Parameter and Uncertainty Estimation in Groundwater Modelling

The data basis on which groundwater models are constructed is in general very incomplete, and this leads to uncertainty in model outcome. Groundwater models form the basis for many, often costly decisions and if these are to be made on solid grounds, the uncertainty attached to model results must be quantified. This study was motivated by the need to estimate the uncertainty involved in groundwater models.Chapter 2 presents an integrated surface/subsurface unstructured finite difference model that was developed and applied to a synthetic case study.The following two chapters concern calibration and uncertainty estimation. Essential issues relating to calibration are discussed. The classical regression methods are described; however, the main focus is on the Generalized Likelihood Uncertainty Estimation (GLUE) methodology. The next two chapters describe case studies in which the GLUE methodology was applied.Capture zone modelling was conducted on a synthetic stationary 3-dimensional flow problem involving river, surface and groundwater flow. Simulated capture zones were illustrated as likelihood maps and compared with a deterministic capture zones derived from a reference model. The results showed that the reference capture zone was predicted within the 95\% prediction zone. The results depended on a subjective criterion for modelfit. However, it was argued that it would be advantageous to base the criterion for model fit on studies of expected errors in prediction.The GLUE methodology was applied to a regional aquifer system that had previously been the target of parameter and uncertainty estimation within the classical regression framework. The prediction intervals achieved with regard to heads and streamflows were validated against observations. In comparison with the results from the regression analysis, a high number of observations were found to fall outside the 95\% prediction bounds. The spatial distribution of the observation points wherevalidation failed indicated errors in the conceptual description of the aquifer system.