Understanding parameter sensitivity and its management implications in watershed‐scale water quality modeling

Abstract

[1] Because of uncertainty and variability in input parameter values, watershed-scale water quality modeling can result in significant output uncertainty. Quantifying this uncertainty is very important for policy makers and stakeholders who rely on the output of these models for watershed management. Given the large number of parameters in these complex models, a preliminary sensitivity analysis is needed before the uncertainty analysis. A few sensitivity studies have been conducted for watershed models, but efficiently selecting critical parameters for the uncertainty analysis remains an issue. This study aimed to (1) develop a framework for systematically conducting a preliminary sensitivity analysis for complex watershed models using generalized sensitivity analysis (GSA) as a global technique and (2) evaluate the relevance of incorporating management concerns in the preliminary sensitivity analysis and its impact on parameter selection. Although the proposed approach is valid for any complex watershed model, for this study the Watershed Analysis Risk Management Framework (WARMF) model was implemented using data from the Santa Clara River in southern California. To simulate hydrology, sediment, and pesticide transport, 121 parameters are needed for a single catchment/reach combination; an efficient selection method is paramount for an uncertainty analysis. The results show that GSA can be implemented efficiently, yielding insights into model and parameter behavior. The sensitivity analysis must consider management concerns early on in the process to identify parameters and parameter values that can influence management decisions. The number of parameters that must be considered in a subsequent uncertainty analysis was significantly reduced. This study also provides guidance for future research on parameter sensitivity and uncertainty in complex watershed models.Citation: Zheng, Y., and A. A. Keller (2006), Understanding parameter sensitivity and its management implications in watershedscale water quality modeling, Water Resour. Res., 42, W05402,

Flow and nutrient transport in intermittent rivers: a modelling case-study on the Vène River using SWAT 2005

“…Yet, decisions regarding catchment management practices can often be made based on modelling studies (Tripathi et al, 2005;Nicolletta-Ripa & Malone, 2006;Santhi et al, 2006;Zheng & Keller, 2006;Volk et al, 2009). Unlike the Total Daily Maximum Loads (TDML) framework of the US Environment Protection Agency (EPA), the SEQ-EAU quality evaluations are based on concentrations which, in this instance, are less than 1 mg L -1 .…”

Application of dynamic parametric sensitivity analysis for identifying contaminant sources in a watershed

“…The predicted time series of water quality in Fulton Creek at the inlet to Fulton Bay was chosen as the objective function response to evaluate the overall effect of parameter variability. The normalized gradient sensitivity response [ ) θ λ(C, i ] at the Fulton Bay inlet was calculated according to equation (6).…”

“…Most previous applications have been in risk analysis, financial planning, and model based policy assessment studies. The principal application of parametric sensitivity analysis for watersheds has been to assess uncertainty and degree of confidence in existing data or models [1][2][3], hydrological parameter optimization [4], source and process identification in watersheds [5] and as an aid in watershed-scale water quality management [6]. In this paper, parametric sensitivity in response to perturbations in constituent fluxes from lake sediments and contiguous tailings was performed.…”

Estimating Uncertainty in the Revised Universal Soil Loss Equation via Bayesian Melding

“…Various techniques exist that attempt to address the issue of uncertainty in water quality estimates. The generalized likelihood uncertainty estimation technique (GLUE) (Beven and Binley 1992), parameter optimization software (PEST) (Doherty 2007), sensitivity analysis (Zheng and Keller 2006), and emulators such as Bayesian Analysis of Computer Code Output (Oakley and O'Hagan 2002) are four such approaches.…”