基于PEST的HSPF水文模型多目标自动校准研究

参数校准是流域水文模型构建的关键步骤之一。针对HSPF水文模型，系统梳理并形成PEST-HSPF多目标自动校准的输入、输出、交互、迭代和终止等流程，以滇池流域中和子流域为研究对象（数据为1999—2010 年日流量），与人工校准（HSPEXP）和单目标校准做对比分析。结果表明：①提出的以日流量、月流量和流量频率的偏差最小化为目标函数、各目标初始偏差的倒数为权重的校准思路，可实现相对人工、单目标更好的模拟精度、变化趋势和一致性；②人工校准擅长于把握径流总量和年际变化趋势，但对于日径流过程（精度和变异性）改进效果有限；③多目标自动校准不仅平衡了3 类目标函数的偏差贡献，还能一定程度上继承各目标函数的优势，在捕捉日月总径流量、整体变化趋势、局部变异性、不同保证率特征值上表现出可靠的特征。

PEST-based Multi-objective Automatic Calibration of Hydrologic Parameters for HSPF Model

Parameter calibration is one of the most important steps in hydrological modeling. A PEST-based multi-objective automatic calibration approach (PEST-HSPF) is proposed specially for Hydrological Simulation Program- Fortran (HSPF) model. The proposed approach consists of five steps: 1) the development of HSPF model framework including wdm and uci files, 2) the generation of time-series preprocess file (tsproc.dat) to transform the HSPF's output to ASCII format, 3) the preparison of parameter file (model.tpl) to adjust model parameter values of HSPF after repeating interations, 4) the setup of PEST's coeffients to generate model.ins and model.pst, and 5) the running of PEST to determine the optimal set of parameters when the discrepancies between the pertinent model-generated numbers and the corresponding measurements are reduced to minimum. Additionally, squared error of daily flows, monthly flows and exceedence times for flow (1%, 5%, 10%, 25%, 50%, 75% and 90%) are suggested to be objective functions, and the weighting functions are assigned to each subobjective function to ensure that the contributions of each to the multiple-objective functions were almost equal. A realworld case study is then conducted for Zhonghe Subwatershed of Lake Dianchi Watershed, which is applied to illustrate its advantages in predictive accuracy over manual and single-objective methods. The results indicate that: 1) ten types of sensitive parameters were automatically calibrated after 734 running of PEST-HSPF, leading to the significant decrease of the error of objective functions to be 45.4% of that at initial stage; therefore, PEST- HSPF results using squared error of daily flows, monthly volumes and flow exceedances weighted by reciprocal of their initial values performed better with respect to goodness-of-fit measures than manual and single-objective results, for example, the R², NSE and index of agreement of daily flows calibration were 0.75, 0.74 and 0.93. 2) Manual method assisted by HSPEXP satisfied total volumes and yearly trend, but cannot capture the small- scale hydrological processes; for example, the R², NSE and index of agreement of daily flows calibration were only 0.45, 0.24 and 0.80; 3) PEST-HSPF with multi-objective functions could not only balance the contribution for three objective functions, but also accurately acquire daily/monthly volumes, temporal trend, regional variations, and specific flows at different exceedance fractions over calibration and verification periods. Although the PEST-HSPF with single-objective function was superior to manual method assisted by HSPEXP, it just focused on the minimization of one type of errors within calibration process. For example, the R², NSE and index of agreement of the PEST-HSPF only with exceedence times for flow were only 0.61, 0.54 and 0.80. However, further research is necessary to understand how to select the parameters, objective functions and the associated weights, since there is currently no guidance available for the use of PEST-HSPF in more applications in watershed modeling. Considering the positive performance reported in this research and the many possible applications of this automatic calibration method, PEST-HSPF offers a new frontier for improving the field of watershed modeling.