Assessing the effects of spatial discretization on large-scale flow model performance and prediction uncertainty

Large-scale physically-based and spatially-distributed models (>100 km2) constitute useful tools for water management since they take explicitly into account the heterogeneity and the physical processes occurring in the subsurface for predicting the evolution of discharge and hydraulic heads for several predictive scenarios. However, such models are characterized by lengthy execution times. Therefore, modelers often coarsen spatial discretization of large-scale physically-based and spatially-distributed models for reducing the number of unknowns and the execution times. This study investigates the influence of such a coarsening of model grid on model performance and prediction uncertainty. The improvement of model performance obtained with an automatic calibration process is also investigated. The results obtained show that coarsening spatial discretization mainly influences the simulation of discharge due to a poor representation of surface water network and a smoothing of surface slopes that prevents from simulating properly surface water-groundwater interactions and runoff processes. Parameter sensitivities are not significantly influenced by grid coarsening and calibration can compensate, to some extent, for model errors induced by grid coarsening. The results also show that coarsening spatial discretization mainly influences the uncertainty on discharge predictions. However, model prediction uncertainties on discharge only increase significantly for very coarse spatial discretizations.