Modeling the impact of floating oyster ( Crassostrea virginica ) aquaculture on sediment − water nutrient and oxygen fluxes

Bivalve aquaculture relies on naturally occurring phytoplankton, zooplankton, and detritus as food sources, thereby avoiding external nutrient inputs that are commonly associated with finfish aquaculture. High filtration rates and concentrated bivalve biomass within aquaculture operations, however, result in intense biodeposition of particulate organic matter (POM) on surrounding sediments, with potential adverse environmental impacts. Estimating the net depositional flux is difficult in shallow waters due to methodological constraints and dynamic processes such as resuspension and advection. In this study, we combined sediment trap deployments with simulations from a mechanistic sediment flux model to estimate seasonal POM deposition, resuspension, and processing within sediments in the vicinity of an eastern oyster Crassostrea virginica farm in the Choptank River, Maryland, USA. The model is the stand-alone version of a 2-layer sediment flux model currently implemented within larger models for understanding ecosystem responses to nutrient management. Modeled sediment−water fluxes were compared to observed denitrification rates and nitrite + nitrate (NO2+NO3), phosphate (PO4) and dissolved O2 fluxes. Model-derived estimates of POM deposition, which represent POM incorporated and processed within the sediment, comprised a small fraction of the material collected in sediment traps. These results highlight the roles of biodeposit resuspension and transport in effectively removing oyster biodeposits away from this particular farm, resulting in a highly diminished local environmental impact. This study highlights the value of sediment models as a practical tool for computing integrated measures of nitrogen cycling as a function of seasonal dynamics in the vicinity of aqua culture operations.


INTRODUCTION
Molluscan aquaculture is thriving in many US coastal bays and estuaries (Newell 2004, Murray & Hudson 2013) with a value of US $ 328 million in 2013 representing a 75% growth rate since 2005.On the US east coast, 315 farms grew US $ 68.3 million worth of eastern oysters and 278 farms grew US $ 64.6 million worth of hard clams in 2013 (NASS 2014).Despite these successes, many suitable locations that can sustain bivalve aquaculture have yet to be utilized.There are several reasons for this slow development, but a major impediment is considerable opposition to the use of public waters for aquaculture due to poten-tial adverse environmental consequences (Langan et al. 2006, NRC 2010).Such opposition is based largely on early finfish aquaculture operations in coastal waters where excess feed and fecal material resulted in over-enrichment of organic matter in underlying sediments and caused adverse effects on benthic communities.Although these practices have improved due to higher feed utilization, better placement of farms (Dudley et al. 2000), and fallowing of sites, public perception of aquaculture practices remains unfavorable.In distinct contrast, bivalve aquaculture farms do not result in the import of new nutrients to the system because bivalves feed solely on naturally available phytoplank ton (i.e.no nutrients are added); how ever, these operations transfer nutrients from the water-column to sediments.Consequently, adverse effects of overenrichment of underlying sediments occur due to deposition of organic matter by bivalve biodeposits (Crawford et al. 2003, Mallet et al. 2006, Mitchell 2006, Cranford et al. 2009, Dumbauld et al. 2009, Forrest et al. 2009, McKindsey et al. 2011).
Bivalves filter large quantities of phytoplankton and detritus from the water column, and unincorporated particulate organic matter (POM) is then transferred as feces and pseudofecal rejecta (collectively termed biodeposits) to the sediment surface (Bayne & Newell 1983, Newell & Langdon 1996, Ward & Shumway 2004, Cranford et al. 2011).Settling biodeposits can have a range of ecosystem effects, depending on the physical and chemical conditions present during and after the settling process.High seston deposition rates (including POM) are often associated with high rates of oxygen uptake and nutrient release by sediments (Mazouni et al. 1996, Souchu et al. 2001).However, if biodeposits settle on sediments that contain oxygen in the surficial layers, NH 4 + regenerated from organic matter is oxidized to NO 2 − and NO 3 − during nitrification, some of which diffuses into anaerobic sediments, where it may be converted to N 2 via denitrification as coupled nitrificationdenitrification (Henriksen & Kemp 1988, Seitzinger 1988, Rysgaard et al. 1994).In cases where suspension-feeding bivalves cause localized over-enrichment of the surrounding sediment, the depletion of oxygen near the sediment -water interface (Newell 2004) allows sulfide accumulation and associated nitrification inhibition, allowing regenerated nitrogen to remain in the system as NH 4 + , potentially supporting further algal and microbial production locally (Kaspar et al. 1985, Asmus & Asmus 1991, Giles et al. 2006).While nitrogen retention may be desirable in nutrient-poor systems, the opposite is true in eutrophic systems, and these adverse effects of bivalve aquaculture have most often been ob served in locations with weak water currents that do not promote the distribution of the biodeposits over a wide area of the underlying sediments (Tenore et al. 1982, Kaspar et al. 1985, Mazouni et al. 1996, Barranguet 1997, Mazouni 2004, Holyoke 2008).
Numerical models are valuable tools for integrating field data and increasing the spatial and temporal scope of investigations into the processes associated with the impacts of shellfish aquaculture on sediments.Many types of models have been used to compute biodeposition associated with shellfish aquaculture (Grant et al. 2005, Weise et al. 2009), stock production (Gangnery et al. 2001, Duarte et al. 2003), carrying capacity (Byron et al. 2011), and ecosystem effects and linkages (Ren et al. 2012).From a biogeochemical perspective, numerical models can be used to quantify the spatial and temporal impacts of aquaculture-derived POM deposition on underlying sediments and how these impacts are related to farm size, culture intensity, and the physical environment of the aquaculture site.Such models are valuable not only for scientific studies of the aquaculture operations themselves, but are more often useful for assessments of production capacity and resource management, including water quality criteria assessment for aquaculture operations.
The purpose of this study was to develop sediment biogeochemical modeling infrastructure capable of predicting changes in nutrient and oxygen cycling in sediments associated with an oyster aquaculture operation in the Choptank River, Chesapeake Bay, USA.This process-based approach allows for the b ackcalculation of realistic patterns of biodeposition by allowing the model to account for non-linear dynamics in nutrient cycling associated with a variable aerobic layer.Model-data comparisons and calibration were then used to examine spatial and temporal variability in sediment biogeochemical processes near the oyster farm, which in turn allowed for the examination of aquaculture-associated POM transport at the site.The final objective was to use the model to develop a nitrogen budget and assess the susceptibility of the site to the impacts of concentrated biodeposition.

Site description
Our study site is an oyster farm (Marinetics) located on the southern shore of the Choptank River in eastern Chesapeake Bay, MD (Fig. 1).The farm oper-ates on a 4-acre oyster lease stocked with up to 5000 floats (0.75 × 1.8 m), each containing several thousand smaller (0.5−3 cm shell height) or hundreds of larger (3−9 cm shell height) cultchless oysters.As oyster biomass in the floats increases during the growing season, oysters are removed from the floats, graded by size, and returned at lower densities.A sampling of some of the 2010 cohort (1.2 million oysters in 1020 trays) in 2011 averaged 663 oysters per tray (mean oyster shell length = 5.4 cm), 422 per tray (7.3 cm), or 93 per tray (8.8 cm).Water temperature in the Choptank River ranged from 1−32°C over the course of the year, with May through October temperatures >15°C, and mean surface temperatures > 25°C from June to August when maximum growth and biodeposition was observed (Table 1).The local tidal range is 0.5 m and salinity ranged from 8 to 14.The site is sheltered from the strongest northwest winds and has sediments ranging from sands to muds, with relatively shallow depths (~0.5−1.5 m at MLLW) (Fig. 1).Sediment trap deployments and sediment -water nutrient and oxygen flux measurements were carried out at 3 sites in and around the farm, including a site inside the oyster float matrix (farm site), a site ~350 m south of the float matrix in about 2 m of water (near-farm site) and an adjacent reference location that is sufficiently distant (~800 m south east) to be uninfluenced by oyster biodeposition, but otherwise similar to the farm site (reference site) (Table 1, Fig. 1).The near-farm site was chosen at a location that we hypothesized would tend to accu mulate biodeposits, if dispersal from the farm was directed towards the land (southward) as op -posed to Riverward (northward).This is supported by the fact that sediments at this site are dominated by fine grained sediment (i.e.silt), which contrasts with the sand-dominated farm and reference sites (Table 1).The reference site was specifically chosen to have similar depth, bottom sediments, and geographical orientation to the farm site (Table 1, Fig. 1).A tidal model of the region, MIKE2D (J.Richardson pers.comm.) also shows very similar tidal currents (velocity and direction) at the farm and reference sites (0.2 m s −1 alongshore at maximum ebb).The reference site allowed us to characterize seasonal processes asso ciated with the cycling of naturally deposited and transported organic matter, but in the absence of significant biodeposition.

Sediment trap deployments
Estimates of POM deposition to the sediments at each study location and time were derived from sediment trap measurements.Sediment traps were deployed for approximately 24 h at 3 sites (farm, near-farm, reference) (Fig. 1) during 4 sampling cruises in 2011 (18 April, 6 June, 1 August, and 29 September).All deployments were during relatively calm conditions to avoid potential damage by large surface waves, and in water deep enough to ensure that the traps were never exposed or directly impacted by oyster floats during low tide.Traps (8 replicates at each site) were constructed from 5 × 30.5 cm (internal diameter × height) PVC pipes that were supported vertically off the bed (off-bottom traps) in an additional piece of pipe (7.6 cm diameter) embedded in cast concrete.The trap openings were approx.30 cm above the sediment.For the 2 sampling periods in August and September, we also deployed 8 additional sediment traps (identical dimensions) that were suspended on monofilament line immediately underneath the oyster floats (under-float traps) to capture biodeposits as the material first settled from the floats.
Sediment traps deployed near the bottom in shallow tidal waters can be strongly influenced by bottom sediment resuspension (Ko et al. 2003).The sediment traps deployed at Marinetics oyster farm collected a combination of several types of organic matter, including background sedimentation, resuspension of both the am bient organic deposits and oyster biodeposits, and sedi mentation of new oyster-derived feces and pseudo-feces from the farm.We used data from several trap deployments to estimate these contributions at each site.The off-bottom traps deployed at the reference site measured a combination of background (ambient) new sedimentation and resuspension unaffected by the oyster farm; these are referred to as the reference fluxes (F ref ).The off-bottom trap deployed at the farm site measured a combination of background new sedimentation and resuspension, new bio deposition from the oyster floats, and resuspension of biodeposits (hereafter F tot ).The traps deployed directly beneath the oyster floats measured new bio deposition from the oyster floats (F s ).F s was adjusted to account for the fact that the oyster floats only occupy a fraction, r, of the surface area of the farm site (Fig. 1), while the off-bottom traps represent an average of the whole bottom area.Over-flight images of the farm were used to calculate an r of 0.43, dividing the area occupied by floats within a representative subregion of the farm by the total area of that subregion.Thus, the magnitude of the biodeposition source term per unit bottom area was estimated as rF s .
Using all 3 sediment trap types and, given the above assumptions, the resuspended biodeposit flux (F res ) was calculated as: This estimate was calculated directly for the August and September 2011 observations at the oyster farm, but not for the April and June 2011 observations since no oyster float traps were deployed during these months, and no direct measurements of rF s were available.To obtain reasonable estimates, we assumed that rF s was a fraction (A) of the total biodeposit flux caught in the off-bottom traps: Eq. ( 2) is based on the idea that greater biodeposition will be accompanied by proportionally greater biodeposit resuspension.Eq. ( 2) is also the simplest possible function that approaches 0 as F tot approaches F ref (i.e. as the influence of oysters approaches 0).Using the data for August and September ( when all 3 trap types were deployed, resulted in estimates of A = 0.57 for August and A = 0.76 for September.We adopted the average value of A = 0.67 to estimate the biodeposit source term for April and June using Eq.(2) (Table 2).Finally, we estimated resuspended biodeposit fluxes using Eq. ( 1), with the results also listed in Table 2.
We also derived estimates of net POM deposition to sediments using a sediment flux model (SFM).This scheme estimates the annual POM deposition, which is the primary input to the biogeochemical model, so that the modeled sediment -water NH 4 + fluxes best fit the observed NH 4 + fluxes (Di Toro 2001).Because such a method ignores the year-to-year carryover of POM, adjusting the annual POM deposition to fit NH 4 + flux over a number of years is a complex estimation problem since every year affects each subsequent year in proportion to the store of organic material that did not undergo diagenesis (Brady et al. 2013).A Hooke-Jeeves pattern search algorithm was utilized to minimize the root mean square error (RMSE) between modeled and observed NH 4 + flux (Hooke & Jeeves 1961, Brady et al. 2013).The algorithm minimizes a cost function; in this application the RMSE in the predicted NH 4 + flux is minimized by varying yearly average de positional fluxes.The pattern search starts with an initial estimate of the POM depositional flux, which in this case is a constant yearly organic matter depositional flux of 35 mmol C m −2 d −1 , a reasonable estimate based on literaturederived values for similar sites in Chesapeake Bay, USA (Roden et al. 1995, Kemp et al. 1999, Hagy et al. 2005).This was followed by exploratory moves (± 30%), changing the POM depositional flux for the first year, recording the direction that reduces the cost function (i.e.NH 4 + flux RMSE for the first year).This process is repeated for each year and the direction of change (± 30% POM depositional flux) that reduces the error is retained for each year.These directions of change that reduce the RMSE for each year of the pattern search are repeated until the RMSE no longer decreases.The individual year-byyear search is then repeated and a new pattern is established and repeatedly applied.If it is not possible to find a new pattern, that indicates a local minimum has been found.At this time, the size of the exploratory moves is reduced from ± 30 to ±10% in 10% intervals and finally ± 5% to converge to a final solution.The only constraint on depositional flux was a minimum of 8.3 mmol C m −2 d −1 , 10% below the lowest depositional flux measured in Chesapeake Bay (Kemp et al. 1999, Hagy et al. 2005).

Sediment -water fluxes of NH
3− , and O 2 were estimated from temporal changes in constituents during 4 sets of core incubations.Sediments were collected and core incubations were performed on 10 April, 6 June, 1 August, and 26 September in 2011.Acrylic cores that were 7 cm (inner diameter) and 30 cm tall were collected by pole coring (Gao et al. 2014) at the 3 study sites (Fig. 1).The methods for core incubation are de scribed in detail in Cornwell et al. (2014).Briefly, triplicate cores from each site that had aerobic overlying water conditions were bubbled with air over night while submersed in a temperature controlled environmental chamber.Core tubes were sealed by acrylic lids with suspended magnetic stirrers attached and time series of solute (NH 4 ) and gas (O 2 , N 2 , Ar) concentrations were determined within the cores over the incubation period.Gas concentrations were measured from high precision N 2 :Ar or O 2 :Ar ratios using membrane inlet mass spectro metry (Kana et al. 1994).Cores were incubated in the dark for 4 time points and then under illumination at ambient irradiances for an additional 3 time points.While the fluxes of N 2 are referred to as denitrification, they are actually the summation of all gaseous N transformation processes and may include processes such as anammox (Rich et al. 2008) or N fixation associated with sulfate reduction (Bertics et al. 2013); fluxes of N 2 O were not measured.
Data for overlying water-column nutrient and O 2 concentrations nearest the sediment -water interface in Chesapeake Bay, which are required boundary conditions for the stand-alone SFM simulations, were retrieved from the Chesapeake Bay Program (CBP) Water Quality database (www.chesapeakebay.net/data_ waterquality.aspx)for a nearby site in the Choptank River (Stn ET5.2; 38.5807°N, 76.0587°W).Measurements of bottom water salinity, dissolved O 2 , NH 4 + , NO 3 − , and PO 4 3− were also made as part of the sediment -water flux measurements and were augmented by CBP data by combining the time series and using piecewise cubic hermite interpolation (PCHIP) to derive daily overlying water-column values.

Sediment biogeochemical modeling
We applied a 2-layer sediment flux model (Di Toro 2001, Brady et al. 2013, Testa et al. 2013) to examine the biogeochemical response of the sediments given the POM depositional loading observed using sediment traps proximal to and distant from aquaculture facilities (Figs. 1 & 2).SFM has accurately simulated sediment -water fluxes of NH 4 + , NO 2 − plus NO 3 − , PO 4 3− , and dissolved silica for diverse chemical and physical environments throughout Chesapeake Bay (Brady et al. 2013, Testa et al. 2013).The model structure for SFM involves 4 general processes: (1) the sediment receives depositional fluxes of POM (C, N), as well as biogenic and inorganic phosphorus and silica from the overlying water, (2) the decomposition of POM produces soluble intermediates that are quan tified as diagenesis fluxes, (3) solutes react, trans fer between solid and dissolved phases, are transported between the aerobic and anaerobic layers of the sediment, or are released as gases (CH 4 , N 2 ), and (4) solutes are returned to the overlying water (Fig. 2).SFM numerically integrates mass-balance equations for chemical constituents in 2 functional layers: an aerobic layer near the sediment -water interface of variable depth (H 1 ) and an anaerobic layer below that is equal to the total modeled sediment depth (0.1 m) minus the depth of H 1 .The model includes an algorithm that continually updates the thickness of the aerobic layer, H 1 .Values for H 1 are computed as the product of the diffusion coefficient (D O 2 , m 2 d -1 ) and the ratio of overlying water (layer 0) O 2 concentration (O 2 (0), mmol m −3 ), to the sediment oxygen demand (SOD, mmol m −2 d −1 ): (3) This relationship was perhaps first suggested by Grote (1934) -quoted by Hutchinson (1957) -and verified by measurements (Jorgensen & Revsbech 1985, Cai & Sayles 1996).The inverse of the second term on the right hand side of Eq. ( 3) is the surface mass transfer coefficient (K L 01 , m d −1 ): (4) The surface mass transfer coefficient controls solute exchange between the aerobic layer and the over lying water column (Fig. 2).The model can, therefore, use the same mass transfer coefficient for all the solutes since differences in the diffusion coefficients between solutes are subsumed in the kinetic para meters fitted to data (Di Toro 2001, Fennel et al. 2009).The model assumes that organic matter mineralization is achieved by denitrification, sulfate reduction, and methanogenesis, thus aerobic respiration is not explicitly modeled.The simulation time-step is 1 h and output is aggregated at 1 d intervals.
The key equations utilized to estimate particulate organic nitrogen deposition include the mass balance equations for NH 4 + in the aerobic and anaerobic layers in Eqs.(5 & 6), respectively: (2) are the NH 4 + concentrations (mmol m −3 ) in the overlying water, aerobic, and anaerobic sediment layers, re spectively.K L01 is the sediment -water mass transfer coefficient (m d −1 ), K L12 is the mass transfer coefficient between the aerobic and anaerobic layer (m d −1 ), k NH 4 + ,1 is the nitrification reaction velocity (m d −1 ), H 1 and H 2 are the depth of the aerobic and anaerobic layers, respectively (in meters), and J N 1 and J N 2 are the aerobic and anaerobic layer diagenesis rates, respectively (mmol N m −2 d −1 ).
The diagenesis of POM is modeled by partitioning the settling POM into 3 reactivity classes, termed the G model (Westrich & Berner 1984).Each class represents a fixed portion of the organic material that reacts at a specific rate (Burdige 1991).For SFM, 3 G classes represent 3 levels of reactivity: G 1 is rapidly reactive (20 d half-life, 65% of settling POM), G 2 is more slowly reactive (1 yr half-life, 20% of settling POM), G 3 (15% of settling POM) is non-reactive in this particular model (Brady et al. 2013).The dia genesis expression for carbon is as follows (similar equations govern diagenesis of particulate organic nitrogen and phosphorus): where POC i is the POC concentration in reactivity class i in the anaerobic layer (mmol m −3 ), k POC,i is the first order reaction rate coefficient (d −1 ), θ POC,i is the temperature coefficient, T is water temperature (°C), ω 2 is the burial velocity (m d −1 ), J POC is the depositional POC flux from the overlying water to the sediment (mmol m −2 d −1 ), and ƒ POC,i is the fraction of J POC that is in the ith G class.The aerobic layer is not included, due to its small depth relative to the anaerobic layer: H 1 ≈ 0.1 cm, while H 2 ≈ 10 cm.Particulate nitrogen (J PON ), phosphorus (J POP ), and silica (J PSi ) deposition is based on Redfield stoichiometry.The details of the model structure and processes for phosphorus, nitrogen, silica, and sulfur (Fig. 2) are reported elsewhere (Di Toro 2001, Brady et al. 2013, Testa et al. 2013).

Model setup, deposition time-series, and inclusion of oyster farm
We ran multiple simulation scenarios at each of the 3 experimental sites (7 simulations in total) to quantify the realized and potential effects of background and oyster culture-derived biodeposition rates (in cluding particulate biogenic C, N, and P) on sediments.Specifically, these scenarios represent different assumptions regarding the amount of POM that is incorporated into the sediment: (1) all of the material collected in off-bottom sediment traps was incorporated into sediments, (2) a fraction of this material is resuspended, and resuspension fluxes are subtracted from the sediment-trap rate (this simulation was only ran for the farm site where data were available), and (3) only material incorporated into sediments is simulated, which excludes POM that is removed from the site via all possible physical transport mechanisms, including advection and bedload transport.Scenario 3 is the optimized, SFM model-predicted, annual mean POM flux and assumes that the NH 4 + flux from the sediment is a good indicator of POM deposition after the model has accounted for diagenesis, nitrification, denitrification, burial, and mixing.The observed deposition of POM measured in sediment traps varied by season and station, where POM deposition was higher in the summer (especially June) and this peak was accentuated near the oyster floats as a result of increased oyster filtration and biodeposition rates.To accommodate this seasonal variability, the annual mean POM fluxes computed by SFM were transformed into seasonally-varying rates to represent the seasonal cycle of the depositional fluxes observed in the sediment traps.To achieve this, the interpolated monthly record was scaled so that the annual sum of the daily interpolated POM flux matched that of the POM depositional flux calculated in SFM.Initial conditions were determined by running the model for a 15 yr simulation period with a deposition record derived for the Choptank River by Brady et al. (2013).The model simulation period began in 1985 (after 15 yr of spin up) and ran through 2005, using the Hooke-Jeeves pattern search algorithm derived from a multi-year nutrient and O 2 flux record from a site close to the farm (38.6307°N, 76.1474°W).Since the farm was established in 2006, we simulated the seasonal deposition rates determined for the 3 stations (from 2011 observations) for each year during the 2006−2011 period.Thus, the total simulation period was from 1985− 2011.

RESULTS
Site-specific differences in organic matter deposition POM deposition was greatest within the farm site, and declined with distance from the farm (Fig. 3).At the farm site, modeled annual average particulate organic nitrogen (PON) deposition was 162.8 mmol N m −2 d −1 .This estimate includes the assumption that all PON collected in the off-bottom sediment traps was incorporated and processed in the underlying sediments (i.e.no resuspension correction).Under the same assumption, the near-farm site (~350 m from the closest oyster float) and reference site PON deposition comprised 48.3% (78.7 mmol N m −2 d −1 ) and 33.3% (54.2 mmol N m −2 d −1 ) of the farm deposition, respectively.By using the resuspension correction on the sediment trap estimates at the farm site, PON loading was reduced from 162.8 to 79.0 mmol N m −2 d −1 .Finally, if the Hooke-Jeeves algorithm (Brady et al. 2013) is used to back-calculate deposition based on observed NH 4 + flux, the resulting PON deposition would be 8.09, 4.08, and 1.92 mmol N m −2 d −1 at the farm, near-farm, and reference sites, respectively.Essentially, these model-based results imply that only 3−5% of the material caught in the off-bottom traps was incorporated into the sediment.However, the results also indicate that PON incorporation into the sediment at the farm site was 4.2 times greater than the reference site.

NH 4 + fluxes
The observed annual average NH 4 + fluxes at the farm, near-farm, and reference sites were 394, 127, and 21.8 µmol N m −2 h −1 , respectively.Only at the farm site did it appear that sediment -water NH 4 + fluxes increased substantially following the implementation of oyster culture (Fig. 4 + fluxes and denitrification rates (Fig. 5).Observed NO 3 − fluxes were −19.6, −4.08, and −6.33 µmol N m −2 h −1 at the farm, near-farm, and reference sites, respectively.The corresponding modeled fluxes were −4.96, 3.89, and 2.76 µmol N m −2 h −1 .Although the mean error was −10.6 µmol N m −2 h −1 , the model captured much of the dynamics of the system as evidenced by a reliability index of 1.33 and model-data correlation coefficient of 0.79 (Table 3).The model correctly captured the relative rankings of the NO 3 − fluxes at the 3 sites, as well as the seasonal signal within each site.For example, if we consider the near-farm site, the combination of higher PON deposition (Fig. 3) relative to the reference site and higher aerobic layer depth and nitrification rate relative to the farm site resulted in the highest NO 3 − fluxes at the near-farm site (as reflected in the observations and the model simulations).It is worth noting that in our simulations here, we re duced the sediment nitrification reaction velocity (Di Toro 2001) from 0.13 to 0.1 m d −1 to yield the optimal model-data agreement.

Sediment oxygen demand
Annual average observed sediment oxygen demand (SOD) at the farm, near-farm, and reference sites was 51.8, 37.9, and 21.4 mmol O 2 m −2 d −1 , respectively.Model results compare favorably to the observations (Fig. 6), as evidenced by annual average modeled SOD of 68.6, 29.5, and 15.2 mmol O 2 m −2 d −1 at the same locations.The Reliability Index across all observations was 1.37 and the mean error only 3.89 mmol O 2 m −2 d −1 (Table 3).While PON deposition was adjusted to match NH 4 + flux, no calibration or ad justments were made to predict SOD.Model and observations support the potential of biodeposition to approximately double the sediment oxygen demand in sediments even though only 3−11% of the potential POM was processed in the sediment.The largest model-observation mismatch came at the farm site August sampling event.Although the overlying water column was not observed to go hypoxic, the model predicts a substantial sulfide build up associated with elevated sulfate reduction rates, with approx.50% higher sulfide concentrations in the anaerobic layer at the farm site compared to the reference site (data not shown).

Denitrification
We compared model estimates of sediment denitrification to observations of net sediment N 2 fluxes measured in dark incubations.The observed annual average (± SD) net N 2 fluxes (which we equate to denitrification) at the farm, near-farm, and reference sites were 55. 8 ± 20.8, 72.8 ± 21.4, and 56.6 ± 16.1 µmol N m −2 h −1 , respectively.Both modeled and observed N 2 fluxes were comparable in magnitude and seasonality across sites, with peak fluxes during June and minima in April and August.For example, modeled denitrification rates, averaged over the same period as the observations, were 57.4 ± 11.7 (mean ± SD) µmol N m −2 h −1 at the farm site and 80.1 ± 9.2 µmol N m −2 h −1 at the reference site.Unlike the NH 4 + fluxes (and also PO 4 3− and O 2 fluxes summarized below), modeled N 2 fluxes were similar in the 3 depositional scenarios, where N 2 fluxes were slightly higher under the lowest PON deposition rates (Fig. 7).Whilst denitrification usually follows an annual cycle that closely matches temperature in oxic environments, with rela tively deep O 2 penetration into sediments (i.e. during the period 2000−2005; Fig. 7 & 8), model computations and observations suggest that denitrification became limited in warm months (August) in the years when PON deposition rates were elevated following the initiation of aquaculture in 2006 (Fig. 7).Such re duced late-summer denitrification (particularly 2007− 2009) was associated with reduced aerobic layer depths (Fig. 8).Despite this seasonal alteration due to higher PON deposition rates, the overall annual magnitude of sediment denitrification did not change in response to the introduction of the aquaculture operation.

Aerobic layer depth
A striking difference between the different POM deposition scenarios is the depth of the modeled aerobic layer.Model simulations suggest that once aquaculture activities began in 2006, the aerobic layer depth decreased markedly, especially at the farm site (Fig. 8).For example, using SFM-derived estimates of deposition, aerobic layer depth at the farm and near-farm sites were ~50% and ~75% of the reference site aerobic layer depth, respectively.At all sites, the seasonal maxima in aerobic layer depth during winter months were particularly reduced (e.g. from 2 to <1 mm at the near-farm site) (Fig. 8).During summer, reductions in the aerobic layer depth occurred despite the absence of severe water-column O 2 depletion in the overlying water, where a YSI ® 6600 sensor deployed within the farm during August of 2011 never recorded O 2 concentrations <125 µM.

PO 4 3− fluxes
Observed PO 4 3− fluxes were relatively small (−10.8 to 9.8 µmol P m −2 h −1 ) throughout our study relative to fluxes typical of deep Chesapeake Bay habitats (e.g.30−100 µmol P m −2 h −1 at depths >10 m), but were consistent with values expected for shallow systems with relatively high O 2 in overlying water (Fig. 9).Although observed PO 4 3− fluxes, when averaged over the 4 sampling dates (April−September), were −1.4,−0.09, and −2.7 µmol P m −2 h −1 at the farm, near-farm, and reference sites, respectively, model-predictions averaged over the same period using previously determined solid-solute partitioning coefficients (Testa et al. 2013) were much higher (24.2, 8.5, 3.5 µmol P m −2 h −1 at the farm, near-farm, and reference sites, respectively).Because the oyster farm site may have iron-rich sediments with a higher capacity to bind PO 4 3− than typical Chesapeake Bay sediments, we increased the phosphorus partitioning coefficients from 300 to 400 l kg −1 in the aerobic layer and from 100 to 200 l kg −1 in the anaerobic layer.As a result, sediment-water PO 4 3− fluxes de creased 82, 88.6, and 98% at the farm, nearfarm, and reference sites, respectively, resulting in mean (± SD) fluxes of 4.5 ± 1.1, 0.98 ± 0.18, and 0.06 ± 0.06 µmol P m −2 h −1 , which were much closer to fluxes observed at the sites (Table 3, Fig. 9).

Farm impacts on nitrogen cycling
Upon comparing changes in nitrogen cycling due to proximity-to-farm associated differences in PON deposition, several key points can be made.Firstly, PON deposition to sediments decreased with in creasing distance from the farm; PON deposition was 420 µmol N m −2 h −1 at the farm relative to 75 µmol N m −2 h −1 at the reference site, resulting in comparably reduced diagenesis rates and sediment -water NH 4 + fluxes (Fig. 10).Secondly, although NO 3 − fluxes were minimal relative to NH 4 + , denitrification, and nitrification (Fig. 10), nitrification and denitrification were elevated at the reference and near-farm sites relative to the farm site.Consequently, the nitrogen recycling efficiency [NRE = NH 4 + flux/(NH 4 + flux + NO 3 − flux + N 2 flux) × 100] at the sites decreased markedly with distance from the farm, where nearly all of the N fluxes were derived from NH 4 + fluxes at the farm site, while NH 4 + was < 30% of all N fluxes at the reference site (Fig. 10).

Nitrogen budget at the farm site
By using the suite of simulations carried out for 2011, we generated a nitrogen budget of the farm site (Fig. 11).Perhaps the most striking feature of the budget is that if we consider the sediment trap measurements used to drive the model, background deposition and oyster biodeposition would potentially result in the deposition of 3714 µmol N m −2 h −1 , while model simulations suggest that only 425 µmol N m −2 h −1 would actually have been deposited to sediments.This suggests that 3289 µmol N m −2 h −1 was exported from the system via some mechanism(s) of horizontal transport under normal conditions and/or during storm events.This represents an export of 88.6% of the material potentially processed within the farm.Of the PON that was incorporated into the sediments, 80% was released back to the water column as NH 4 + , while only 10% was lost permanently via denitrification and 10% was buried (Fig. 11).Because this budget includes annual mean fluxes and discounts changes in storage within the sediment, slight imbalances in the budget exist (Fig. 11).2).This increase is almost certainly due to the filtering of ambient particulates by cultured oysters and subsequent excretion of feces and pseudofeces resulting in the sinking of biodeposits to sediments.Although this conclusion is unsurprising, the fact that observed POM deposition rates were 3 times larger at the farm than at a nearby reference site (~700 m to the southeast) reveals that POM originating from the farm does not accumulate substantially in the immediate surrounding areas.In addition, model results indicate that POC deposition at the reference site did not increase following the initiation of farming in 2006 (Fig. 3).These findings indicate that large amounts of POM originally concentrated within the farm are widely distributed over a short period.

DISCUSSION
Likely mechanisms for this substantial POM export include resuspension and subsequent horizontal trans port during wind events, added to normal tidal ad vective transport.Dominant ebb tides at this location transport POM northward, away from the farm and into the open Choptank River estuary (Fig. 1), where the material is effectively dispersed.Both wave ob servations and wave modeling (data not shown) indicate that winds of modest strength (5−10 m s −1 ) from the NW to the NE (relatively common) or similar winds from the SE (less common) are capable of generating large enough waves (0.5 m) to resuspend significant amounts of sediment in these shallow waters.None of the 24 h sediment trap deployments occurred during such a wind wave event, however.Thus, although the farm concentrates large amounts of POM that sink below the cages, there does not appear to be a large 'footprint' of the farm in terms of excessive POM accumulation in adjacent sediments (Table 1).
As a result of the sensitivity of this site to wave and tide-induced resuspension and transport, we calculated that only 11% of the organic material settling below the oyster cages is eventually processed within the sediment at the farm (Fig. 11).Marinetics oyster farm is thus situated in an ideal location for floating oyster aquaculture, as the physical setting (hydrodynamics and sediment transport) prevents the accumulation of large amounts of organic matter that otherwise would severely impact the sediments at the site (Figs. 4,6 & 8), as has been observed elsewhere (Weise et al. 2009).In other locations, suspended shellfish farms (e.g.mussels, oysters) located at sites with slower currents and less resuspension suffered measurable degradation (e.g.sulfide production, nutrient release) with high POM accumulation (Grant et al. 2005, Holyoke 2008).
Although we conclude that the impact of the oyster culture at the Marinetics farm site on sediments is substantially lower than the potential impact given retention of all oyster-derived POM on site, we did observe higher NH 4 + effluxes and O 2 influxes at the farm, consistent with comparable studies elsewhere (Giles et al. 2006).Model results suggest that O 2 influxes and NH 4 + effluxes began to substantially increase a year after aquaculture was initiated ( 2007), especially at the farm site where both model simulations and observations indicated that they were 3−10 times higher than the reference site.These patterns should be expected, given elevated respiration under conditions of elevated organic carbon availability.The farm NH 4 + efflux computed from the observed O 2 influx, assuming Redfield stoichiometry and a respiratory quotient of 1, was 367 µmol m −2 h −1 , which is close to the actual NH 4 + flux of 394 µmol N m −2 h −1 and implies that the majority of the N released during diagenesis at the farm is released back to the watercolumn, consistent with an elevated 'N recycling efficiency' (Fig. 10).In contrast, the NH 4 + efflux computed stoichiometrically at the reference site was 130 µmol N m −2 h −1 greater than observed, indicating that denitrification was removing N from the system.This latter pattern is also consistent with the fact that denitrification was similar across all sites (Fig. 5), thus removing a relatively larger fraction of inorganic N from the reference site, which had lower NH 4 + fluxes.Comparable sediment denitrification rates, but different NH 4 + and O 2 fluxes between reference and aquaculture sites have been observed elsewhere in Chesapeake Bay (Higgins et al. 2013).It is worth noting that our analysis considered only sedi ment denitrification, and does not consider N processes occurring within the aggregations of the cultured oysters themselves, where NH 4 + efflux, denitrification, and overall nitrogen cycling may be enhanced relative to adjacent sediments (Duarte et al. 2003, Kellogg et al. 2013).
The similar denitrification rates across the 3 varying POM deposition and NH 4 + /O 2 flux environments suggest feedbacks within the nitrogen cycle that prevent denitrification inhibition under increasingly anaerobic conditions.In many enriched, muddy Chesapeake Bay sediments, low oxygen penetration restricts nitrification and thus coupled nitrification-denitrification (Kemp et al. 1990).Although benthic macrofauna can elevate O 2 penetration in sediments (Mayer et al. 1995), none of the sediment cores collected at the farm site contained these organisms.SFM simulations revealed that the aerobic layer depth was substantially reduced as POM loading to sediments and sediment O 2 uptake increased (both spatially and temporally), but these declines in aerobic layer depth were not substantial enough to severely limit nitrification (Fig. 10) and sediment -water NO 3 − fluxes were relatively small across all sites.Although denitrification was reduced at the farm and reference sites in August when aerobic layer depth was at its seasonal minimum, even the model simulations forced with the largest POM deposition rates did not substantially alter computed denitrification rates.Denitrification rates likely re mained high because overlying water O 2 concentrations were always relatively high (>94 µM observed at the farm site over a 3 wk period in August 2011) in the simulations, allowing for the persistence of an aerobic layer and nitrification despite high POM deposition.If sediment oxygen uptake was as high as predicted under the highest POM deposition scenarios (200 mmol O 2 m −2 d −1 ), the O 2 stock in a 0.5 m water column (assuming 187 µM O 2 ) would be depleted in 11 h (also assuming minimal air-sea exchange).Assuming that maximum POM deposition is possible under very quiescent periods, these results suggest that the overlying water could have become hypoxic during dark hours at the farm and severely diminish the aerobic layer, limiting denitrification (Testa & Kemp 2012); however, the present study is lacking in these observations.Microphytobenthic production may also influence nutrient and oxygen cycling at the sites we visited in this study.Although this site is extremely shallow, the water column is highly turbid (k d = 0.7−3.9 from April to September), limiting the amount of light that can reach the bottom.Measurements of surface and bottom PAR indicate that the sediments at the farm site received between 12.5 and 148 µE m −2 s −1 during April to September (roughly 1−45% of surface light) and < 25% of surface light in June to August.This amount of light could support microphytobenthic algal production at the sediment surface, where sediment cores that were incubated in 225 µE m −2 s −1 of light for 3−4 h showed 15−200% declines in sediment O 2 uptake and 21−106% declines in sediment NH 4 + release, yet variable impacts on N 2 , and NO 3 − fluxes (data not shown).Thus, the sediments at the farm study site may release less N and consume less O 2 during periods where light reaches the bottom and if significant benthic algal growth occurred.Given that we simulated microphytobenthic growth, the SFM deposition rate would likely underestimate the true POM deposition, given that less of the N processed in the sediment would be realized as NH 4 + if benthic algal N uptake occurred.Given the high rates of resuspension and biodeposit sinking at the farm site, it is unlikely that significant benthic algal production occurred.In addition, the measured denitrification rates (9−115 µmol N m −2 h −1 ) were somewhat high for Chesapeake Bay given the observed NO 2 − plus NO 3 − concentrations in the overlying water (1−11 µM), and if benthic algal photosynthesis was high, it would likely out-compete nitrifying bacteria for NH 4 + and thus suppress coupled nitrification-denitrification (Risgaard-Petersen 2003).
PO 4 3− fluxes were also similar across all study sites and averages over April to September approached zero at all sites.On the one hand, this should be expected given a well-oxygenated water-column and persistent aerobic layer, which would support a highsorption capacity for PO 4 3− given abundant oxidized iron and manganese (Sundby et al. 1986).On the other hand, particulate organic phosphorus deposition was 4 times higher at the farm than at the reference site and therefore, some increases in PO 4 3− were expected.Therefore, sediments at all sites appear to retain substantial amounts of Fe-bound PO 4 3− .Our initial simulations used solid-solute partitioning coefficients that are characteristic of conditions for Chesapeake Bay (Testa et al. 2013), and these simulations predicted modest PO 4 3− efflux from sediments, while observations suggested zero flux or net influx.These dis crepancies were particularly apparent at the farm site (Fig. 9).The fact that modeled PO 4 3− fluxes better agreed with observations upon an increase in these partitioning coefficients (see 'Materials and methods') suggests that these sites (especially the farm) may have a higher effective sorption capacity than adjacent sediments.One explanation for this pattern may be that oysters, in filtering large amounts of water containing high suspended solids, effectively concentrate large amounts of inorganic material in the sediments below the site (sediment trap material was only 2% carbon).Thus, oysters appeared to be concentrating Fe-rich material and enhancing sorption locally, a finding consistent with oyster biodeposit studies on other Choptank River sediments (Holyoke 2008) suggesting that oysters may engineer their local environment.Many oyster culture operations result in the accumulation of shell material in sediments underlying the site (Langan et al. 2006), but we did not observe much shell in the sediments at the farm site.It is possible that an abundance of calcium carbonate could also enhance the sorption of phosphate or the precipitation of calcium-phosphate minerals (Millero et al. 2001).Although the exact mechanisms may be unknown, oyster-induced biogeochemical changes occurring within aquaculture sites may need to be incorporated into sediment biogeochemical models to adequately capture the dynamics of phosphorus and other elements.
The ability to derive realistic deposition magnitudes using a sediment flux model is a novel tool that should help aquaculturists understand both the physics of depositional transport and the potential biogeochemical perturbation of a site by aquaculture operations.Combining sediment trap deployments (to estimate both biodeposit deposition and resuspension) and model simulations (to estimate incorporation of biodeposits into sediments) provided a fairly comprehensive view of particle transport at the farm.Given a modest field program to measure watercolumn nutrient and oxygen concentrations and a few sediment -water NH 4 + flux rates, SFM could be used as a management and site-selection tool to predict the potential for adverse effects from shellfish farming.Additional benefits to this approach are the ability to simulate sediment oxygen uptake and denitrification reasonably well with minimal calibration.Residuals between observed and modeled PO 4 3− fluxes suggest a yet unincorporated sediment phosphorus process may be at work at aquaculture sites, including interactions between biodeposition, iron, manganese, and sulfur cycling (Holyoke 2008).In order to improve this tool, we recommend further studies that examine phosphorus sorption potential at shellfish aquaculture farms, the impact of short-term oxygen excursions on nutrient cycling, and the incorporation of coupled sediment -water column models into aquaculture site selection tools.For this latter tool, feedbacks associated with sediment nutrient-release, primary production, oyster growth, and subsequent biodeposition could be examined.
Fig. 1.(a) Map showing location of the farm, near-farm and reference sites in the Choptank River, (b) a tributary on the eastern shore of Chesapeake Bay.Contours represent local bathymetry relative to mean lower low water, and white circles represent locations of sediment trap deployments and sediment -water nutrient and oxygen exchange measurements Fig. 2. Flow diagram illustrating the state variables, transport processes, and reactions in the sediment flux model, including those for nitrogen, phosphorus, and silica (left panel) and carbon, sulfur, methane, and oxygen (right panel).PN, PC, PSi, and PP are particulate nitrogen, carbon, silica, and phosphorus, respectively, and ω 12 is the particle mixing rate between layers 1 and 2. Please refer to 'Materials and methods: sediment biochemical modeling' for parameter definitions Fig. 3. Time-series (1985−2011) of particulate organic carbon (POC) deposition to sediments at (a) reference, (b) near-farm, and (c) farm sites.Green and red shading indicates pre-and post-oyster farming periods, respectively.Red line: uncorrected raw sediment trap estimates, black line: farm site sediment trap estimates corrected for resuspension, blue line: deposition estimated by sediment flux model (SFM) by fitting deposition to the observed sediment -water NH 4 + fluxes

Fig. 6 .
Fig. 6.Time-series (2000−2011) of modeled (lines) and observed (red circles) O 2 fluxes at (a) reference, (b) near-farm, and (c) farm sites.Green and red shading indicates pre-and post-oyster farming periods, respectively.Red line: uncorrected raw sediment trap estimates, black line: farm site sediment trap estimates corrected for resuspension, blue line: deposition estimated by sediment flux model (SFM) by fitting deposition to the observed sediment -water NH 4 + fluxes Fig. 7. Time-series (2000−2011) of modeled (lines) and observed (red circles) N 2 fluxes at (a) reference, (b) near-farm, and (c) farm sites.Green and red shading indicates pre-and post-oyster farming periods, respectively.Red line: uncorrected raw sediment trap estimates, black line: farm site sediment trap estimates corrected for resuspension, blue line: deposition estimated by sediment flux model (SFM) by fitting deposition to the observed sediment -water NH 4 + fluxes

Table 3 .
Root mean square error (RMSE), reliability index (RI), mean error (ME), correlation coefficient (r), and relative error (RE) for model-data comparison of sediment -water NO 3