Why Should Practitioners be Concerned about Predictive Uncertainty of Groundwater Management Models?

Numerical models are now commonly used to define guidelines for the sustainable management of groundwater resources. Despite significant advances in inverse modeling and uncertainty analysis, most of groundwater management models are still calibrated by manual trial and error and disregard predictive uncertainty. There is a gap between recent advances in inverse modeling and current practices in operational groundwater modeling. The disinterest of water practitioners for this issue can be explained by unawareness, lack of relevant and reliable datasets, difficulties of implementation and prohibitive computation times. The purpose of this study is to convince water practitioners and water managers that uncertainty analysis is not just a smart, optional add-on to a groundwater model, but rather a critical and necessary step. So as to broaden the audience of this paper out of the community of specialists, we use a simple didactic illustration and propose realistic, practical solutions. Based on a synthetic model, we highlight that if we follow common practices (parameter calibration solely against observed groundwater heads), our knowledge of the unknown parameters is not sufficient to constrain the predicted value of interest (the sustainable yield). This is a critical issue since management models are likely to be used for the design of legal frameworks. After this illustration, we argue that calibration algorithms should become a routine process to bring the uncertainty analysis to the forefront. We promote the use of a linear uncertainty analysis as a diagnostic tool for large real world groundwater management models. When uncertainty is high, stakeholders should encourage the collection of multiple data sets to expand the calibration data set and gather prior information on parameter values.