Comparing two ocean biogeochemical models of Chesapeake Bay with and without the sulfur cycle instead highlights the importance of particle sinking, burial, organic matter, nitrification and light attenuation

A number of models have been developed to simulate hypoxia in the Chesapeake Bay, but these models vary in complexity and in which processes they represent. In this study we implement a previously published biogeochemical code (BioRedoxCNPS) developed for open-ocean waters that includes “cryptic” microbial sulfur cycling within the ChesROMS physical model of the Chesapeake Bay. Sulfur cycling can increase rates of denitrification and anammox in anoxic waters, but the net impacts of such changes on oxygen, nitrate and ammonium are not understood. We compare the results to a physically identical simulation with an estuarine biogeochemical cycling code previously implemented and calibrated in the Bay (ECB). The ECB code neglects sulfur cycling but includes burial of particulate organic matter (POM) and cycling of dissolved organic matter (DOM) and uses different values for many parameters governing phytoplankton growth and particle dynamics. Although the BioRedoxCNPS model produces a better simulation of oxygen and nitrate at a key test site this turns out not to be due to the inclusion of sulfur cycling. Instead, large differences in modeled oxygen and ammonium are largely due to whether or not the biogeochemical codes include cycling of DOM and sedimentary burial of POM. Changes in light attenuation produce large changes in nitrate. Changes in parameters used in both biogeochemical codes (in particular particle sinking velocities) tended to compensate the other differences in model construction. The quantitative impacts of these choices for simulating Chesapeake Bay have not previously been documented in the peer-reviewed literature. Predictions of hydrogen sulfide from our merged model were very sensitive to the choice of parameters and light attenuation. This suggests that observations of hydrogen sulfide could help to constrain these processes in future models.