湟水流域SWAT模型构建及参数不确定性分析

SWAT Model Construction and Uncertainty Analysis on Its Parameters for the Huangshui River Basin

敏感性分析和不确定性分析是分布式水文模型参数校准和模型构建的先决条件。以位于青藏高原与黄土高原过渡地带的青海湟水流域为例,基于SWAT模型的拉丁超立方和单次单因子(LH-OAT)采样方法和SWAT-CUP程序的拉丁超立方采样方法进行模型参数敏感性分析,同时以SWAT-CUP的P因子和R因子进行模型不确定性分析,最终结合手动调参和自动率定算法,构建湟水流域分布式水文模型。结果表明:湟水流域日尺度模拟中,率定期模型平均确定系数为0.7,平均效率系数为0.68;验证期平均确定系数为0.65,平均效率系数为0.53,可以满足应用要求;PSO算法在水文模型率定中总体表现良好;基于SWAT模型和SWAT-CUP程序的单次率定用时分别为3.5min和2.2min,SWAT-CUP程序明显快于SWAT;分析结果还表明:模型不确定性和模型率定结果精度并不一致,但模型不确定性决定着模型验证期的结果精度,为确保验证期精度,必须降低模型不确定性;模型自动率定中,相同迭代次数下,SWAT模型对于流量较大的子流域率定效果较好。

Sensitivity analysis（SA） and uncertainty analysis（UA） are prerequisites for parameter calibration and dis tributed hydrologic model development. Taking the Huangshui River basin located in Qinghai Tibetan Plateau and Loess Plateau as a case, firstly, based on SWAT model and SWAT-CUP program, sensitivity analysis was carried out by combing Latin-Hypercube（LH） and One-Factor-At-a-Time（OAT ） sampling with Latin hypercube sampling; secondly, uncertainty analysis was conducted by using P-factor and R factor of SWAT -CUP program; finally, parameter calibration and distributed hydrologic model building for the Huangshui River basin were performed by the coupled method of manual and auto-calibration. The results showed that the average determination coefficient was 0.7 and the average Nash-Sutcliff was 0.68 for the calibration. The average determination coefficient was 0. 65 and the average Nash-Sutcliff was 0.53 for the validation, the result could meet the application requirements. PSO algorithm had a good performance in the model calibration. The time for per calibration was 3.5 min and 2.2 min for SWAT model and SWAT-CUP, respectively, obviously the SWAT CUP＇s calibration time was shorter than that of SWAT model. Uncertainty analysis results were not always accordance with the calibration result precision, but it determined the validation precision of the model. So for the validation of the model, the uncertainty of the model should be cut down. Under the same times of the model calibration, SWAT model had a better runoff simulation value for the larger subbasin than the smaller one.