I recently found myself reviewing the opening chapter of my very first hydrology textbook—Physical Hydrology by S. Lawrence Dingman. Here, Dingman expresses that hydrology can be broadly described as having two principal focuses: 1) the global hydrologic cycle—studying quantity and distribution of water across components of the global water system and 2) the land phase of the hydrologic cycle. Dingman’s first focus is one of the main goals of continental-scale modeling. As I read this passage, I came to an exciting realization: the ideas I had learned about so early in my graduate studies outlined the importance of the large-scale hydrology research I would inevitably take on.
Traditionally, hydrologic modeling has been most feasible to implement at the catchment scale—larger simulations are computationally demanding and require vast amounts of data. This approach is problematic to our understanding of the global (or even regional) hydrologic cycle because connections between various aspects of the water cycle extend beyond watershed boundaries.
Climate, earth system, and land surface modeling communities have been using models at continental and global scales for decades, with general circulation models (global climate models) being one of the first methods to numerically capture the hydrologic cycle across spatial scales. However, utilizing process-based hydrologic models at these large spatial scales is relatively new and notoriously difficult. In 2011, Wood et al. (2011) addressed this challenge and identified the expansion of hyper-resolution capabilities in continental and global hydrologic models as one of the Grand Challenges in hydrology.
Model developers have worked to address this Grand Challenge, building hyper-resolution hydrologic simulations to run over vast scales. However, model performance evaluation for large models is necessary and lacking. As a second year PhD student at Princeton University, my current research consists of evaluating performance of the two process-based, high-resolution, continental-scale hydrologic models that have successfully simulated hydrologic processes over the Continental United States—ParFlow-CONUS and the NOAA National Water Model based on the NCAR WRF-Hydro model. Although these models differ somewhat in how they represent processes (for example, ParFlow-CONUS has a significantly deeper subsurface representation), they are similar enough to make a reasonable comparison. My work presents the first intercomparison of continental-scale models that are considered hyperresolution for the spatial extent they represent (Figure 1).
As a proof of concept for this intercomparison, I analyzed simulations of streamflow for both ParFlow-CONUS and a configuration of the National Water Model, and then compared the results to USGS streamflow measurements at 2,200 gauge locations across the US. I found that even with the differences between the models, both successfully matched over half of the USGS gauges in streamflow timing and volume (Figure 2) and had similar patterns of performance across the US. Generally, the Eastern portion of the US had better simulated streamflow than the Central and Western US (Figure 3).
The initial results of this research really point to the WHAT of model performance—we aimed to understand differences between the models and assess their general performance. But the hardest aspect to pin down in this comparison is WHY we see the behavior that we do. The difficulty in understanding the reasoning behind model performance highlights one reason why continental-scale modeling is such a challenge. Unlike smaller models, it is computationally expensive to attempt traditional sensitivity analyses or validation methods, even when utilizing high performance supercomputers.
Parsing out different kinds of model bias was a difficult task, but our team pared it down to these possibilities: 1) meteorological forcing inputs, 2) model physics and parameters, 3) topography and resolution, and 4) anthropogenic influence. My analysis explores how these factors effect model results, particularly considering the spatial discrepancies in simulated streamflow, seen in Figure 2. For example, poorer streamflow performance in the Central and Western US is likely attributed to hard-to-represent, complex topography in the Rocky Mountains, documented precipitation and temperature biases in NLDAS-2 meteorological forcing (ultimately impacting modeled snow water equivalent), and the influence of dams and groundwater pumping which are not represented in either model. Better understanding how these factors influence simulation results will encourage future work and fuel model development.
Through this research, I have realized that the principal focuses of hydrology that Dingman described can only be addressed though the work of countless individuals and the result of collaboration between disciplines. Through these types of intercomparisons and model evaluations, the community can come together to improve continental hydrologic models attempting to address the Grand Challenges of hydrology and use them to gain a better understanding of the global hydrologic cycle. I am excited to continue this work with this community in order to better understand model performance and to use these large-scale models to study the effects of climate-induced land use and land cover change on continental and regional water balances.
Want to learn more about large-scale hydrologic modeling? Check out some resources!
The emergence of global-scale hydrology. WRR. Eagleson, 1986.
Global hydrology 2015: State, trends, and directions. WRR. Bierkens, 2015.
Tijerina, D., Condon, L.E., FitzGerald, K., Dugger, A., Forrester, M., Sampson, K., Gochis, D., Maxwell, R.M. (2020) Continental Hydrologic Intercomparison Project (CHIP), Phase 1: a framework for large-scale hydrology model comparisons. (Manuscript in preparation).