Evaluating Simulated Groundwater Surface Water Interactions across the Continental US with an Integrated Hydrologic Model (Invited Presentation)

Integrated hydrologic models are increasingly deployed across large spatial domains at high resolution. These tools have the potential to provide novel insights into hydrologic connections across spatial scales, and to inform global earth systems modeling. However, the scale and computational demands of large, physically based models are not well suited to the model calibration and validation approaches traditionally applied to watershed scale applications. Furthermore, even within relatively well monitored countries such as the US, the existing point observations are sparse relative to the outputs that can be generated by high resolution gridded models. Point observations can also be biased spatially resulting in the overrepresentation of some hydrologic settings relative to others. For example, the concentration of groundwater observations in areas with groundwater pumping. Novel approaches are needed to evaluate the ability of simulations to reproduce physically meaningful spatial structures and temporal signatures, in addition to point comparisons. Here, we focus on the evaluations of groundwater surface water interactions across the continental US. We use the fully integrated model ParFlow, which combines 3D variably saturated groundwater flow with overland flow and land surface processes, solving the coupled water energy balance from the bedrock to the top of the canopy. The model provides high-resolution (1 km²) outputs over a large spatial extent (~6.3 million km²) that are used to characterize groundwater surface water exchanges across a wide range of hydroclimatic settings and spatial scales, not feasible with other approaches. Simulations include baseline historical climate simulations in addition to perturbed warming simulations. Model outputs are evaluated using more than 1.2 million groundwater and surface water observation available over the simulation period, in addition to multiple gridded reanalysis products. Combining multiple metrics, we classify model performance relative to likely sources of bias such as uncertainty in model inputs, limitations of spatial resolution, missing processes or features. Aggregating results to watershed units we characterize watershed function using the Budyko framework and evaluate model ability to reproduce partitioning relationships in watersheds ranging from 100 to 1,000,000 square kilometers.