Evaluation of a Stepwise, Multiobjective, Multivariable Parameter Optimization Method for the APEX Model. Issue 4 (1st July 2014)

• Record Type: Journal Article
• Authors: Senaviratne, G.M.M.M. Anomaa; Udawatta, Ranjith P.; Baffaut, Claire; Anderson, Stephen H.
• Is Part Of: Journal of Environmental Quality. Volume 43:Issue 4(2014)
• Page Start: 1381
• Page End: 1391
• Publication Date: 2014-07-01
• Subjects: Agricultural ecology -- Periodicals; Environmental engineering -- Periodicals; Pollution -- Periodicals; 630
• DOI: 10.2134/jeq2013.12.0509
• Languages: English
• ISSNs: 0047-2425

Abstract

Abstract : Hydrologic models are essential tools for environmental assessment of agricultural nonpoint‐source pollution. The automatic calibration of hydrologic models, though efficient, demands significant computational power, limiting their application. The study objective was to develop and evaluate a stepwise, multiobjective, multivariable automatic calibration method for the Agricultural Environmental Policy eXtender (APEX) model for simulating runoff, sediment, total phosphorus (TP), and total nitrogen (TN). The most sensitive parameters were grouped according to the process they primarily affect (runoff, sediment transport, soil biological activity, TP transport, and TN transport) and were optimized separately and consecutively. Two multiobjective functions comprising combinations of coefficient of determination ( r 2 ), regression slope, and Nash‐Sutcliffe coefficient (NSC) and a global objective function, the Generalized Likelihood Uncertainty Estimation, were considered to select the optimal parameter combination. A previously manually calibrated and validated APEX model for three adjacent row‐crop field‐size watersheds in northeast Missouri was used as the baseline. The greatest improvements in model performance for sediment, TP, and TN, but not for runoff, were found after runoff parameter optimization, indicating that runoff parameter optimization was crucial for good simulation of sediment and nutrients. The r 2 values for sediment, TP, and TN improved from 0.59–0.87 to 0.77–0.94. The NSC values for TP also improved after soil biological activity and TP parameter optimizations, but subsequent optimizations did not improve sediment or TN simulations. The objective function based on r 2, slope, and NSC outperformed the other objective functions. Modelers can benefit from this cost‐efficient optimization technique (2570 runs for 23 parameters).