Autocalibration of HSPF for Simulation of Streamflow Using a Genetic Algorithm

Abstract

Hydrologic models are essential to watershed planning and management, particularly in the San Antonio River watershed where competition for scarce water resources is a challenge. As a result, the calibration and validation of hydrologic models are essential steps for their successful application. In this study, we examined the use of a loosely coupled genetic algorithm (GA) as an autocalibration tool for optimization of model parameters for the Hydrologic Simulation Program-Fortran (HSPF), a model frequently used in surface hydrology and water quality modeling. The GA-HSPF model is a more objective and less time-consuming alternative to traditional trial-and-error methods. The objective function was optimized by minimizing the mean absolute error (MAE) between corresponding simulated and observed average daily streamflow in the San Antonio River watershed. The MAE was used to evaluate the fitness of the parameter set in the GA. The calibrated model parameters (LZSN, INFILT, AGWRC, UZSN, DEEPFR, LZETP, and INTFW) were selected based on a sensitivity analysis from a previous study. Goodness-of-fit of the GA calibrated model was evaluated using the Nash-Sutcliffe coefficient of efficiency, MAE, root mean square error, flow duration curves, wavelet analysis, and total volume error. Overall simulation time with 2000 model simulations was 11 days, which can be improved significantly under parallel computing, as GA-HSPF simulations are highly independent. The objective function ceased improvement after approximately 250 simulations, with a minimized MAE of 25.8 m 3 /s. With the exception of DEEPFR, all optimized model parameter values were within the range cited in the literature. Nash-Sutcliffe coefficients in all simulations were above 0.5, suggesting that the simulated flows were in good agreement with the observed flows. Visual comparison between observed and simulated stream flow using time series and flow duration curves showed that the GA calibrated model was unable to simulate peak flow events accurately, particularly in the 0% to 10% exceedence range. It is hypothesized that the storage-based routing scheme in HSPF limits its ability to predict peak flows in this watershed. Comparison between observed and simulated flows in the wavelet domain indicated that the GA calibrated model was not able to preserve the scale and location of some high frequencies, but the scale and location of lower frequencies were preserved. The cyclic nature of the streamflow in this watershed was more prominent in lower frequencies. While overall flow rates were adequately predicted using a GA-HSPF approach, future work in this watershed needs to focus on multi-objective optimization that optimizes both volumes and peak flows. The GA-HSPF model offers an objective and efficient method for calibration and validation, a useful tool in watershed planning efforts.

Automatic Calibration Tool for Hydrologic Simulation Program-FORTRAN Using a Shuffled Complex Evolution Algorithm

“…Calibration is the process of adjusting selected model parameters to minimize the difference between the simulated and observed variables of interest [1,2]. Parameter calibration is necessary when using spatially-lumped hydrologic models like the Hydrological Simulation Program-FORTRAN (HSPF) [3,4]. Model calibration may be performed manually, or the processes can be automated using an optimization algorithm [5,6].…”

“…In routine HSPF applications, the model is typically manually calibrated with initial parameter estimates and thoughtful adjustments [4,5,12,13]. With HSPF, manual calibration assistance is provided by decision support software, the Expert System for the Calibration of HSPF (HSPEXP), which has been developed to provide guidance for parameter adjustment [1].…”

“…Iskra and Droste [14] found that the random multiple search method (RSM) and the Shuffled Complex Evolution method (SCE-UA) could find a parameter set providing better model performance statistics than with PEST employing the GLM algorithm. Sahoo et al [4] calibrated the hydrologic components of HSPF using a generic algorithm (GA), but it has been suggested that the GA required greater computing resources and time for parameter calibration than SCE-UA, making running the model less efficient [18][19][20][21][22].…”

“…Previous studies examining auto-calibration for hydrologic models showed that the auto-calibration method did not always lead to successful calibration in terms of solution robustness and computational efficiency due to the limitations of the algorithm used [4,5,14,36]. In this research, we have linked the HSPF model with the SCE-UA algorithm in a parallel computing framework supported by R (HSPF-SCE) with the purpose of providing an alternative and efficient tool for automated parameter calibration of HSPF.…”

Optimization of Calibration Parameters for an Event Based Watershed Model Using Genetic Algorithm

“…They achieved a sufficiently accurate Pareto set and a good diversity in the obtained front with the use of Non-dominated Sorting Differential Evolution (NSDE) for the model calibration. Sahoo et al (2010) examined the use of a loosely coupled GA as an auto calibration tool for optimization of model parameters for the Hydrologic Simulation Program-Fortran (HSPF). The objective function was optimized by minimizing the mean absolute error between corresponding observed and simulated average daily stream flow in the San Antonio River watershed.…”