Assessing the impacts of technical and structure choices on groundwater model performance using a complex synthetic case

According to the European Water Framework Directive (2000/60/EC) and the specific Groundwater Directive (2006/188/EC), Member States have to manage groundwater at the groundwater body scale and in an integrated way. Given the objectives of “good quantitative and qualitative status” of groundwater for 2015 stated by the Directive, end-users want to know the quantitative and qualitative evolution of groundwater for several scenarios. Physically-based and spatially-distributed groundwater flow and transport models constitute useful management tools in this context since they take explicitely into account the heterogeneity and the physical processes occuring in the subsurface for predicting system responses to future stress factors. However, at such a scale, groundwater flow and transport modelling is challenging due to (1) the complexity of geological and hydrogeological contexts, (2) the uneven level of characterisation knowledge, and (3) the representativity of measured parameters. Furthermore, such models require long execution times. As a consequence, a series of choices and simplifications are made for dealing with these issues. Therefore, the outstanding question is to know whether endusers’expectations can be met in spite of such choices and simplifications. This work focuses on choices and simplifications related to spatial discretisation and saturation–pressure relations in the unsaturated one. The influence of stress factor time resolution is also tested. Considering this general context, the objective of the present work is to evaluate the influence of some model technical (spatial discretisation) and structure (saturation–pressure relations) uncertainties on model results, parameter sensitivities, and optimisation performance in order to provide guidelines for model development. This is performed using a synthetic case inspired by typical groundwater bodies of Wallonia (Belgium). This synthetic case is used for obtaining reference observations in terms of flow rates and hydraulic heads. These reference observations are then compared with their simulated equivalent produced by simplified models differing by their spatial discretisation, their saturation–pressure relations in the unsaturated zone, or the time resolution of their stress factors. The simplified models are then ranked using several performance criteria measuring the discrepancies between reference observations and their simulated equivalent. This ranking leads to guidelines for large-scale groundwater flow model development with respect to typical end-users’ expectations. Whatever the time resolution of stress factors, the quantitative and qualitative analyses performed indicate that coarsening horizontal spatial discretisation deteriorates mainly the simulation of flow rates, coarsening vertical spatial discretisation deteriorates mainly the simulation of hydraulic heads, and (over)simplifying saturation–pressure relations in the unsaturated zone significantly impair the simulation of both flow rates and hydraulic heads. Although optimisation can compensate for errors induced by model technical and structure uncertainties, the improvement of model fit is limited, especially for the coarsest models. Furthermore, with respect to end-users’ expectations, the weighted least-square objective function is not always the most relevant criteria for optimising models. Therefore, it is essential to use specific performance criteria for evaluating model performance depending on the objectives of the study. The ideal would be to develop an end-users objective function for including such performance criteria in the optimisation process and stop the optimisation process once performance criteria would have reached the values specified by the end-users with respect to the objectives of the study