Revisiting the Relationship Between Data, Models, and Decision‐Making

Abstract

We hydrologists can do a better job of supporting water-resources decision-making. I will argue that we can do this by recognizing that decision makers use qualitative, multiple-narrative approaches. So, rather than providing single-model predictions with quantitative uncertainties, we should develop teams of rival models that inform decision makers about what is known, what is possible, and what is unknown. This requires that we build ensembles of models that include biased, advocacy models that directly represent stakeholders' interests or concerns. From this inclusive platform, we can speak objectively and clearly about the risks that drive stakeholders' decisions. Furthermore, we will be promoting more appropriate use of the scientific method in making informed water-resources decisions.

Defensible Model Complexity: A Call for Data‐Based and Goal‐Oriented Model Choice

“…Marshall () highlights the need for creativity and stakeholder feedback in model building. As Ferré () points out, there can be conflicting stakeholder interests that might call for different levels of minimum model complexity. In such situations, building a set of rival models not only serves to explore prediction space, but also “complexity space.” Acknowledging that minimum model complexity is a subjective decision, it is even more important that it becomes common practice to develop, communicate, and defend this decision as transparently and formally as possible (Peeters ), and to find creative ways to involve stakeholders in this process (Castilla‐Rho ).…”

The Present State and Future Application of Cloud Computing for Numerical Groundwater Modeling

“…Potentially, distribution of parallel tasks for the calibration process can be modified to take advantage of spot pricing, and become robust in a DDCE (Schumacher et al unpublished data). This could facilitate the parallel calibration of multiple alternative conceptual models as described in Ferre () and be useful in forecast first calibration methods (White ).…”

“…The multimodeling approach discussed in Ferre (), can address these issues by creating models based on alternative conceptualizations designed by teams of modelers to minimize or maximize predictions of different stake holder interest, which may be sufficient for qualitative hypothesis testing and decision support. However, a small number of alternate conceptual models created by geologists, will not fill, or adequately sample, the space of plausible models for quantitative analysis (Refsgaard et al ).…”

Potential Use of Multimodels in Consulting to Improve Model Acceptance and Decision Making

“…Based on the nature of stakeholder biases and structural noise, multimodels are a logical next step in the continuous evolution of modeling workflows. Potential benefits of the use of multimodels in a consulting environment include: The ability to address the inherent distrust of stakeholders who have an alternative conceptualization in mind by explicitly including advocacy models to represent their concerns (Ferre ). A better exploration of the magnitude and sources of prediction uncertainty due to structural noise by testing multiple conceptualizations, structures, and parameterizations. Establishment of an early warning of an undesired outcome in scenarios where hydraulic properties of discrete pathways, such as faults or open wellbores, have the potential to change or express themselves over time. …”

“…Adopting an approach such as Akaike Information Criterion (AICc; Poeter and Anderson ) is attractive if the model predictions are being applied to a design or engineering problem. When model predictions are being used to address competing stakeholder interests, Ferre () presents a strong argument for the use of advocacy models in a model ensemble in a 5 step process referred to as DIRECT. This approach focusses on defining utility functions for each stakeholder and the use of discriminating model predictions.…”