Abstract. As three-dimensional (3-D) aquatic ecosystem models are used more frequently for operational water quality forecasts and ecological management decisions, it is important to understand the relative strengths and limitations of existing 3-D models of varying spatial resolution and biogeochemical complexity. To this end, 2-year simulations of the Chesapeake Bay from eight hydrodynamic-oxygen models have been statistically compared to each other and to historical monitoring data. Results show that although models have difficulty resolving the variables typically thought to be the main drivers of dissolved oxygen variability (stratification, nutrients, and chlorophyll), all eight models have significant skill in reproducing the mean and seasonal variability of dissolved oxygen. In addition, models with constant net respiration rates independent of nutrient supply and temperature reproduced observed dissolved oxygen concentrations about as well as much more complex, nutrient-dependent biogeochemical models. This finding has significant ramifications for short-term hypoxia forecasts in the Chesapeake Bay, which may be possible with very simple oxygen parameterizations, in contrast to the more complex full biogeochemical models required for scenario-based forecasting. However, models have difficulty simulating correct density and oxygen mixed layer depths, which are important ecologically in terms of habitat compression. Observations indicate a much stronger correlation between the depths of the top of the pycnocline and oxycline than between their maximum vertical gradients, highlighting the importance of the mixing depth in defining the region of aerobic habitat in the Chesapeake Bay when low-oxygen bottom waters are present. Improvement in hypoxia simulations will thus depend more on the ability of models to reproduce the correct mean and variability of the depth of the physically driven surface mixed layer than the precise magnitude of the vertical density gradient.

[1] The overall size of the ''dead zone'' within the main stem of the Chesapeake Bay and its tidal tributaries is quantified by the hypoxic volume (HV), the volume of water with dissolved oxygen (DO) less than 2 mg/L. To improve estimates of HV, DO was subsampled from the output of 3-D model hindcasts at times/locations matching the set of [2004][2005] stations monitored by the Chesapeake Bay Program. The resulting station profiles were interpolated to produce bay-wide estimates of HV in a manner consistent with nonsynoptic, cruise-based estimates. Interpolations of the same stations sampled synoptically, as well as multiple other combinations of station profiles, were examined in order to quantify uncertainties associated with interpolating HV from observed profiles. The potential uncertainty in summer HV estimates resulting from profiles being collected over 2 weeks rather than synoptically averaged $5 km 3 . This is larger than that due to sampling at discrete stations and interpolating/extrapolating to the entire Chesapeake Bay (2.4 km 3 ). As a result, sampling fewer, selected stations over a shorter time period is likely to reduce uncertainties associated with interpolating HV from observed profiles. A function was derived that when applied to a subset of 13 stations, significantly improved estimates of HV. Finally, multiple metrics for quantifying bay-wide hypoxia were examined, and cumulative hypoxic volume was determined to be particularly useful, as a result of its insensitivity to temporal errors and climate change. A final product of this analysis is a nearly three-decade time series of improved estimates of HV for Chesapeake Bay.

The three-dimensional UnTRIM San Francisco Bay-Delta model was applied to simulate tidal hydrodynamics and salinity in the San Francisco Estuary (estuary) using an unstructured grid. We compared model predictions to observations of water level, tidal flow, current speed, and salinity collected at 137 locations throughout the estuary. A quantitative approach based on multiple model assessment metrics was used to evaluate the model's accuracy for each comparison. These comparisons demonstrate that the model accurately predicted water level, tidal flow, and salinity during a 3-year simulation period that spanned a large range of flow and salinity conditions. The model is therefore suitable for detailed investigation of circulation patterns and salinity distributions in the estuary. The model was used to investigate the location, and spatial and temporal extent of the low-salinity zone (LSZ), defined by salinity between 0.5 and 6 psu. We calculated X2, the distance up the axis of the estuary to the daily-averaged 2-psu near-bed salinity, and the spatial extent of the LSZ for each day during the 3-year simulation. The location, area, volume, and average depth of the low-salinity zone varied with X2; however this variation was not monotonic and was largely controlled by the geometry of the estuary. We used predicted daily X2 values and the corresponding daily Delta outflow for each day during the 3-year simulation to develop a new equation to relate X2 to Delta outflow. This equation provides a conceptual improvement over previous equations by allowing the time constant for daily changes in X2 to vary with flow conditions. This improvement resulted in a smaller average error in X2 prediction than previous equations. These analyses demonstrate that a well-calibrated three-dimensional (3-D) hydrodynamic model is a valuable tool for investigating the salinity distributions in the estuary, and their influence on the distribution and abundance of physical habitat.

Sediment dispersal in the Adriatic Sea was evaluated using coupled three‐dimensional circulation and sediment transport models, representing conditions from autumn 2002 through spring 2003. The calculations accounted for fluvial sources, resuspension by waves and currents, and suspended transport. Sediment fluxes peaked during southwestward Bora wind conditions that produced energetic waves and strengthened the Western Adriatic Coastal Current. Transport along the western Adriatic continental shelf was nearly always to the south, except during brief periods when northward Sirocco winds reduced the coastal current. Much of the modeled fluvial sediment deposition was near river mouths, such as the Po subaqueous delta. Nearly all Po sediment remained in the northern Adriatic. Material from rivers that drain the Apennine Mountains traveled farther before deposition than Po sediment, because it was modeled with a lower settling velocity. Fluvial sediment delivered to areas with high average bed shear stress was more highly dispersed than material delivered to more quiescent areas. Modeled depositional patterns were similar to observed patterns that have developed over longer timescales. Specifically, modeled Po sediment accumulation was thickest near the river mouth with a very thin deposit extending to the northeast, consistent with patterns of modern sediment texture in the northern Adriatic. Sediment resuspended from the bed and delivered by Apennine Rivers was preferentially deposited on the northern side of the Gargano Peninsula, in the location of thick Holocene accumulation. Deposition here was highest during Bora winds when convergences in current velocities and off‐shelf flux enhanced delivery of material to the midshelf.