During the first year of the US IOOS Modeling Tested, The Estuarine Hypoxia Team (EHT) focused on model-data and model-model comparisons of five different 3-D hydrodynamic models and six different hydrodynamic-hypoxia model combinations, for both a wet and a dry year (2004 – 2005), including sensitivity studies in response to refined environmental forcings and grid resolution. The data used for comparison were profiles of temperature, salinity and dissolved oxygen collected by the EPA Chesapeake Bay Program at ~40 monitoring stations distributed throughout the Bay and sampled every 2 to 4 weeks. The five hydrodynamic models utilized (and associated lead modelers) were CH3D (L. Linker/C. Cerco, EPA/USACE CBP), EFDC (J. Shen, VIMS), ChesROMS (R. Hood/W. Long, UMCES), CBOFS (L. Lanerolle, NOAA-CSDL), and UMCES-ROMS (M. Li/Y. Li, UMCES).

The first six hypoxia-hydrodynamic model combinations we compared (and their associated EHT investigators) are (1) CH3D-ICM, where ICM is a complex multi-component ecosystem formulation (L. Linker/C. Cerco, EPA/USACE CBP), (2) ChesROMS-BGC, where BGC is a NPZD-type biogeochemical approach (W. Long/R. Hood, UMCES), (3) EFCD-1eqn, where 1eqn is a one-equation respiration model driven by sediment oxygen demand (J. Shen, VIMS), (4) ChesROMS-1term, where 1term is an everywhere constant net respiration formulation (M. Scully, ODU), (5) ChesROMS-DD, which adds linear depth-dependence to (4) (M. Scully, ODU), and (6) CBOFS-1term, which connects the 1term approach to CBOFS rather than to ChesROMS (L. Lanerolle, NOAA-CSDL). Hypoxia models (3) through (6) assume the organic matter supplying respiration is unlimited, and thus these four models are independent of nutrient input. As a tool for model intercomparison and model-data skill assessment, we have utilized Target diagrams, whose advantages include: their visual simplicity (the closer the model output approaches the Target's “bull's-eye”, the better), relatively straightforward mathematics (based on root-mean-square differences (RMSD)), and the large amount of output that can be summarized on a single diagram.

Major results from the hydrodynamic model comparison for Chesapeake Bay include a demonstration that the five 3-D models all do reasonably well in capturing fundamental aspects of the hydrodynamics, although the precise depth and intensity of stratification at the pycnocline continues to be a universal challenge. Temperature was simulated very well by every one of the models. The CH3D and EFDC models did slightly better in reproducing bottom salinity and density stratification, whereas the CH3D and ChesROMS models did slightly better in reproducing pycnocline depth. All five models under-predict the strength and variability of salinity stratification. The 3 ROMS-based models (ChesROMS, UMCES ROMS, CBOFS) demonstrated remarkably similar skill in reproducing bottom salinity, stratification and pycnocline depth, indicating that this skill is not highly dependent on horizontal grid resolution. The model behavior for salinity stratification and associated error was remarkably similar in terms of its temporal and spatial variability across all five of the models.

It is possible that a fundamental structural aspect common to all five models may be responsible for the similar errors in stratification. Somewhat surprisingly, the models' many innate differences do not appear to result in large differences in skill, including differences in grid resolution (which vary by a factor 5), different vertical coordinate systems (z-grid vs. sigma-grid), and differences in the nature of model forcing (e.g., modeled vs. observed winds, shelf climatology vs. nudging by bay-mouth observations, and a full watershed model vs. simple point discharges). Recent simulations suggest that potential culprits may include turbulence closure and/or advection formulations present in all five hydrodynamic models.

Major results of the hypoxia model-data comparisons include a demonstration that the six hydrodynamic-hypoxia model combinations tested to date all do reasonably well in capturing of the seasonal variability of the dissolved oxygen field. By a narrow margin, the EFDC-1eqn model performed best in reproducing bottom dissolved oxygen, whereas ChesROMS-1DD performed best in reproducing hypoxic volume in 2004 and ICM performed best in reproducing hypoxic volume in 2005. However, the differences in skill among the various hypoxia models were generally small. These results in aggregate confirm that all the models tested are near the cutting edge of the best present knowledge with regards to accurately and efficiently hindcasting oxygen for a large estuary of national importance. Another significant finding with regards to future modeling strategies is the result that the ensemble hindcast for hypoxic volume using multiple models (ICM, CBOFS2-1term, plus 3 ChesROMS DO models) was more accurate than the hindcast from any one model alone.

Scientifically, key results to date include the ability of models with highly simplified biology (e.g., a constant net respiration rate independent of nutrient input) to reproduce the seasonal hypoxia cycle about as well as much more complex, nutrient-dependent ecological models. Reproduction of seasonal variation in DO was found to not be dependent on the seasonal cycle in respiration rate, nor the seasonal cycle in fresh water input, nor the seasonal cycle in density stratification. In fact, all the models reproduced the observed seasonal cycle in bottom DO better than they reproduced the seasonal cycle in observed stratification. Further sensitivity analyses are currently being conducted with this set of models in order to examine the sensitivity of seasonal variations in DO to wind speed and direction.