基于SWAT-TGRU耦合模型的伊洛河流域径流模拟研究

Runoff simulation in Yiluo River Basin based on SWAT-TGRU coupled model

摘要

【目的】考虑流域地理空间异质性及多站点径流信息在上、下游的传递，对于提升流域径流模拟准确性至关重要。【方法】构建了基于SWAT与树状门控循环单元(TreeGRU)的耦合径流模拟模型，通过SWAT模型模拟流域内各子流域的水文物理过程，获取降水、蒸散发、地表径流等物理变量输出，并将这些变量及上游水文站的模拟径流数据作为输入，从而构建SWAT-TGRU耦合模型。模型利用树状结构实现信息从上游到下游的传递，充分考虑河流网络中水文站的时空关系。【结果】结果表明，SWAT-TGRU模型在伊洛河流域的潭头、龙门镇、卢氏和黑石关四个水文站的径流模拟中，纳什效率系数(NSE)分别较SWAT模型提升49%、11%、13%和15%。【结论】与SWAT-GRU和SWAT-TLSTM模型相比，SWAT-TGRU模型在多站点径流模拟中表现出更高的精度和稳定性，有效避免了未采用树状结构时模型复杂度显著上升的劣势，为复杂流域多站点径流模拟提供了一种高效、准确的新方法。

Abstract

[Objective] Considering the spatial heterogeneity of river basins and the transmission of multi-station runoff information between upstream and downstream is essential for improving runoff simulation accuracy in river basins. A coupled runoff simulation model based on the Soil and Water Assessment Tool(SWAT) and Tree-Structured Gated Recurrent Unit(TreeGRU) was established. [Methods] The SWAT model was employed to simulate the hydrological and physical processes in sub-basins to obtain outputs of key physical variables such as precipitation, evapotranspiration, and surface runoff. These variables, along with simulated runoff data from upstream hydrological stations, were used as inputs to establish the SWAT-TGRU coupled model. By incorporating a tree-structured framework, the model enabled the transmission of hydrological information from upstream to downstream, while fully accounting for the spatiotemporal relationships among hydrological stations in the river network. [Results] The SWAT-TGRU model improved the Nash-Sutcliffe Efficiency(NSE) at four hydrological stations of Tantou, Longmenzhen, Lushi, and Heishiguan by 49%, 11%, 13%, and 15%, respectively, in the Yiluo River Basin, compared to the SWAT model. [Conclusion] Compared with SWAT-GRU and SWAT-TLSTM models, the SWAT-TGRU model exhibits higher accuracy and stability in multi-station runoff simulations. It effectively overcomes the significant increase in model complexity that arises when non-treed-structured approaches are not utilized, providing an efficient and accurate method for multi-station runoff simulation in complex river basins.

关键词

径流模拟 / SWAT模型 / 树状GRU / 伊洛河流域 / 多站点 / 影响因素

Key words

runoff simulation / SWAT Model / tree-structured GRU / Yiluo River Basin / multi-station / influencing factors

基金资助

国家自然科学基金项目(52109034)

唐仲英基金会项目(K4050723175)

中国博士后科学基金项目(2023M732891)

陕西省博士后科学基金项目(2023BSHYDZZ64)