Hierarchical calibration and intelligent optimization of the Xin'anjiang model for flood simulation and forecasting

Flood simulation and forecasting are fundamental to effective watershed management, particularly under the combined influences of climate change and intensified human activities. Conventional calibration strategies for hydrological models often struggle to maintain stable performance across floods of different magnitudes due to scale dependent differences in runoff generation and routing processes. To address this limitation, this study proposes a hierarchical calibration framework that explicitly links model parameter estimation to flood magnitude using segmented net rainfall thresholds. Based on observed flood events in the Yuecheng Catchment, thresholds of 25 mm, 13.5 mm, and 12.5 mm were identified for small, medium, and large floods, respectively. DE, SA, and WOA were employed to calibrate the parameters of the Xin'anjiang model within this hierarchical framework, and their impacts on flood simulation performance were systematically evaluated. The results demonstrate that the proposed hierarchical calibration framework substantially improves model adaptability and simulation accuracy across different flood magnitudes. Mean NSE values generally exceeded 0.94, while mean MARE remained below 15%. Compared with conventional uniform parameter calibration, the proposed approach significantly reduced peak discharge errors for large floods (from 6.34% to 3.20%) and decreased peak timing errors (from 0.71 h to 0.14 h). Among the optimization methods, DE exhibited the most balanced and stable overall performance, SA showed advantages in maintaining runoff volume consistency, particularly for medium floods, and WOA achieved satisfactory peak magnitude simulations but remained sensitive to timing deviations. The proposed framework provides a transferable reference for improving hydrological simulation and flood forecasting in mountainous catchments under changing climatic and environmental conditions.