Last year, I wrote here about our corporate sponsorship of the 2016 Darcy Lecture series, in which Ty Ferré, Ph.D., a professor in the University of Arizona’s Department of Hydrology and Water Resources, travels to research institutions around the world presenting his lecture all year long. Dr. Ferré’s blog breaks down the statistics of the lecture: 52 talks; 41,292 miles in the air; 2,109 miles on a train or bus; 3,400 miles in a rental car; and 3,046 people in the unofficial audience. At the National Ground Water Association’s Annual Summit at the end of April, I got to be one of those audience members listening to this year’s Darcy Lecture.
Rethinking the relationship between data, models, and decision making
This year’s presentation was focused on making a case for analyzing uncertainty to inform decision making. As scientists, uncertainty isn’t something many of us are very comfortable with, and our processes for most of our projects include going to great lengths to reduce or eliminate uncertainty so that we can make the best decisions possible for our clients. However, in hydrogeology as in life, we’re often required to make decisions in the face of some degree of uncertainty, and Ferré’s lecture discussed the ways we can analyze this uncertainty and become more comfortable acting without definitive answers.
Ferré reminded the audience that uncertainty is something we’re used to dealing with in our daily lives, and we often don’t worry too much about the fact that we’re not entirely sure what’s going to happen. Think of a weather forecast—while sometimes we know there’s a 100% chance of sun or rain, it’s more often the case that we might have a 30% chance of rain. Depending on your comfort with taking your chances, you might grab an umbrella on your way out the door in the morning, or you might go without and risk getting rained on. It all depends on how risk-averse you are, and this same thinking extends to the way we carry out projects.
What this lecture highlighted was the importance of collecting discriminatory data for the problems we’re trying to solve, in other words, the data that best test a hypothesis, and Ferré proposes a framework for improving the way hydrogeologic studies are conducted using all the data available to us.
Analyzing uncertainty to satisfy stakeholders
In a project where groups of stakeholders have different interests, concerns, and sensitivities to uncertainty, science can be helpful in the decision-making process by looking for discriminatory data that can test hypotheses against one another. This data can help scientists to create models that simulate possible outcomes of given courses of action; though these are critical to decision making, they contain a significant amount of uncertainty. However, better models will more comprehensively recreate the data collected, which means that determining which models are “better” comparatively can change as more data is collected, particularly when the data most important for model refinement, overall uncertainty reduction, and subsequent enhanced decision-making is collected.
When presented with the same collection of information based on sound science, many people will make different decisions. This means that as scientists, we need to develop a number of models based on the data that shows the variety of potential hydrologic impacts on stakeholders’ interests based on a given proposed solution. Since you can only collect so much data, many models might be equally good, which means that a certain amount of bias ultimately comes into play in interpreting results. By integrating the potential models with one another, you can decrease the impact of bias on the final decision making process and define more clearly the probabilities of outcomes that will affect stakeholders. One of the ways to do this, Ferré says, is to maintain an open view of competing models that advocate for different interpretations, without trying to find the average or best one, simply holding those competing voices simultaneously to consider the possibilities of the uncertain system we’re working with.
In the end, the lecture suggests that while we will always have uncertainty, we should take risk aversion into account in decision making. To reduce fears of unwanted outcomes, develop multiple models that show how these outcomes can be avoided with a given solution. The findings from multiple models are more difficult to brush aside than findings from a single one. Particularly when during the modeling process we are provided an opportunity to improve our understanding and therefore our models by collecting discriminatory data. The goal is to use data that test multiple hypotheses and show the full scope of potential outcomes, refining the likelihood of those outcomes with further data collection and integrated modeling.
This is an approach I’m excited to see gain traction in the industry and one that can hopefully help us give our clients even more peace of mind on the projects we work on.
