Assessing the impacts of technical uncertainty on coupled surface/subsurface flow model predictions using a complex synthetic case

According to the EU Water Framework Directive, Member States have to manage surface water and groundwater at the water body scale and in an integrated way. Flow and transport models constitute useful management tools in this context since they can predict system responses to future stresses. However, numerical modelling at such a scale faces specific issues linked to (1) the representation of the geological and hydrogeological complexity, (2) the uneven level of characterisation knowledge, (3) the representativity of measured parameters and variables in the field, and (4) the CPU time needed for solving the numerical problem. Assumptions and simplifications made for dealing with these issues can lead to a series of models differing by their complexity and by the reliability of their predictions. Consequently, modellers have to find a compromise between complexity and reliability.
The main objective of this research is to estimate the impacts of technical uncertainty, which is the uncertainty related to the numerical implementation, on groundwater flow model predictions. To reach that objective, the methodology consists in comparing reference predictions (hydraulic heads and flow rates) of a complex and close to reality synthetic case with the predictions provided by a series of simplified models (coarse spatial discretisation, coarse time discretisation, simplified law in the unsaturated zone).
The synthetic case reflects the main characteristics found in groundwater bodies of South Belgium (Condroz region of Wallonia), characterised by a succession of limestone synclines and sandstone anticlines. The numerical model is developed with the fully-integrated surface/subsurface flow and transport code HydroGeoSphere using a mesh refined along the surface water network (153027 nodes and 269872 elements). A 5-year reference transient simulation, with daily stress factors is performed. The simulated hydraulic heads and flow rates constitute the reference observations and predictions for the comparison with the simplified models.
The simplified models tested differ by their horizontal (500 m vs. 1000 m element size) and vertical (8 layers vs. 3 layers) spatial discretisations, their time discretisation (daily vs. monthly stress factors), and the type of constitutive law used for simulating the unsaturated flow (linear vs. van Genuchten). The models are run with the same parameter values than those used in the reference model to evaluate the deterioration in model predictions due to technical uncertainty. Additionally, some of the models are calibrated with the inverse modelling code PEST to distinguish how far a model calibration can possibly compensate for technical uncertainty. Then, predictions from each simplified model are compared with the reference predictions of the synthetic case. Then, the simplified models are ranked using several model performance criteria. Results of this research provide guidelines for the numerical implementation of groundwater flow models at the water body scale with respect to specific groundwater management objectives.