一种SWAT模型参数识别的改进方法

参数的敏感性分析和不确定性分析是分布式水文模型构建的先决条件。在辽河流域建立SWAT模型,利用SWAT-CUP中的SUFI-2算法进行参数的率定,在此基础上提出一种更为简便的参数识别方法。将研究区域辽河干流的主要支流分别进行参数识别,再将SWAT-CUP中率定的最佳参数的从TXINOUT文件中提取出来,分别覆盖到SWAT模型中各对应支流子流域的TXINOUT文件中,即可得到按主要支流经过参数识别后的SWAT模型,避免了SWAT-CUP调参工具涉及众多子流域导致参数识别过于复杂的问题。结果表明,辽河干流主要支流招苏台河、清河、柴河等子流域主要水文断面率定期的平均纳什效率系数分别为0.60、0.65、0.68,验证期分别为0.60、0.72、0.77,参数率定的结果相对于全局调参有较大的改进。采用本文提出的参数识别方法,可以解决SWAT-CUP全局调参时上下游断面结果难以同时匹配或伪匹配的问题,又可以避免分区参数识别时对不同子流域的土地利用类型、土壤类型以及坡度等参数的繁琐设定,同时降低了SWAT模型手动调参的复杂程度,可较好地应用到SWAT模型参数识别过程中。

An improved method for parameter identification of SWAT model

The sensitivity analysis and uncertainty analysis of model parameters are the preconditions for the construction of distributed hydrological model. In this study, a SWAT model was established for the Liao River Basin, and the parameters were calibrated via the SUFI-2 algorithm in SWAT-CUP. On this basis, a more convenient parameter identification method was proposed. The parameters of the main tributaries of Liao River were identified first, and then the best parameters calibrated from SWAT-CUP were applied to the TXINOUT files of each corresponding tributary sub-basin in the SWAT model. Thus a calibrated SWAT model was established with its parameters identified according to the main tributaries. This avoided the complexity of parameter identification in the original procedures of SWAT-CUP tool. The results showed that the average Nash coefficients at main hydrological stations in Zhaosutai River, Qing River, and Chai River were 0.60, 0.65, and 0.68 respectively in the calibration period, and 0.60, 0.72, and 0.77 respectively in the validation period. The results of parameter calibration improved much compared with the global parameter calibration. This parameter identification method can solve the problem that SWAT-CUP global calibration results cannot match both the upstream and downstream sections or the problem of pseudo match. It can avoid complicated parameter settings for land use type, soil texture, and slope of different sub-basins. At the same time, it can reduce the complexity of manual adjustment of SWAT model parameters. The method can be applied well to the process of SWAT model parameter identification.