On the Inclusion of Surface-Water Observations in the Groundwater Model Calibration Process

Numerical groundwater models are commonly used to predict base-flow contribution to local surface-water features in shallow water table environments. Surface-water systems in these models are typically conceptualized as specified-head or head-dependent groundwater boundary conditions. Alternatively, the surface-water system can be explicitly simulated using a variety of solutions to the shallow surface water flow equations, allowing the inclusion of additional surface-water observations. To evaluate the potential effect surface-water system conceptualization has on predictive uncertainty, two highly-parameterized groundwater models were evaluated. One model was assigned head-dependent flux boundaries to represent the surface-water system, while the other model employed an implicitly coupled, dynamic surface-water solution. A dataset generated from a single synthetic model was used to evaluate the predictive uncertainty of both models. The dataset consists of 365 daily groundwater levels from six locations and surface-water discharge from one location Predictive uncertainty was evaluated using a bi-objective Pareto trade-off analysis, in which the maximum decrease in predicted base flow was identified for both models within a given level of calibration error. A more accurate conceptualization of the surface-water system alone is shown to have negligible effect on the calibration-constrained range of predicted base flow, indicating that groundwater-level calibration targets alone contain little information relative to the prediction. However, when combined with surface-water discharge observations, dynamic representation of the surface-water and groundwater system reduces the predictive uncertainty by constraining the prediction between observations from both the groundwater and surface-water system.