We live in an era of Big Data. Hydrogeologists work at the confluence of ever-faster computers, sophisticated modeling packages, parameter estimation software, geographical information systems, remote sensing, and geophysical techniques. Does the existence of all this information and technology mean that we have the black box that accepts our information and *poof* produces accurate predictions—objectively honoring the data and obviating human intervention? In his new book, The Signal and the Noise: Why So Many Predictions Fail–But Some Don't, political and sports prediction guru Nate Silver suggests that successful predictions include subjective judgment. His track record predicting national election outcomes for The New York Times is impressive. He models elections using a combination of state and national polls. The way the polls are combined is a matter of subjective judgment. Performance by polls in past elections is a relatively objective measure, but other information about relevance and quality is subject to interpretation. Scientists might squirm at the notion of releasing their grip on the grail of impartial objectivity. And yet, we know that modeling includes subjective decisions, informed by judgment. Should we ignore this veiled subjectivity and pretend that cold hard data provide all the information we need? Or is it time to recognize and embrace the value of subjectivity? Consider improvements in short-term weather forecasting—especially hurricanes. In weather, like groundwater, the physics is well understood so the models are process-based—not just correlations. Yet, as in groundwater, the true complexity is not fully represented and observations of outcomes are still uncertain. What are the secrets to their success in forecasting? First is running multiple models with different conceptualizations and initial conditions. Second is a post audit of predictions with actual outcomes. Finally, weather forecasters incorporate a certain amount of subjective judgment—gleaned from insights about the modeled system, past model performance, and even quirks of specific software. Big Data helps, but judgment plays a vital role in interpreting the data and in culling valuable signals out of the noise. Weather prediction is deeply informed by models, but improvements are due, in part, to recognition that not all data are equally informative and that judgment calls are needed as to how data and models are combined. Silver says: “The numbers have no way to speak for themselves. We speak for them.” Modelers make many decisions to determine how observations are used in both model construction and parameter estimation. It would be foolish to ignore the data, but it may be equally foolish to assume there is only a single way for the data to “speak.” Silver is a Bayesian. The Bayesian perspective provides a way to combine the objective and the subjective through conditionality and updating. Conditionality is the formal recognition that predictions depend upon everything that feeds them. This includes “hard” data like observations of heads and flows; “soft” data perhaps related to conceptual information about the depositional environment; and seemingly intangible information like software choice, discretization, and so on. Updating occurs when soft data augment hard data and when old information is updated with new information. In the Bayesian perspective, subjectivity is not taboo. For example, changes to the information upon which a model is conditioned, not surprisingly, also change the predictions. These concepts should be familiar to experienced modelers. In groundwater projects, models evolve as we rule out the ridiculous and reinforce the plausible. If we acknowledge and render explicit this subjectivity, this evolution in the modeling approach is not jarring, but naturally becomes part of the process. Progress in prediction will depend on bringing all information, subjective and objective, to the conversation. Taking advantage of both kinds of information, predictions will be more robust, and our analysis more complete. When “speaking for the data,” it is up to us to choose the data that really have something meaningful to say. Note: Opinions expressed are those of the author and not necessarily those of the National Ground Water Association.
Read full abstract