Uncertainty assessment in watershed‐scale water quality modeling and management: 2. Management objectives constrained analysis of uncertainty (MOCAU)

Abstract

[1] Watershed-scale water quality models are increasingly used to support management decision making. However, significant uncertainty in model output remains an unaddressed issue. In our first study, a framework for assessing the uncertainty in watershed modeling and management was developed, and the application of the generalized likelihood uncertainty estimation (GLUE) approach was examined. The influence of subjective choices (especially the likelihood measure) in a GLUE analysis, as well as of availability of observational data, was investigated. On the basis of GLUE, we developed a new Bayesian approach of uncertainty analysis, specifically for management-oriented watershed water quality modeling, as introduced in this paper. The approach, named management objectives constrained analysis of uncertainty (MOCAU), inherits GLUE's equifinality ideology while explicitly considering management objectives and observational uncertainty. It has many unique features that have not been covered (or have not been explored in great detail) by previous GLUE studies. A series of experiments was conducted to investigate the performance of MOCAU. The results show that MOCAU can be applied efficiently, generating accurate uncertainty estimates for management applications. Subjective assumptions in the uncertainty analysis are explicit and realistic, on the basis of management objectives such as nonattainment frequency of water quality objectives. MOCAU also yields insights into watershed model structure improvement and strategic data collection to reduce uncertainty. Besides water quality modeling, MOCAU can also be applied to other complicated modeling problems where errors are significant, observational data is limited, and management objectives are involved. Citation: Zheng, Y., and A. A. Keller (2007), Uncertainty assessment in watershed-scale water quality modeling and management: 2. Management objectives constrained analysis of uncertainty (MOCAU), Water Resour. Res., 43, W08408,

River loads of suspended solids, nitrogen, phosphorus and herbicides delivered to the Great Barrier Reef lagoon

“…We recommend that approaches for assimilating different types of information (models and monitoring data) are developed (Wikle and Berliner, 2007;Reichle, 2008), in order to harness the strengths of both approaches, to quantify the uncertainty in both simultaneously, and ultimately to provide an even stronger scientific basis for estimating and reporting river pollutant loads with estimates of uncertainty (Romanowicz et al, 2006). Representing all such sources of uncertainty (parameter, model and data) in deterministic models, for example by employing Bayesian techniques (Zheng and Keller, 2007), would enable transparent, objective and repeatable comparisons with other load estimates, but is not a trivial task for models like SedNet due to the number of model inputs.…”

Uncertainty Assessment in Watershed-Scale Water Quality Modeling and Management

“…In Zheng and Keller [2007b], we showed that for diazinon in a southern California watershed, a simple model adequately represented the observational error:…”

“…However, the study did not evaluate GLUE's performance from a [Zheng and Keller, 2007a]. We therefore developed an approach, Management Objectives Constrained Analysis of Uncertainty (MOCAU), tailored specifically for management-oriented watershed modeling [Zheng and Keller, 2007b]. MOCAU inherits GLUE's equifinality ideology, while explicitly considering management objectives and observational uncertainty.…”

“…Given its focus on management objectives, MOCAU generates more accurate uncertainty estimates than GLUE for management applications [Zheng and Keller, 2007b]. The key assumptions needed for the uncertainty analysis are based on actual regulatory (or stakeholder) decisions: the management objectives.…”

Markov Chain Monte Carlo (MCMC) uncertainty analysis for watershed water quality modeling and management

“…For different management objectives, different sets of random parameters may need to be considered in the UA. Eventually, the UA has to be management-oriented, as suggested by Zheng and Keller (2007b). Nevertheless, how to determine an appropriate set of random parameters for the MCMC UA deserves a separate study.…”