Implications of model resolution and parameterization for modeling water management impacts on land-surface, surface, and subsurface processes

Abstract:

Advancements in hydrologic modeling are critical for predicting future water availability, especially in regions with significant water management or anticipated increases in drought risk. Historically, hydrologic models chosen for water resource studies have included highly simplified surface and subsurface process representations due to data and computational resource limitations. Recently, there have been significant advancements in computing performance and accessibility along with continued development of fully integrated, processed-based hydrologic models, including parallelization and advanced non-linear solvers. These advancements have opened the door for more detailed and higher resolution watershed models with improved surface and subsurface process representations. However, increasing model complexity without first understanding the physical effects of model resolution, discretization, and parameterization is unlikely to improve model performance and may increase model uncertainty.

In this study, we conduct a comprehensive analysis on the impact of hydrologic model configuration on modeled land-surface, surface, and subsurface processes. To compare study results over different scales, a suite of simulations using a fully integrated hydrologic model was run at the single column, hillslope, and watershed scale with variable lateral and vertical resolutions, surface and subsurface parameterizations, and water management activities, including irrigation and groundwater pumping. Results were analyzed using a combination of active subspaces and machine learning to evaluate which model configuration choices have the most impact on a variety of model outputs, including evaporative fluxes, surface water and groundwater flows, infiltration, and water storage changes. Additionally, we evaluated how the impacts of modeled water management activities are affected by model configuration. Though model simplifications are still necessary due to data and computational limitations, this study provides a valuable tool to evaluate which model resolution, parameter, or process simplifications will have the least impact on critical model outputs, as well as insight into the uncertainty introduced by these model simplification choices.