Flood simulation in prairie agricultural regions is challenging due to low-relief topography, extensive surface storage, and spatially heterogeneous rainfall-runoff response. This study presents an integrated high-resolution flood modeling framework that combines GIS-based spatial analysis, flood frequency analysis of the observed annual peak flow record, the physically based HEC-HMS hydrological model, and a data-driven Adaptive Neuro-Fuzzy Inference System (ANFIS) optimized by the Imperialist Competitive Algorithm (ICA). The ANFIS-ICA model was developed as a computationally efficient surrogate to emulate HEC-HMS-simulated hydrographs at 3-minute temporal resolution under 10-, 50-, and 100-year return periods. The framework was applied to 16 sub-basins in Maple Creek, a representative prairie region in Saskatchewan, Canada. Based on spatial datasets processed in QGIS, the HEC-HMS model was first calibrated against observed peak discharges from selected historical flood events, and then used to generate training data for the ANFIS-ICA model incorporating rainfall and lagged flow inputs. Model evaluation showed excellent agreement between simulated and reference hydrographs, with NSE and R2 values exceeding 0.95 and RMSE below 0.02 m3/s across training, validation, and testing datasets, indicating strong generalization and high surrogate fidelity. Hydrograph analysis revealed spatially variable runoff behavior, characterized by relatively rapid flow propagation in sub-basins with limited surface storage and attenuated responses in sub-basins exhibiting stronger surface storage effects, reflecting the heterogeneous hydrological nature of prairie environments. Peak flows were classified into flood risk categories, enabling sub-basin-level vulnerability mapping to inform targeted flood mitigation, early warning, and adaptive agricultural management. The proposed framework provides a validated and potentially transferable tool for efficient, localized flood risk assessment in data-sparse prairie agroecosystems. However, its performance remains dependent on the accuracy of HEC-HMS simulations and the availability of observed peak flow data, and full hydrograph validation is constrained by the lack of high-temporal-resolution discharge records. Graphical Abstract The graphical abstract visually summarizes the integrated workflow developed in this study for flood modeling in prairie agroecosystems. The top panel presents the data collection stage, where spatial datasets including a high-resolution DEM and land-use layers for the Maple Creek were combined with observed hydrometric data and design storms for the 10-, 50-, and 100-year return periods. In the methodology panel, QGIS was employed to classify land cover and derive hydrological parameters to support HEC-HMS watershed setup across 16 sub-basins. The physically based HEC-HMS model produced hydrographs that served as both reference flood responses and input data for training the ANFIS optimized by the ICA. The flood frequency analysis based on Weibull plotting positions and LP3 distribution fitting, enabled design peak flow estimation under different return periods. The results panel highlights model performance evaluation, where ANFIS-ICA accurately emulated the calibrated HEC-HMS hydrographs, demonstrating high surrogate fidelity and substantial computational efficiency. Hydrograph comparisons illustrate characteristic prairie hydrological behavior, including rapid rising limbs and storage-controlled runoff response. The final output depicts flood vulnerability classification maps for each sub-basin and return period, supporting spatially explicit agricultural flood risk assessment and improved flood management in prairie regions.

Hybrid AI-Hydrologic Flood Modeling in Prairie Agricultural Watersheds

期刊论文
发表日期：

2026年4月20日

查看全文

Amin Hassanjabbar,
Xin Zhou,
Todd Han,
Kevin McCullum,
Peng Wu

摘要

Flood simulation in prairie agricultural regions is challenging due to low-relief topography, extensive surface storage, and spatially heterogeneous rainfall-runoff response. This study presents an integrated high-resolution flood modeling framework that combines GIS-based spatial analysis, flood frequency analysis of the observed annual peak flow record, the physically based HEC-HMS hydrological model, and a data-driven Adaptive Neuro-Fuzzy Inference System (ANFIS) optimized by the Imperialist Competitive Algorithm (ICA). The ANFIS-ICA model was developed as a computationally efficient surrogate to emulate HEC-HMS-simulated hydrographs at 3-minute temporal resolution under 10-, 50-, and 100-year return periods. The framework was applied to 16 sub-basins in Maple Creek, a representative prairie region in Saskatchewan, Canada. Based on spatial datasets processed in QGIS, the HEC-HMS model was first calibrated against observed peak discharges from selected historical flood events, and then used to generate training data for the ANFIS-ICA model incorporating rainfall and lagged flow inputs. Model evaluation showed excellent agreement between simulated and reference hydrographs, with NSE and R² values exceeding 0.95 and RMSE below 0.02 m³/s across training, validation, and testing datasets, indicating strong generalization and high surrogate fidelity. Hydrograph analysis revealed spatially variable runoff behavior, characterized by relatively rapid flow propagation in sub-basins with limited surface storage and attenuated responses in sub-basins exhibiting stronger surface storage effects, reflecting the heterogeneous hydrological nature of prairie environments. Peak flows were classified into flood risk categories, enabling sub-basin-level vulnerability mapping to inform targeted flood mitigation, early warning, and adaptive agricultural management. The proposed framework provides a validated and potentially transferable tool for efficient, localized flood risk assessment in data-sparse prairie agroecosystems. However, its performance remains dependent on the accuracy of HEC-HMS simulations and the availability of observed peak flow data, and full hydrograph validation is constrained by the lack of high-temporal-resolution discharge records.

Graphical Abstract

The graphical abstract visually summarizes the integrated workflow developed in this study for flood modeling in prairie agroecosystems. The top panel presents the data collection stage, where spatial datasets including a high-resolution DEM and land-use layers for the Maple Creek were combined with observed hydrometric data and design storms for the 10-, 50-, and 100-year return periods. In the methodology panel, QGIS was employed to classify land cover and derive hydrological parameters to support HEC-HMS watershed setup across 16 sub-basins. The physically based HEC-HMS model produced hydrographs that served as both reference flood responses and input data for training the ANFIS optimized by the ICA. The flood frequency analysis based on Weibull plotting positions and LP3 distribution fitting, enabled design peak flow estimation under different return periods. The results panel highlights model performance evaluation, where ANFIS-ICA accurately emulated the calibrated HEC-HMS hydrographs, demonstrating high surrogate fidelity and substantial computational efficiency. Hydrograph comparisons illustrate characteristic prairie hydrological behavior, including rapid rising limbs and storage-controlled runoff response. The final output depicts flood vulnerability classification maps for each sub-basin and return period, supporting spatially explicit agricultural flood risk assessment and improved flood management in prairie regions.