What we talk about when we talk about uncertainty. Toward a unified, data-driven framework for uncertainty characterization in hydrogeology

In this manuscript, we compare and discuss different frameworks for hydrogeological uncertainty analysis. Since uncertainty is a property of knowledge, we base this comparison on purely epistemological concepts. In a detailed comparison between different candidates, we make the case for Bayesianism, i.e., the framework of reasoning about uncertainty using probability theory. We motivate the use of Bayesian tools, shortly explain the properties of Bayesian inference, prediction and decision and identify the most pressing current challenges of this framework. In hydrogeology, these challenges are the derivation of prior distributions for the parametric uncertainty, typically hydraulic conductivity values, as well as the most relevant paradigm for generating subsurface structures for assessing the structural uncertainty. We present the most commonly used paradigms and give detailed advice on two specific paradigms; Gaussian multivariate random fields as well as multiple-point statistics, both of which have benefits and drawbacks. Without settling for either of these paradigms, we identify the lack of open-access data repositories as the most pressing current impediment for the advancement of data-driven uncertainty analysis. We detail the shortcomings of the current situation and describe a number of steps which could foster the application of both the Gaussian as well as the multiple-point paradigm. We close the manuscript with a call for a community-wide initiative to create this necessary support.