Impacts of Spatially Variable Soil Depth Characterization on NOAA National Water Model Streamflow Prediction in the Semi-Arid Southwest U.S.

In arid and semi-arid landscapes, dominant controls on water partitioning can be vastly different than in more humid environments. Factors like drought-deciduous vegetation, taproots, hydraulic redistribution, hydrophobicity, channel seepage, and shallow soils can dominate hydrologic response in water-scarce basins, but these processes are not generally represented in global models. Active soil depths in particular can vary dramatically in dry environments due to erosion processes and high spatial variability in biomass. Observations clearly show this heterogeneity impacts local water storage and runoff partitioning, with implications for predicting shorter-term floods and longer-term droughts. New national and global datasets now give us a more spatially complete picture of subsurface characteristics across large domains, allowing us to quantify the impacts of incorporating improved soil depth characterization as compared to the traditional uniform soil depth assumption of many global land surface models.

The NOAA National Water Model (NWM), built on the WRF-Hydro community modeling system, is an operational hydrologic forecasting model predicting major components of the water cycle in realtime across the contiguous U.S. (CONUS). WRF-Hydro was originally developed as a tool to bridge the gap between highly efficient and scalable global models and process-complex catchment models, so it provides an efficient framework for testing the value of this new soil depth parameterization on hydrologic prediction. In this study, we integrate the global SoilGrids (Hengl et al. 2017) and Pelletier et al. (2016) soil depth products separately into an NWM configuration of WRF-Hydro and compare against a benchmark simulation using a fixed 2-meter uniform soil depth. We run a spinup and 5-year simulation for each configuration, then evaluate differences in streamflow, recharge, and evapotranspiration fluxes at locations across the semi-arid Southwest U.S. We first summarize results within the simple context of uncertainty provided by the different soil depth products. We then evaluate regional and seasonal differences in water partitioning between the model experiments. Finally, we quantify computational costs of the additional model complexity over the CONUS-scale NWM implementation. Results of these model experiments will inform NWM version 2.1 implementation, as well as other regional to global-scale model development activities focused on hydrologic prediction.