Toward Reproducible Environmental Modeling for Decision Support: A Worked Example

Abstract

A fully worked example of decision-support-scale uncertainty quantification (UQ) and parameter estimation (PE) is presented. The analyses are implemented for an existing groundwater flow model of the Edwards aquifer, Texas, USA, and are completed in a script-based workflow that strives to be transparent and reproducible. High-dimensional PE is used to history-match simulated outputs to corresponding state observations of spring flow and groundwater level. Then a hindcast of a historical drought is made. Using available state observations recorded during drought conditions, the combined UQ and PE analyses are shown to yield an ensemble of model results that bracket the observed hydrologic responses. All files and scripts used for the analyses are placed in the public domain to serve as a template for other practitioners who are interested in undertaking these types of analyses.

Risk‐Based Wellhead Protection Decision Support: A Repeatable Workflow Approach

“…Recognizing the importance of uncertainty, a public/private educational partnership seeking to improve the use of models in decision-making states the importance of uncertainty from the initial discussions of a project (the Groundwater Modelling Decision Support Initiative; https://gmdsi.org/about/manifesto/). Our workflow is implemented using PEST++ (White et al 2020a(White et al , 2020b to incorporate both parameter and observation uncertainty using ensemble methods. The importance of repeatability and transparency (e.g., Goecks et al 2010;Peng 2011;Morin et al 2012;Johansson 2015) is also growing in the hydrologic community (Anderson et al 2015, chapter 11.1).…”

“…We depart from the synthetic Freyberg model covered in Hunt et al (2020) and White et al (2020aWhite et al ( , 2020b and extend the workflow to include model construction and an applied decision-support outcome in a regulatory context. We illustrate our workflow using a real world example in a humid temperate climate of southeast New York, USA.…”

“…Hunt et al (2020) conclude with a recognition that additional utility of uncertainty for forecasts and repeatability could be gained through scripting. White et al (2020aWhite et al ( , 2020b addressed the inclusion of automated forecast uncertainty analyses needs through additional capabilities in PEST++ version 5 and demonstrated their use on an enhanced "Freyberg model," where the problem was made more complex in space and time. White et al (2020aWhite et al ( , 2020b applied this approach for repeatable data assimilation and uncertainty analysis to a real world example in the Edwards aquifer in Texas.…”

Early Uncertainty Quantification for an Improved Decision Support Modeling Workflow: A Streamflow Reliability and Water Quality Example

“…Notwithstanding the challenges associated with formulating and verifying a conservative prior, a number of strategies to circumvent the effects of a non-conservative prior have been postulated in previous studies. These include: adopting high parameter dimensionality (e.g., Hunt et al, 2007;Knowling et al, 2019), with parameterization expressing system uncertainty at different spatial and temporal scales (e.g., White et al, 2020a;McKenna et al, 2020), and processing or transforming simulated outputs to minimize uncertainty, and thereby also the effects of an inadequate prior (e.g., Sepúlveda and Doherty, 2015;Knowling et al, 2019). Deploying such strategies is an important component of "Model definition" and prior formulation in "Prior uncertainty quantification" in the proposed workflow.…”

“…The case study contributes to the small body of real-world decision support worked examples that are currently available in the international literature (e.g., Kunstmann et al, 2002;Enzenhoefer et al, 2014;Sepúlveda and Doherty, 2015;Brouwers et al, 2018;Sundell et al, 2019;White et al, 2020a).…”