Review of strategies for handling geological uncertainty in groundwater flow and transport modeling

The geologically related uncertainty in groundwater modeling originates from two main sources: geological structures and hydraulic parameter values within these structures. Within a geological structural element the parameter values will always exhibit local scale heterogeneity, which can be accounted for, but is often neglected, in assessments of prediction uncertainties. Strategies for assessing prediction uncertainty due to geologically related uncertainty may be divided into three main categories, accounting for uncertainty due to: (a) the geological structure; (b) effective model parameters; and (c) model parameters including local scale heterogeneity. The most common methodologies for uncertainty assessments within each of these categories, such as multiple modeling, Monte Carlo analysis, regression analysis and moment equation approach, are briefly described with emphasis on their key characteristics. Based on reviews of previous studies, assessments are made on the relative importance of the three uncertainty categories for different types of model predictions. Furthermore, the strengths, limitations and interactions of these methodologies are discussed and conclusions are made with respect to identifying key subjects for which further research is needed. When all sources of uncertainty are analyzed by exploring model parameter and local scale heterogeneity uncertainty for several plausible geological model structures the joint uncertainties can be assessed by use of model averaging techniques, such as Bayesian Model Averaging (BMA). General challenge in model averaging with respect to choosing mutually exclusive and collectively exhaustive choice models, as well as to assign weights when models are used beyond their calibration base, are discussed.

► Uncertainty from geological structure; model parameters; local scale heterogeneity.
► Lessons learned from previous studies on relative importance of these uncertainties.
► Challenges in model averaging.