基于GR4J与LSTM耦合的清江流域径流模拟对比分析

Comparative analysis of runoff simulation in the Qingjiang River basin based on the coupling of GR4J and LSTM

摘要

为评估传统水文模型与数据驱动模型在山区流域中的径流模拟性能，本文以清江流域为研究区域，基于2014-2023年逐日降水与径流资料，对传统水文模型 (GR4J) 与长短时记忆网络模型 (LSTM) 的性能进行比较；在此基础上构建两类融合物理机制与数据驱动的混合模型 (松耦合模型 Hybrid1 将 GR4J 模拟径流作为 LSTM 输入特征，属于单向信息融合；紧耦合模型 Hybrid2 在 LSTM 输出径流的同时，进一步生成 GR4J 模型参数，并将其反馈至 GR4J 模型，形成“LSTM-参数-GR4J”的闭环交互结构)，并分析混合建模策略对山区径流模拟性能的提升效果。结果表明：在观测资料受限条件下，采用单一年份资料进行模型率定与训练，LSTM 模型整体表现优于 GR4J；采用丰枯年组合资料 (2017 年丰水年与 2019 年枯水年) 进行模型率定与训练，GR4J 与 LSTM 在验证期内的纳什效率系数 (NSE) 分别为 0.742 和 0.790，Kling-Gupta 效率系数 (KGE) 分别为 0.543 和 0.771，显著优于采用单一年份资料的结果，且模拟性能接近于使用较长时间序列资料 (2014-2019 年) 的水平。混合建模策略进一步提升了径流模拟精度，其中 Hybrid2 整体优于 Hybrid1。Hybrid2 在验证期 (2020-2023 年) 的 NSE 和 KGE 分别为 0.795 和 0.771，较 GR4J 模型分别提高 10.7% 和 52.7%，较 LSTM 模型分别提高 3.7% 和 14.7%。此外，各模型径流模拟性能呈现明显的季节性差异，在夏季及汛期表现较好，其中 Hybrid2 在汛期的 NSE 和 KGE 均超过 0.8，具有良好的洪水期径流模拟能力。因此，采用单一年份、丰枯年组合水文资料训练模型具有可行性，Hybrid2 能有效提升山区径流模拟精度，可为山区流域径流模拟预报提供参考。

Abstract

To evaluate the runoff simulation performance of traditional hydrological and data-driven models in mountainous basins, this study takes the Qingjiang River basin as the research area and compares the performance of the conceptual hydrological model (GR4J) and the Long Short-Term Memory network (LSTM) based on the daily precipitation and runoff data from 2014 to 2023. On this basis, two hybrid models integrating physical mechanisms and data-driven approaches are developed: a loosely coupled model (Hybrid1) and a tightly coupled model (Hybrid2). In Hybrid1, the runoff simulated by GR4J is used as an input feature for LSTM, representing a one-way information fusion strategy. By contrast, Hybrid2 not only outputs runoff through LSTM but also generates GR4J model parameters, which are subsequently fed back into GR4J, thereby forming a closed-loop interaction structure of "LSTM-parameters-GR4J". The improvements in simulation performance achieved by hybrid modeling strategies are further assessed. Results show that, under limited data conditions, when models are calibrated using single-year data, LSTM generally outperforms GR4J. When a combination of wet and dry years (2017 and 2019) is used for calibration, GR4J and LSTM achieve Nash-Sutcliffe efficiency (NSE) values of 0.742 and 0.790, and Kling-Gupta efficiency (KGE) values of 0.543 and 0.771, respectively, during the validation period. These results are significantly better than those obtained using single-year data and are comparable to those achieved with longer time series data (2014-2019). The hybrid modeling strategy further improves runoff simulation accuracy, with Hybrid2 outperforming Hybrid1 overall. In particular, Hybrid2 attains NSE and KGE values of 0.795 and 0.771, respectively, during the validation period (2020-2023), representing improvements of 10.7% and 52.7% over GR4J, and 3.7% and 14.7% over LSTM. In addition, model performance exhibits clear seasonal variability, with all models achieving optimal accuracy in summer and the flood season. Hybrid2 achieves NSE and KGE values exceeding 0.8 in the flood season, demonstrating strong capability for flood-period runoff simulation. Overall, the results demonstrate the feasibility of training models using single-year and wet-dry year combination data, and confirm that Hybrid2 effectively enhances runoff simulation accuracy in mountainous basins, providing a valuable reference for runoff simulation and flood forecasting in similar regions.

关键词

GR4J / LSTM / 物理机制与数据驱动耦合 / 山区径流模拟 / 评估

Key words

GR4J / LSTM / coupling of physical mechanisms and data-driven approaches / runoff simulation in mountainous basins / evaluation

参考文献

[1] 敖天其,刘臻,吴玲玲,等. 2025.分布式水文模型在高原山地流域的研究与应用进展[J].高原山地气象研究,45(3):14-23.

[2] Ao T Q, Liu Z, Wu L L, et al. 2025. Research and application progress of distributed hydrological models in plateau mountainous basins [J].Plateau and Mountain Meteorology Research,45(3):14-23 (in Chinese). doi:10.3969/j.issn.1674−2184.2025.03.002

[3] 包红军,林建,曹爽,等. 2020.基于流域地貌的中小河流致洪动态临界面雨量阈值研究[J].气象,46(11):1495-1507.

[4] Bao H J, Lin J, Cao S, et al. 2020. Topography-based dynamic critical arearainfall threshold for small to middle-sized river flood warning [J].Meteorological Monthly,46(11):1495-1507 (in Chinese). doi:10.7519/j.issn.1000−0526.2020.11.010

[5] 邓超,孙培源,尹鑫,等. 2015.基于概念性水文模型与长短时记忆模型耦合的汉江上游流域径流模拟[J].湖泊科学,37(1):279-292.

[6] Deng C, Sun P Y, Yin X, et al. 2015. Runoff simulation in the upper Han River Basin using physics-informed machine learningmodel [J].Journal of Lake Sciences,37(1):279-292 (in Chinese). doi:10.18307/2025.0143

[7] 甘衍军,徐晶,赵平,等. 2017.暴雨致洪预报系统及其评估[J].应用气象学报,28(4):385-398.

[8] Gan Y J, Xu J, Zhao P, et al. 2017. Introduction and evaluation of a rainstorm-caused flood forecasting system [J].Journal of Applied Meteorological Science,28(4):385-398 (in Chinese). doi:10.11898/1001-7313.20170401

[9] 高顺年,高铭,王振海. 2026.青海大通“8·18”山洪灾害的天气特点及其气象应急服务对策[J].暴雨灾害,45(1):142-148.

[10] Gao S N, Gao M, Wang Z H. 2026. Synoptic characteristics and corresponding meteorological emergency service strategies based on a mountain flood disaster event in Datong of Qinghai Province on August 18, 2022 [J].Torrential Rain and Disasters,45(1):142-148 (in Chinese). doi:10.12406/byzh.2024−140

[11] 郭良,丁留谦,孙东亚,等. 2018.中国山洪灾害防御关键技术[J].水利学报,49(9):1123-1136.

[12] Guo L, Ding L Q, Sun D Y, et al. 2018. Key techniques of flash flood disaster prevention in China [J].Journal of Hydraulic Engineering,49(9):1123-1136 (in Chinese). doi:10.13243/j.cnki.slxb.20180728

[13] 胡乐怡,蒋晓蕾,周嘉慧,等. 2014.考虑记忆时间的 LSTM 模型在赣江流域径流预报中的应用[J].湖泊科学,36(4):1241-1251.

[14] Hu L Y, Jiang X L, Zhou J H, et al. 2014. Application of LsTM considering time steps in runoff prediction of Ganjiang River Basin [J].Journal of Lake Sciences,36(4):1241-1251 (in Chinese). doi:10.18307/2024.0454

[15] 黄天文,焦飞,伍志方. 2024.一种基于迁移学习和长短期记忆神经网络的降水预报方法[J].暴雨灾害,43(1):45-53.

[16] Huang T W, Jiao F, Wu Z F. 2024. A precipitation forecast method based on transfer learning and Long Short Term Memory [J].Torrential Rainand Disasters,43(1):45-53 (in Chinese). doi:10.12406/byzh.2023−118

[17] 李德龙,程先云,杨浩,等. 2013.人工智群算法在水文模型参数优化率定中的应用研究[J].水利学报,44(z1):95-101.

[18] Li D L, Cheng X Y, Yang H, et al. 2013. Study on artificial inteifigence optimization algorithms forauto-calibration of hydrological models [J].Journal of Hydraulic Engineering,44(z1):95-101 (in Chinese). doi:10.3969/j.issn.0559−9350.2013.z1.014

[19] 李国英. 2022.加快建设数字孪生流域提升国家水安全保障能力[J].中国水利,(20):1.

[20] Li G Y. 2022. Accelerate to build the digital twin river basins and enhance China's capacity to ensure water security [J].China Water Resources, (20):1 (in Chinese). doi:10.16616/j.cnki.11-4446/TV.2022.09.01

[21] 刘攀,郑雅莲,谢康,等. 2021.水文水资源领域深度学习研究进展综述[J].人民长江,52(10):76-83.

[22] Liu P, Zheng Y L, Xie K, et al. 2021. Review of research progress in deep learning for hydrology and water resources [J].Yangtze River,52(10):76-83 (in Chinese). doi:10.16232/j.cnki.1001−4179.2021.10.012

[23] 庞轶舒,张俊,秦宁生,等. 2022.长江上游夏季径流量年际增量预测模型及检验[J].应用气象学报,33(1):115-128.

[24] Pang Y S, Zhang J, Qin N S, et al. 2022. Forecast model of interannual increment for summer runoff and its verification in the upper reaches of the Yangtze River [J].Journal of Applied Meteorological Science,33(1):115-128 (in Chinese). doi:10.11898/1001−7313.20220110

[25] 涂勇,吴泽斌,何秉顺. 2020. 2011-2019 年全国山洪灾害事件特征分析[J].中国防汛抗旱,30(Z1):22-25.

[26] Tu Y, Wu Z B, He B S. 2020. Analysis on the characteristics of flash flood disasters in China from 2011 to 2019 [J].China Flood & Drought Management, 30(S1):22-25 (in Chinese). doi:10.16867/j.issn.1673−9264.2020239

[27] 夏军,石卫. 2016.变化环境下中国水安全问题研究与展望[J].水利学报,47(3):292-301.

[28] Xia J, Shi W. 2016. Perspective on water security issue of changing environment in China [J].Journal of Hydraulic Engineering,47(3):292-301 (in Chinese). doi:10.13243/j.cnki.slxb.20150937

[29] 许钦,金晨,张坤,等. 2025.耦合深度学习与水文模型的喀斯特地区径流模拟方法[J].水科学进展,36(4):634-645.

[30] Xu Q, Jin C, Zhang K, et al. 2025. Runoff simulation in Karst regions by integrating deep learning with physically-basedhydrological models [J].Advances in Water Science,36(4):634-645 (in Chinese). doi:10.14042/j.cnki.32.1309.2025.04.008

[31] 雍斌,张建云,王国庆. 2023.黄河源区水文预报的关键科学问题[J].水科学进展,34(2):159-171.

[32] Yong B, Zhang J Y, Wang G Q. 2023. Key scientific issues of hydrological forecast in the headwater area of Yellow River [J].Advances in Water Science, 34(2):159-171 (in Chinese). doi:10.14042/j.cnki.32.1309.2023.02.001

[33] 张国涛,崔鹏,张晨笛. 2025.中国山洪灾害防控理论与技术: 进展、挑战与对策[J].地理科学进展,44(7):1315-1333.

[34] Zhang G T, Cui P, Zhang C D. 2025. Theory and technology of flash flood disaster prevention and control in China: Progress, challenges, and counter measures [J].Progress in Geography,44(7):1315-1333 (in Chinese). doi:10.18306/dlkxjz.2025.07.001

[35] Cho K, Kim Y. 2022. Improving streamflow prediction in the WRF-Hydro model with LSTM networks [J].Journal of Hydrology,605:127297. doi:10.1016/j.jhydrol.2021.127297

[36] Devia G K, Ganasri B P, Dwarakish G S. 2015. A review on hydrological models [J].Aquatic Procedia,4:1001-1007. doi:10.1016/j.aqpro.2015.02.126

[37] Gao S, Huang Y, Zhang S, et al. 2020. Short-term runoff prediction with GRU and LSTM networks without requiring time step optimization during sample generation [J].Journal of Hydrology,589:125188. doi:10.1016/j.jhydrol.2020.125188

[38] Gauch M, Mai J, Lin J. 2021. The proper care and feeding of CAMELS: How limited training data affects streamflow prediction [J].Environmental Modelling & Software,135:104926. doi:10.1016/j.envsoft.2020.104926

[39] Hochreiter S, Schmidhuber J. 1997. Long short-term memory [J].Neural Computation,9(8):1735-1780. doi:10.1162/neco.1997.9.8.1735

[40] Karpatne A, Atluri G, Faghmous J H, et al. 2017. Theory-guided data science: A new paradigm for scientific discovery from data [J].IEEE Transactions on Knowledge and Data Engineering,29(10):2318-2331. doi:10.1109/TKDE.2017.2720168

[41] Kraft B, Jung M, Körner M, et al. 2022. Towards hybrid modeling of the global hydrological cycle [J].Hydrology and Earth System Sciences,26(6):1579-1614. doi:10.5194/hess−26−1579−2022

[42] Lin G F, Chou Y C, Wu M C. 2013. Typhoon flood forecasting using integrated two-stage Support Vector Machine approach [J].Journal of Hydrology,486:334-342. doi:10.1016/j.jhydrol.2013.02.012

[43] Lu D, Konapala G, Painter S L, et al. 2021. Streamflow simulation in data-scarce basins using bayesian and physics-informed machine learning models [J].Journal of Hydrometeorology,22(6):1421-1438. doi:10.1175/JHM−D−20−0082.1

[44] Moriasi D, Arnold J, Van Liew M, et al. 2007. Model evaluation guidelines for systematic quantification of accuracy in watershed simulations [J].Transactions of the ASABE, 50. doi:10.13031/2013.23153

[45] Noori N, Kalin L. 2016. Coupling SWAT and ANN models for enhanced daily streamflow prediction [J].Journal of Hydrology,533:141-151. doi:10.1016/j.jhydrol.2015.11.050

[46] Nourani V. 2017. An Emotional ANN (EANN) approach to modeling rainfall-runoff process [J].Journal of Hydrology,544:267-277. doi:10.1016/j.jhydrol.2016.11.033

[47] Okkan U, Ersoy Z B, Ali Kumanlioglu A, et al. 2021. Embedding machine learning techniques into a conceptual model to improve monthly runoff simulation: A nested hybrid rainfall-runoff modeling [J].Journal of Hydrology,598:126433. doi:10.1016/j.jhydrol.2021.126433

[48] Perrin C, Michel C, Andréassian V. 2003. Improvement of a parsimonious model for streamflow simulation [J].Journal of Hydrology,279(1):275-289. doi:10.1016/S0022−1694(03)00225−7

[49] Perrin C, Oudin L, Andreassian V, et al. 2007. Impact of limited streamflow data on the efficiency and the parameters of rainfall-runoff models [J].Hydrological Sciences Journal,52(1):131-151. doi:10.1623/hysj.52.1.131

[50] Sorooshian S, Gupta V K, Fulton J L. 1983. Evaluation of Maximum Likelihood Parameter estimation techniques for conceptual rainfall-runoff models: Influence of calibration data variability and length on model credibility [J].Water Resources Research,19(1):251-259. doi:10.1029/WR019i001p00251

[51] Sun W, Wang Y, Wang G, et al. 2017. Physically based distributed hydrological model calibration based on a short period of streamflow data: case studies in four Chinese basins [J].Hydrology and Earth System Sciences,21(1):251-265. doi:10.5194/hess−21−251−2017

[52] Wang K H, Altunkaynak A. 2012. Comparative case study of rainfall-runoff modeling between SWMM and fuzzy logic approach [J].Journal of Hydrologic Engineering,17(2):283-291. doi:10.1061/(ASCE)HE.1943−5584.0000419

[53] Xiang Y, Li L, Chen J, et al. 2019. Parameter uncertainty of a snowmelt runoff model and its impact on future projections of snowmelt runoff in a data-scarce deglaciating river basin [J].Water,11(11):2417. doi:10.3390/w11112417

[54] Xiang Z, Yan J, Demir I. 2020. A rainfall-runoff model with LSTM-based sequence-to-sequence Learning [J].Water Resources Research,56(1):e2019WR025326. doi:10.1029/2019WR025326

[55] Yaseen Z M, El-shafie A, Jaafar O, et al. 2015. Artificial intelligence based models for stream-flow forecasting: 2000-2015 [J].Journal of Hydrology,530:829-844. doi:10.1016/j.jhydrol.2015.10.038

[56] Yu Q, Jiang L, Wang Y, et al. 2023. Enhancing streamflow simulation using hybridized machine learning models in a semi-arid basin of the Chinese loess Plateau [J].Journal of Hydrology,617:129115. doi:10.1016/j.jhydrol.2023.129115

基金资助

国家自然科学基金项目(U2340201)

广西重点研发计划项目(AB25069132)

全国暴雨研究开放基金项目(BYKJ2025Q34，BYKJ2025M27，BYKJ2024Z11)

中国气象局创新发展专项(CXFZ2026J133)