mandi; and in the Eastern sub-basin, Manali, Bajaura, Aut, Kasol, and Manikaran. SHAP analysis identified drainage density, distance to river, slope, and rainfall as the most significant flash flood drivers. The study provides valuable insights for formulating flood mitigation strategies and improving resource management across BRSB.