Objective This study aims to construct a runoff prediction model integrating physical mechanisms and data-driven advantages to enhance the physical interpretability and simulation robustness of flood prediction, thereby providing technical references for watershed flood early warning and disaster prevention and mitigation. Methods Taking the middle and upper reaches of the Tingjiang River Basin as the study area, the HEC-HMS hydrological model was employed to conduct refined simulations of individual heavy rainfall events. The spatial heterogeneity of the initial loss parameter (Ia) and its regulatory role in runoff generation mechanisms were systematically analyzed. Ia was then incorporated as a key physical parameter into the input layer of LSTM, establishing an Ia-LSTM coupled prediction model. Results (1) HEC-HMS-optimized Ia values effectively characterized the physical mechanisms of rainfall-runoff in the river basin. Significant spatial differences in Ia were observed across sub-basins during different individual events, and its values were jointly controlled by multiple synergistic factors. (2) The HEC-HMS physical hydrological model demonstrated reliable performance in peak capture and peak timing synchronization, while the LSTM data-driven model exhibited superior trend fitting capabilities for the overall runoff process, reflecting complementary strengths between physical and data models across different dimensions. (3) The Ia-LSTM model significantly enhanced the robustness of short- to medium-term rainfall-runoff prediction. Its performance degradation with extended lead times was substantially smaller than that of the LSTM and BOLSTM models. The NSE for a 12-hour lead time remained at 0.71, with an overall NSE of 0.849. Conclusion Embedding the physical parameter Ia of runoff generation and concentration into LSTM effectively enhances the capability of data-driven models to represent rainfall-runoff physical mechanisms. The Ia-LSTM model improves short- to medium-term prediction robustness while significantly mitigating performance degradation associated with lead time extension.
WangX K, YangP, SunT, et al. Study on division early warning of flash flood disaster caused by rainstorm in mountainous small watersheds[J]. Advanced Engineering Sciences, 2021,53(1):29-38.
LiuQ Y, ChengW M, SunD Y, et al. Distribution characteristics of historical mountain flood in China[J]. Journal of Geo-Information Science, 2017,19(12):1557-1566.
LiuY, ZhangJ Y, BaoZ X, et al. Study of floods in experimental watersheds of different geomorphic types in eastern monsoon region of China[J]. Journal of Hydroelectric Engineering, 2022,41(3):22-31.
YueJ Y, LiuZ, MaoQ N, et al. Research on flood simulation by integrating satellite and ground precipitation with CNN-LSTM model[J]. China Rural Water and Hydropower, 2025,67(7):52-59.
ZhangH G, GuoS L, WangC J, et al. Review and research of calibration techniques for conceptual hydrological model[J]. Engineering Journal of Wuhan University, 2004,48(3):18-22,26.
ZhangH G, GuoS L, LiC Q. Preliminary study on uncertainty of quantitative precipitation forecast in Three Gorges Reservoir area[J]. Yangtze River, 2006,52(4):39-41.
ZhaiX Y, GuoL, LiuR H, et al. Development and application of China flash flood hydrological model: case study in small and medium-sized catchments of Anhui Province[J]. Journal of Basic Science and Engineering, 2020,28(5):1018-1036.
ShenZ Y, DingY S, KongQ. Application study of coupling rainfall-runoff modeling and floodplain inundation mapping[J]. Journal of Geo-Information Science, 2021,23(8):1473-1483.
XuY H, WuQ, LiC Q, et al. Simulation of the flood process in the middle reaches of the Yellow River by a long-short term memory(LSTM)neuro network[J]. Journal of Beijing Normal University (Natural Science), 2020,56(3):387-393.
[23]
KratzertF, KlotzD, BrennerC, et al. Rainfall-runoff modelling using Long Short-Term Memory(LSTM) networks[J]. Hydrology and Earth System Sciences, 2018,22(11):6005-6022.
QiH X, PengT, ZhiX F, et al. Progress and reflection on application of deep learning techniques in flood forecasting[J]. Meteorological Monthly, 2025,51(4):446-459.
[30]
CuiZ, ZhouY L, GuoS L, et al. Effective improvement of multi-step-ahead flood forecasting accuracy through encoder-decoder with an exogenous input structure[J]. Journal of Hydrology, 2022,609:127764.
KangY, AiH R, PengR J, et al. Adaptive bandwidth flood interval forecasting based on error-corrected hybrid model[J]. Water Resources Protection, 2025,41(5):106-114,131.
[33]
YoungC C, LiuW C, WuM C. A physically based and machine learning hybrid approach for accurate rainfall-runoff modeling during extreme typhoon events[J]. Applied Soft Computing, 2017,53:205-216.
[34]
Ali KumanliogluA, FistikogluO. Performance enhan-cement of a conceptual hydrological model by integrating artificial intelligence[J]. Journal of Hydrologic Engineering, 2019,24(11):04019047.
[35]
TajikiM, SchoupsG, Hendricks FranssenH J, et al. Recursive Bayesian estimation of conceptual rainfall-runoff model errors in real-time prediction of streamflow[J]. Water Resources Research, 2020,56(2):e2019WR025237.
SongS K, WangH L, LangZ L, et al. Mountain flood simulation of small basin in Taihang Mountains using HEC-HMS model: a case study of Luluochuan River Basin[J]. Journal of Catastrophology, 2023,38(1):117-124.
LaiR W, LaiM H, SuY Q. Temporal and spatial differentiation of soil and water loss sensitivity in Changting County[J]. Ecological Science, 2021,40(6):116-124.
TianJ H, FanS Q, WangW C, et al. Comparison between FFMS model and HEC-HMS model applied in rainstorm and flood simulation[J]. Journal of Yangtze River Scientific Research Institute, 2021,38(3):59-65.
YuanL T. Prediction and early warning of mountain flood disasters in Xihe River Basin based on HEC-HMS hydrological model[D]. Dalian, Liaoning: Liaoning Normal University, 2019.
[48]
夏弘.深度学习和水动力学模型结合的中小河流洪水预报研究[D].济南:山东大学,2023.
[49]
XiaH. Study on flood forecasting of small and medium-sized rivers based on deep learning and hydrodynamic model[D]. Jinan: Shandong University, 2023.
XuD M, WangY Y, WangW C. Prediction of annual runoff using long short-term memory neural network model based on Bayesian optimization algorithm[J]. Water Resources and Power, 2022,40(12):42-46.
HuangY F, ZhangX, DengL K, et al. Runoff simulation and time-lag change analysis in the Han River Basin based on machine learning[J]. Water Resources Protection, 2024,40(6):173-180.