Ocean Circulation Causes Strong Variability in the Mid-Atlantic Bight Nitrogen Budget
Abstract
Understanding of nitrogen cycling on continental shelves, a critical component of global nutrient cycling, is hampered by limited observations compared to the strong variability on a wide range of time and space scales. Numerical models have the potential to partially alleviate this issue by filling spatiotemporal data gaps and hence resolving annual area-integrated nutrient fluxes. In this study, a three-dimensional biogeochemical-circulation model was implemented to simulate the Mid-Atlantic Bight (MAB) nitrogen budget. Model results demonstrate that, on average, MAB net community production (NCP) was positive (+0.27 Tg N/year), indicating net autotrophy. Interannual variability in NCP was strong, with annual values ranging between 0.19 and 0.41 Tg N/year. Along-shelf and across-shelf horizontal transport fluxes were the other dominant terms in the nitrogen budget and were primarily responsible for this NCP variability. The along-shelf current transported nitrogen from the north (0.65 Tg N/year) into the MAB, supplementing the nitrogen entering from terrestrial inputs (0.27 Tg N/year). However, NCP was highest in the year when total water volume transport and inorganic nitrogen input was strongest across the continental slope in the southern MAB, rather than when terrestrial inputs were greatest. Interannual variability of NCP appears to be linked to changes in the positions of the Gulf Stream and Slope Water Gyre. Overall, the strong spatiotemporal variability of the nitrogen fluxes highlights the importance of observations throughout all seasons and multiple years in order to accurately resolve the current status and future changes of the MAB nitrogen budget.
Key Points
- A biogeochemical-circulation model is used to quantify the nitrogen budget of the Mid-Atlantic Bight (MAB)
- Horizontal along-shelf and across-shelf fluxes dominate the spatiotemporal variability of net community production (NCP) in the MAB
- Highest NCP is found in a year when inorganic nitrogen entering from across the continental slope is high and terrestrial inputs are low
Plain Language Summary
Portions of the ocean adjacent to land masses play a particularly important role in global nutrient cycling; however, strong spatial and temporal variability in these shallow regions of the ocean make it difficult to quantify nitrogen fluxes from observations alone. Here we use a computer simulation to estimate the fluxes and transformations of inorganic and organic nitrogen in Mid-Atlantic U.S. coastal waters. The coastal circulation flows southward providing roughly two thirds of the inorganic nitrogen to this region, with the remaining third entering from rivers and estuaries. Nitrogen transport across the continental slope is highly variable, directed into the system in some years and out in others. The net community production of the system, that is, the conversion of inorganic to organic nitrogen through photosynthesis minus respiration, is also highly variable. This strong interannual variability is primarily due to the highly variable fluxes of inorganic nitrogen entering this region from across the continental slope where the Gulf Stream approaches the continental slope, rather than the variability in terrestrial inputs. Overall, the strong variability of this region highlights the importance of collecting observations throughout all seasons and during multiple years in order to accurately resolve coastal nutrient budgets.
1 Introduction
Until recently, global biogeochemical cycling studies have underestimated the importance of continental shelves because these regions make up a relatively small proportion of total sea surface area (~8%). However, evidence is accumulating that continental margins may contribute significantly to global carbon and nitrogen cycles (C. T. A. Chen et al., 2003; Jahnke, 2010; Muller-Karger et al., 2005; Najjar et al., 2018). In spite of the relatively small area of continental shelves, observational studies often lack the space and time coverage that is needed to obtain annual estimates of continental shelf nitrogen fluxes. Coupled physical-biological numerical models can provide the space and time resolution that is needed to compute continental shelf nutrient fluxes and budgets (e.g., Cahill et al., 2016; Fennel et al., 2006, 2008; Hofmann et al., 2011, 2008; Kishi et al., 2007; Sasai et al., 2006; Schrum et al., 2006; St-Laurent et al., 2017; Wakelin et al., 2012; Xue et al., 2013). Such models are particularly useful for coastal systems where nutrient budgets are significantly impacted by anthropogenically changing terrestrial nutrient inputs and spatially and temporally varying circulation patterns.
The connection of coastal shelf waters to terrestrial and riverine inputs and oceanic waters provides an integration across systems that is critical in biogeochemical cycling and nutrient budgets. Increasing utilization of anthropogenic fertilizers and management actions in coastal watersheds have caused changes in riverine input of nitrogen to coastal waters over the last century (Galloway et al., 2004; Yang, Tian, Friedrichs, Hopkinson, et al., 2015). Modified ocean circulation due to mesoscale events (Mahadevan & Archer, 2000; McGillicuddy et al., 1998; Oschlies & Garcon, 1998; Siegel et al., 1999; S. Zhang et al., 2018) and/or climate change (Caesar et al., 2018; Ezer, 2015; Saba et al., 2016) also has the potential to impact nutrient inputs and cycling in coastal waters. Such changes in nitrogen inputs impact continental shelf net community production (NCP), defined here as the difference between the generation of organic nitrogen through photosynthesis and the conversion of organic nitrogen to inorganic nitrogen through respiration and remineralization processes. Nitrogen inputs to coastal systems can also be denitrified, buried, or advected to the open ocean. Quantifying each of these nitrogen fluxes, and their temporal and spatial variability, is critical to gaining a better understanding of both regional and global biogeochemical cycles.
The objective of this study is to quantify the spatiotemporal variability of key nitrogen fluxes in the Mid-Atlantic Bight (MAB) by implementing a three-dimensional coupled biogeochemical-circulation model. This region (Figure 1) is of particular interest because of the effect of interannual variability of physical processes (S. Zhang et al., 2018; W. G. Zhang & Gawarkiewicz, 2015), including climate-related changes in precipitation, temperature, and circulation (K. Chen et al., 2015; K. Chen & Kwon, 2018; Ji et al., 2007; Schulte et al., 2018) on biogeochemical cycling. This area is remotely affected by two large-scale currents (Linder & Gawarkiewicz, 1998; Linder et al., 2006): the northward flowing, warm, and low-nutrient Gulf Stream current and the southward flowing, cold, and nutrient-rich Labrador slope current, as well as the cyclonic Slope Water Gyre that occurs in between these two currents (Figure 1). This study builds on earlier efforts to better understand nitrogen fluxes in the MAB (e.g., Fennel, 2010) by forcing the coastal ocean model with inputs from terrestrial (Tian et al., 2015; Yang, Tian, Friedrichs, Hopkinson, et al., 2015; Yang, Tian, Friedrichs, Liu, et al., 2015) and atmospheric nitrogen models (St-Laurent et al., 2017) and including dissolved organic nitrogen (DON) cycling as well as burial and resuspension (Druon et al., 2010).
This paper is structured as follows. Section 2 first describes the numerical models and nitrogen budget calculations, and then section 3 examines the results of a 5-year simulation of the MAB. These results describe the following: (1) the 5-year averaged nitrogen fluxes, (2) their seasonal variability, (3) their spatial variability, and (4) their interannual variability. Section 4 discusses these results in the context of past nitrogen flux studies of the MAB, examines the mechanisms causing the strong spatiotemporal variability of the NCP, and provides suggestions for future work. Lastly, the primary findings are briefly summarized in section 5.
2 Methods
2.1 Biogeochemical Ocean Circulation Model
The 3-D circulation model employed in this study is based on the Regional Ocean Modeling System (ROMS, Shchepetkin & McWilliams, 2005), a modeling system widely used for shelf circulation and coupled physical-biological applications (e.g., Cahill et al., 2016; Druon et al., 2010; Fennel et al., 2006; Hermann et al., 2009; He et al., 2011; Hofmann et al., 2011, 2008; St-Laurent et al., 2017; Xue et al., 2013; Wilkin, 2006). As in Hofmann et al. (2011), Cahill et al. (2016), and St-Laurent et al. (2017), ROMS is configured for the U.S. eastern continental shelf (Figure 1a) with a horizontal resolution of ~9 km and run for the mid-2000s (2004–2008 in this case). Bulk formulae (Fairall et al., 2003) are used to compute air-sea fluxes using air temperature, pressure, humidity, and winds from 3-hourly National Center for Environmental Prediction reanalysis fields (Mesinger et al., 2006). Physical and biological tracers are advected using the Multidimensional Positive Definite Advection Transport Algorithm (Smolarkiewicz & Margolin, 1998). Subgrid-scale vertical diffusivity is estimated using the parameterization from Mellor and Yamada (1982). All model formulations and parameters are identical to those described in St-Laurent et al. (2017) except for horizontal diffusion of temperature and salinity, which was increased slightly to improve simulations of the physical fields.
This circulation model is coupled to the USECoS (U.S. Eastern Continental Shelf) biogeochemical model that includes prognostic equations for two size classes of phytoplankton (P) and zooplankton (Z), semilabile and refractory DON, nitrate, ammonium, oxygen, and small and large carbon and nitrogen detrital pools. Although the model includes the full inorganic and organic carbon cycle as well (see St-Laurent et al., 2017, supporting information for detailed equations), it does not influence the nitrogen budget and thus is not discussed here. The degree of biogeochemical complexity (i.e., number of plankton size classes) chosen for this model was based on the model intercomparison results of Xiao and Friedrichs (2014a). By assimilating satellite-derived particulate organic carbon and size-fractionated chlorophyll concentrations, Xiao and Friedrichs (2014b) demonstrated that this “2P2Z” model had the highest level of model skill compared to simpler models with single phytoplankton and zooplankton size classes and compared to a more complex model with three phytoplankton state variables. Dissolved organic carbon (DOC) and DON represent key components of the nitrogen and carbon budget (Carlson et al., 2010; Fasham et al., 1999; Hansell & Carlson, 2001) yet are frequently not included in continental shelf nutrient cycle simulations (Fennel et al., 2006, 2008; He et al., 2011; Xue et al., 2013). As in Druon et al. (2010), DOC and DON are partially decoupled in order to allow for variations of C:N ratios due to preferential degradation of DON or C:N ratios of different sources (Hopkinson & Vallino, 2005). Full equations of the USECoS model are provided in the supporting information of St-Laurent et al. (2017).
An important nutrient source for the MAB is from the land, and this terrestrial linkage was included as part of the model setup through inputs from the 31 rivers included in the model domain. As in Feng et al. (2015) and St-Laurent et al. (2017), daily riverine freshwater and nutrient inputs are derived from the Dynamic Land Ecosystem Model (DLEM; Tian et al., 2010). DLEM is a process-based mechanistic terrestrial model, which provides explicit flux estimates of freshwater (Yang, Tian, Friedrichs, Liu, et al., 2015), carbon (Tian et al., 2015), and nitrogen (Yang, Tian, Friedrichs, Hopkinson, et al., 2015) from terrestrial ecosystems to river networks and eventually to the estuaries and coastal oceans of the U.S. eastern continental shelf. These fluxes are simulated through generation of surface runoff after precipitation events, leaching of these materials to river networks, transformation of carbon and nutrients in rivers and lakes, and transport from upstream areas to coastal regions. Extensive skill assessment with observations from the U.S. Geological Survey (Tian et al., 2015; Yang, Tian, Friedrichs, Hopkinson, et al., 2015; Yang, Tian, Friedrichs, Liu, et al., 2015) shows that DLEM successfully captures seasonal and interannual variability in fluxes of freshwater, nitrogen, and carbon to the coastal zone of the eastern United States. In the Chesapeake Bay watershed, DLEM skill is similar to that of the Chesapeake Bay Watershed Model (Shenk & Linker, 2013) in terms of reproducing observed hydrological transports and loads of nitrate (Yang, Tian, Friedrichs, Hopkinson, et al., 2015; Yang, Tian, Friedrichs, Liu, et al., 2015).
Another important, yet often neglected, source of nitrogen to the coastal ocean is from atmospheric deposition. During the summer in the Gulf Stream region, atmospheric nitrogen deposition can increase surface nitrogen concentrations by 14% and increase new surface primary production by 22% (St-Laurent et al., 2017). In the Chesapeake Bay, atmospheric deposition has the same gram-for-gram impact on low oxygen concentrations (hypoxia) as riverine nitrogen loading (Da et al., 2018). Here atmospheric deposition estimates were obtained from the U.S. Environmental Protection Agency's Community Multiscale Air Quality (CMAQ) model (Appel et al., 2011; Gantt et al., 2015), as in St-Laurent et al. (2017), and are provided to the ocean model as 3-hourly fluxes of nitrate, ammonium, and DON to the ocean surface.
Boundary and initial conditions were similar to those used in St-Laurent et al. (2017). The open boundary conditions for the physical fields were obtained from 3-day averages of the Hybrid Coordinate Ocean Model North Atlantic assimilative model (Chassignet et al., 2007). The boundary conditions for nitrate and oxygen concentrations were derived from the World Ocean Atlas monthly climatology (Garcia et al., 2014a, 2014b), while boundary conditions for other biological variables were set to constant small values throughout the simulation period. These approximations do not deleteriously affect simulations within the MAB area since all boundaries were placed far from the study area. The initial conditions for the simulation were derived by repeatedly running the model over the period 2004–2008 (with the December 2008 state used to reinitialize the model on January 2004). As documented in St-Laurent et al. (2017), model skill has been extensively assessed based on nutrient concentrations from the World Ocean Atlas (Garcia et al., 2014a) and surface chlorophyll concentrations from the Sea-Viewing Wide Field-of-View Sensor (National Aeronautics and Space Administration, 2014).
2.2 Nitrogen Budget Calculations
(1)
(2)
(3)3 Results
3.1 Model Estimates of the 5-Year Mean Nitrogen Budget
The distributions of the various forms of nitrogen within the MAB provide insights into variability of the nitrogen fluxes. Surface concentrations of DIN (defined as the sum of the modeled nitrate and ammonium state variables), DON (defined as the sum of the semilabile DON and refractory DON state variables) and PON (defined as the sum of the phytoplankton, zooplankton, and detrital state variables) were higher within the MAB than in waters located off the continental shelf (Figures 2a, 2c, and 2e). Concentrations were particularly high where estuaries and rivers flowed to the shelf, such as in the vicinity of the Chesapeake Bay, Delaware Bay, and Long Island Sound plumes. Concentrations of DON were high throughout most of the MAB (Figure 2c), whereas high concentrations of DIN were restricted closer to the shore (Figure 2a). Throughout much of the MAB, surface concentrations of PON were approximately half that of DON (Figures 2c and 2e). At 100-m depth, concentrations of DIN (Figure 2b) were high in the region north of the Gulf Stream and in the Slope Water Gyre south of Cape Cod (Figure 1b), due to shoaling isopycnals north of the Gulf Stream that upwell deep nitrate-rich waters (Csanady & Hamilton, 1988). In contrast, DON and PON concentrations were relatively low at 100 m (Figures 2d and 2f). Neither DON nor PON concentrations show a gradient across the Gulf Stream at depth (Figures 2d and 2f), whereas DIN concentrations throughout the water column show a strong gradient across the Gulf Stream with higher concentrations to the north than to the south, (Figures 2a and 2b).
Temporal averaging of equation 2 provides an estimate for the 2004–2008 mean inorganic nitrogen budget (Figure 3). Over this 5-year period, DIN sources (0.39 Tg N/year) essentially balance DIN sinks (0.37 Tg N/year). Most of the DIN source (85%) is from horizontal transport (E&RDIN + NEDIN) with the northeast transport dominating, and the remaining 15% coming from atmospheric deposition. NCP is positive, indicating overall net autotrophy of the MAB, and accounts for 73% of the DIN sink, with the remaining losses split roughly equally between denitrification and horizontal transport across the shelf break (CSDIN). Hence, the first-order DIN balance is one of horizontal transport inputs balancing NCP.
Horizontal transport and NCP also play dominant roles in the 2004–2008 mean organic nitrogen budget (Figure 3). Of the TON sources, which total 0.87 Tg N/year and balance the TON sinks, horizontal transport from the northeast (NETON) and NCP account for 51% and 31%, respectively. Horizontal transport from estuaries and rivers (E&RTON) accounts for most of the remaining 18%, with a very small role played by atmospheric deposition (St-Laurent et al., 2017). Nearly all (95%) of the TON sink is due to cross-shelf transport with burial accounting for the remaining 5%. Even though PON concentrations are much lower than DON concentrations at the northeast and continental shelf boundaries (Figure 2), PON and DON contribute roughly equally to transport across these boundaries.
Quantifying the water budget of the MAB is helpful for interpreting the nitrogen budget (Table 1). On average over the 5 years analyzed, water enters the MAB via estuaries and rivers (~7%) and from the northeastern along-shelf current (~93%) and flows out across the continental slope (Table 1). Although precipitation, evaporation, and changes in the average volume of water within the MAB cause small residuals between the horizontal transport sources and sinks on an annual basis (generally <2 mSv), over the 5-year simulation the amount of water entering from the landward side of the MAB (E&R = +6 ± 1 mSv) roughly balances the sum of the fluxes across the northeast and continental slope boundaries (NE + CS = −7 ± 2 mSv; Table 3). This is because the CS and NE fluxes are typically large and have opposite sign in each of the 5 years analyzed. On average, the flux across the continental slope (CS = −87 ± 60 mSv) is 7 mSv larger than that entering from the northeast (NE = 80 ± 59 mSv). The large standard deviations on these estimates represent the significant interannual variability (see section 3.4) resulting from episodic events such as warm core rings interacting with the continental shelf (K. Chen et al., 2014). This general circulation pattern follows the concept of a “leaking” MAB shelf (Lozier & Gawarkiewicz, 2001) characterized by a decrease in along-shelf flow from the northeastern boundary of the MAB to Cape Hatteras (Levin et al., 2018). The strong interannual variability of both the along-shelf and the cross-shelf transports is also consistent with those determined from previous studies (e.g., K. Chen & He, 2015).
| Year | Estuaries & rivers (E&R) | Northeast (NE) | Continental slope (CS) | Ocean flux (NE + CS) |
|---|---|---|---|---|
| 2004 | +7 | +93 | −100 | −7 |
| 2005 | +6 | +154 | −160 | −6 |
| 2006 | +7 | +64 | −73 | −9 |
| 2007 | +5 | −7 | +4 | −3 |
| 2008 | +6 | +97 | −105 | −8 |
| 2004–2008 Mean | +6 | +80 | −87 | −7 |
| ± Std. Dev. | ±1 | ±59 | ±60 | ±2 |
3.2 Seasonal Variability of Nitrogen Fluxes
The nitrogen budget terms showed varying levels of seasonal variability throughout the year (Figure 4). Atmospheric deposition was the least seasonally variable, with fluxes varying by only about 0.02 Tg N/year. Burial and denitrification both demonstrated a larger seasonal cycle, with highest fluxes in the spring when primary production is high and lowest fluxes in winter (Figures 4a, 4c, and 4e). As described above, nitrogen input from the land was the smallest of the three horizontal transport fluxes (Figure 3); however, some seasonal variability is still apparent (Figures 4b, 4d, and 4f). Organic nitrogen input from the estuaries and rivers to the MAB is greatest in the summer, which is when the greatest productivity of organic matter within the coastal estuaries typically occurs (Harding et al., 2002). In contrast, input of DIN to the MAB from the estuaries and rivers is typically least in summer, since during this time of year (1) DIN is being rapidly taken up inside the estuaries and (2) river runoff is typically low.
Seasonal variability was particularly prominent in the along-shelf and across-shelf physical horizontal fluxes (Figures 4b, 4d, and 4f). Specifically, horizontal fluxes of both organic and inorganic nitrogen from the NE were highest in spring, which is also when the along-shelf water transport is greatest (Figure 4g). From September to December, the fluxes across the northeastern MAB boundary were smallest and even reversed direction (flowing out of the MAB) during some of this time, again consistent with the water transports across this boundary (Figure 4g). It is important to note, however, that this seasonal variability in along-shelf and across-shelf transport varies considerably from year to year, making it difficult to assess the skill of model results with observations such as those of Lentz (2008a, 2008b). The seasonality in TN flux across the continental slope roughly mirrored that across the northeastern MAB boundary (Figure 4b). Organic nitrogen was directed across the shelf and out of the MAB all year long with the highest flux being in spring (Figure 4f). The flux of inorganic nitrogen across the continental slope was highly variable throughout the year (Figure 4d) and did not always follow the organic nitrogen or water transport fluxes (Figure 4g). This overall seasonal variability can be put into perspective by comparing the results shown in Figure 4 to the interannual variability (standard deviations) presented in Tables 1-3.
| Year | Horizontal fluxa | Atmospheric deposition | Burial | Denitrification | NCP | ∂/∂t | |||
|---|---|---|---|---|---|---|---|---|---|
| DIN | DON | PON | DIN | DON | PON | DIN | TON | TN | |
| 2004 | +0.01 | −0.10 | −0.12 | +0.06 | +0.01 | −0.04 | −0.06 | +0.26 | −0.24 |
| 2005 | +0.28 | −0.09 | −0.08 | +0.05 | +0.01 | −0.03 | −0.05 | +0.21 | +0.07 |
| 2006 | +0.26 | −0.08 | −0.09 | +0.07 | +0.01 | −0.04 | −0.06 | +0.19 | +0.07 |
| 2007 | +0.38 | −0.19 | −0.20 | +0.06 | +0.01 | −0.04 | −0.06 | +0.41 | −0.05 |
| 2008 | +0.49 | −0.12 | −0.15 | +0.06 | +0.01 | −0.05 | −0.06 | +0.29 | +0.18 |
| 2004–2008 Mean | +0.28 | −0.12 | −0.13 | +0.06 | +0.01 | −0.04 | −0.06 | +0.27 | +0.006 |
| ± Std. Dev. | ±0.18 | ±0.04 | ±0.05 | ± 0.01 | ±0.001 | ±0.01 | ±0.007 | ±0.09 | ±0.16 |
- Note. The area of the MAB as defined here is 8.5714 × 104 km2. DIN = dissolved inorganic nitrogen; DON = dissolved organic nitrogen; PON = particulate organic nitrogen; NCP = net community production, TON = total organic nitrogen..
- a Horizontal flux includes all three terms listed in Table 3.
| Year | Estuaries and rivers (E&R) | Northeast (NE) | Continental slope (CS) | ||||||
|---|---|---|---|---|---|---|---|---|---|
| DIN | DON | PON | DIN | DON | PON | DIN | DON | PON | |
| 2004 | +0.15 | +0.05 | +0.07 | +0.40 | +0.22 | +0.15 | −0.54 | −0.37 | −0.34 |
| 2005 | +0.13 | +0.05 | +0.08 | +0.30 | +0.39 | +0.29 | −0.15 | −0.53 | −0.45 |
| 2006 | +0.12 | +0.05 | +0.12 | +0.16 | +0.20 | +0.22 | −0.02 | −0.33 | −0.43 |
| 2007 | +0.08 | +0.05 | +0.09 | −0.16 | +0.05 | +0.09 | +0.46 | −0.29 | −0.38 |
| 2008 | +0.10 | +0.05 | +0.12 | +0.33 | +0.30 | +0.28 | +0.06 | −0.47 | −0.55 |
| 2004–2008 Mean | +0.12 | +0.05 | +0.10 | +0.21 | +0.23 | +0.21 | −0.04 | −0.40 | −0.43 |
| ± Std. Dev. | ±0.03 | ±0.00 | ±0.02 | ±0.22 | ±0.13 | ±0.09 | ±0.36 | ±0.10 | ±0.08 |
- Note. DIN = dissolved inorganic nitrogen; DON = dissolved organic nitrogen; PON = particulate organic nitrogen.
The NCP flux showed the largest seasonal amplitude. This term, representing the conversion of inorganic nitrogen to organic nitrogen (Figures 4d and 4f), was greatest during the spring bloom period and least in the time period directly following the spring bloom, which is a time of rapid recycling. A smaller positive peak of NCP occurred in fall (September) and was again followed by a period of net heterotrophy (negative NCP) from October to December (Figure 4f). The seasonal variability of NCP was largely responsible for the seasonality of the rate of change terms, particularly for organic nitrogen, that is, NCP ≈ ∂TON/∂t (Figure 4f).
3.3 Spatial Variability of Nitrogen Fluxes
Atmospheric deposition, NCP, burial, and denitrification varied spatially throughout the study region (Figure 5). Of these flux terms, atmospheric deposition was the least spatially variable within the MAB (Figure 5a), though farther offshore a high band of atmospheric deposition occurs over the Gulf Stream due to very high precipitation in that region (St-Laurent et al., 2017). In contrast, NCP was highly variable within the MAB, with the highest values occurring in the south and along the continental slope and lower values occurring in the midshelf portion of the MAB (Figure 2b). High NCP also occurred northeast of the MAB near Cape Cod (Figure 1b), while a region of low NCP occurred just southwest of this region in the northernmost portion of the MAB. Like NCP, the spatial variability of burial and denitrification within the MAB (Figures 5c and 5d) was also much larger than that of atmospheric deposition. The largest fluxes generally occurred in the shallowest waters closest to shore. Regions of higher burial and denitrification generally occurred where resuspension was low and/or where NCP was high (Figure 5b).
The 5-year mean horizontal transport fluxes of DIN (Figure 6a) and TON (Figure 7a) into and out of the MAB were also characterized by strong spatial variability. The flux of nitrogen from the estuaries and rivers was, as expected, largest near the outflow of major coastal water bodies, such as Pamlico Sound, Chesapeake Bay, Delaware Bay, and Long Island Sound (Figure 1b) and small in between (Figures 6a and 7a). The nitrogen entering the MAB from the northeast occurred over a much shorter along-boundary distance and showed less spatial variability across this boundary. Along the continental slope (CS), DIN entered from the south near the confluence of the Gulf Stream and the Slope Water Gyre (36.5°N, 74.5°W; Figure 6a) and exited the MAB along the continental slope north of the Slope Water Gyre. Horizontal transport fluxes of TON were directed out of the MAB across almost the entire continental slope (Figure 7a). The one exception was the small flux of organic matter into the MAB in the southern portion of the domain where the Gulf Stream comes closest to the outer shelf before turning out to the open ocean. In general, the transport of nitrogen into the MAB from the rivers and from the northeast and out of the MAB across the continental shelf was dictated by the total volume water transport of the system (Figure 8a).
3.4 Interannual Variability of Nitrogen Fluxes
Some of the terms in the nitrogen budget demonstrated strong interannual variability, while other terms showed much less interannual variability (Figure 9). The differences between the maximum and minimum annual burial, denitrification, and atmospheric nitrogen deposition fluxes were all very small (~0.02 Tg N/year; Table 2). In comparison, the flux of nitrogen from the estuaries and rivers was more variable, with the lowest flux occurring in 2007 (E&RTN = 0.22 Tg N/year), and the highest occurring in 2006 (E&RTN = 0.29 Tg N/year; Figure 9 and Table 3), and with the variability being much greater for PON and DIN than for DON (Table 3). Although the total E&R flux varied from year to year, the locations of greatest flux stayed relatively constant (Figures 6 and 7). Specifically, the locations of greatest export in each year continued to be in the vicinity of the outflow of major coastal water bodies, such as Pamlico Sound, Chesapeake Bay, Delaware Bay, and Long Island Sound.
The horizontal advective flux from the northeast and across the continental slope was considerably more variable than the flux from estuaries and rivers. This large variability in the NETN + CSTN flux (−0.22 ± 0.16 Tg N/year) was primarily driven by the highly variable DIN flux across the continental slope, (CSDIN = −0.04 ± 0.36 Tg N/year), which overall was directed seaward in some years (2004–2006) and landward in others (2007 and 2008; Table 3 and Figure 6). Specifically, as the convergence region between the Gulf Stream and Slope Water Gyre migrated north and south from one year to another, the location of maximum flux of DIN into the MAB along the continental slope also shifted north and south (Figures 6b–6f). This location was farthest north in 2007 due to the presence of a Gulf Stream meander in the region (Figure 6e), which was also the year with the greatest total flux of DIN into the MAB along the continental slope (Table 3). Although the flux of DIN across the northeastern boundary of the MAB was generally directed into the MAB, 2007 was an anomalous year in that the net flux of DIN across this boundary was directed out of the MAB. In terms of water flux (Table 1 and Figure 8), 2007 was also an anomalous year, in that it was the only year in which the net water transport was directed out of the MAB along the northeastern boundary from the northeast and into the MAB across the continental slope. Although the individual CS and NE fluxes were anomalous in 2007 for both DIN and water transport, the overall flux was very similar in 2007 (CSTN + NETN = −0.23 Tg N/year) to the 5-year mean (CSTN + NETN = −0.22 Tg N/year; Table 3).
NCP also varied considerably throughout the 5 years (NCP = 0.27 ± 0.09 Tg N/year; Figure 8 and Table 2). The annual mean NCP was highest in 2007 (>0.4 Tg N/year). Interestingly, this was the year of lowest terrestrial input to the MAB but greatest input of DIN across the continental slope (Table 3). The net horizontal flux of organic nitrogen out of the MAB was also largest in 2007 (Table 2). Analogously, the mean annual NCP was lowest in 2005 and 2006, which was also the year of lowest horizontal flux of organic nitrogen out of the MAB (−0.17 Tg N/year in both years.)
4 Discussion
4.1 Comparison of Modeled Nitrogen Budget to Prior Estimates
The modeled nitrogen flux terms computed for the 5-year mean (Figure 3) are generally consistent with previous observations and model simulations, given the 5-year standard deviations of the model estimates (Tables 2 and 3). This agreement is particularly noteworthy since (1) in highly variable areas like the MAB observations with sufficient spatial and temporal coverage to estimate annual flux estimates for the full region are lacking and (2) models are inherently associated with significant errors due to uncertain parameterizations. The simulated burial rates estimated over the full MAB area (8.5714 × 104 km2) yield an annual mean areal burial rate of 0.47 g N·m−2·year−1. Assuming a Redfield C:N ratio, this rate is highly consistent with the carbon burial rate reported in Najjar et al. (2018) for the MAB (2.8 g C·m−2·year−1 = 0.49 g N·m−2·year−1). Denitrification rates vary substantially throughout the MAB (Figure 5d; Laursen & Seitzinger, 2002; Seitzinger & Giblin, 1996). The rate estimated in this study (~0.1 Tg N/year) was lower than the model results of Fennel (2010) for 2004–2005 (~0.4 Tg N/year), a difference that may result from the more sophisticated sediment boundary conditions applied in this study that account for resuspension and burial (Druon et al., 2010). Also, Fennel (2010) did not include DON or a second smaller class of phytoplankton as part of the model structure.
Horizontal transport fluxes across the MAB boundaries are difficult to estimate from observations alone due to their strong spatiotemporal variability and are thus most frequently estimated by modeling studies or mass balance approaches. For example, using an estuarine mass balance approach, Herrmann et al. (2015) estimated that 1.1–1.3 Tg C/year of organic carbon entered the MAB from estuaries. When converted with a Redfield ratio, this yields 0.19–0.23 Tg N/year, corresponding reasonably well with the estimate obtained here (0.15 ± 0.05 Tg N/year; Figure 3). Using a combined remote sensing and neural network approach for the years 2010–2012, Mannino et al. (2016) estimated the DOC flux from estuaries and rivers to the MAB as 0.8 Tg C/year. Assuming a C:N ratio for DOC of 15 mole C: mole N (Hopkinson & Vallino, 2005; Mannino et al., 2016), this yields a DON flux of 0.06 Tg N/year, which is similar to the estimate computed here of 0.05 Tg N/year (Table 3). Mannino et al. (2016) also computed a net outward horizontal flux of DOC of 2.6 Tg C/year for the MAB. Again converting this to DON using the C:N ratio of 15 mole C: mole N yields a net flux out of the MAB of −0.20 Tg N/year. This is only slightly larger than the 5-year mean (2004–2008) computed here by summing the transport in from the northeast, estuaries and rivers, and out across the continental slope (−0.12 ± 0.04 Tg N/year; Table 2). Considering the strong interannual variability of these flux estimates, which range from a minimum in 2006 (−0.08 Tg N/year) to a maximum in 2007 (−0.19 Tg N/year; Table 2), and the very different approaches taken in deriving these estimates, the agreement between them provides increased confidence in these results.
The horizontal transport fluxes derived here are also similar to those of Fennel (2010), which used a similar physical model but a simpler biogeochemical model (no burial or resuspension, no dissolved organic matter, and only one phytoplankton size class) and used U.S. Geological Survey information directly without linking to a sophisticated terrestrial ecosystem model. Despite these differences, their estimate of TN entering from the estuaries and rivers (+0.22 Tg N/year) is consistent with that estimated here (+0.27 Tg N/year), especially given that their estimate did not include input of DON (~0.05 Tg N/year; Table 3). The modeled estimates for horizontal transport fluxes into and out of the MAB (Figure 3 and Table 2) also compare reasonably well to those in Fennel (2010). The net horizontal fluxes of DIN and PON calculated by Fennel (2010), 0.41 and −0.07 Tg N/year, respectively, were similar to those found in this study (0.28 ± 0.18 and −0.13 ± 0.05 Tg N/year; Table 2), especially considering that different years were analyzed in the analyses. The similarity suggests that, in spite of large interannual variability and differences in model structure, the basic processes controlling these fluxes are well represented in these models.
Long-term averaged estimates of net ecosystem production are almost nonexistent for the MAB; however, the modeled annual total production of organic nitrogen (and carbon) through photosynthesis (Figure 10a) was found to show a reasonable agreement to values in the literature. The modeled estimates of primary production obtained from this study (3.7 ± 0.5 Tg N/year; 22 ± 3 Tg C/year) were lower than historical estimates (30 Tg C/year) based on National Marine Fisheries Surveys conducted in the late 1970s and early 1980s (Najjar et al., 2018; O'Reilly et al., 1987). However, the estimate derived here is similar to estimates based on satellite productivity algorithms for the specific years analyzed here (K. Hyde, personal communication July 23, 2018; 24 Tg C/year). In addition, the model estimated areal primary production rate at the inner shelf (21 mol C·m−2·year−1) was within the range of primary production rates (6–45 mol C·m−2·year−1) observed in the southern MAB (Filippino et al., 2011).
4.2 Spatiotemporal Variability of NCP
This study is a first attempt to quantify the spatiotemporal variability of NCP, which is a critical term in the nitrogen budget. The model simulations show that NCP is highly variable over the MAB shelf, with the greatest rates occurring in the south (Figure 5b), where the Gulf Stream approaches the MAB boundary and intrusions of deep high nutrient water (S. Zhang et al., 2018) stimulate high productivity (Figure 10a). In general, NCP was higher along the continental slope than in the midshelf portion of the MAB (Figure 2b), again due to higher productivity stimulated by deep nutrient entrainment into the MAB euphotic zone via isopycnal shoaling along the shelf break (Ryan et al., 1999; Schollaert et al., 2004; S. Zhang et al., 2018). The resulting enhanced biological productivity at the shelf break is thus a subsurface feature and not evident from satellite-derived surface productivity estimates (Kuring et al., 1990), which tend to decrease monotonically away from the coast.
The simulated spatial distributions of productivity and respiration (Figures 10a and 10b) each show similar patterns, but with opposite signs. This indicates that for the most part, organic nitrogen generated through primary production is respired locally. Exceptions exist, such as the region of high NCP associated with the relatively shallow topography south of Cape Cod (Figures 1b and 5b), an area known for strong upwelling and high productivity (Blanton et al., 1981; W. G. Zhang & Gawarkiewicz, 2015). The simulated patterns show this is a region of high productivity (Figure 10a) but low respiration (Figure 10b), leading to particularly high NCP (Figure 5b). In this case organic matter produced in this region is advected downstream (southwestward) where it is remineralized in the northern coastal MAB in a region of net heterotrophy, that is, where respiration exceeds primary production (Figures 10a and 10b).
Seasonal variability of NCP is large (Figures 4d and 4f) and is matched in magnitude only by the time rate of change terms (∂/∂t) as inorganic nitrogen is converted to organic nitrogen through primary production in the spring (∂DIN/∂t < 0 and ∂TON/∂t > 0) and organic nitrogen is then converted back into inorganic nitrogen through respiration and remineralization processes in the early summer (∂TON/∂t < 0 and ∂DIN/∂t > 0). Thus, NCP is greatest in the spring and typically becomes negative in the early summer, as recycling begins to dominate (Figure 4d). This pattern reflects the interaction between primary production and respiration (Figure 10c), with both following the same pattern of being largest in the spring and smallest in winter (Xu et al., 2013). However, spring primary production is slightly larger than respiration causing net autotrophy (NCP > 0) and summer respiration is slightly larger than productivity causing net heterotrophy (NCP < 0). The same cycle occurs again in the late summer/early fall, with a shorter time period of primary production exceeding respiration (net autotrophy) followed by a time during which respiration exceeded primary production (net heterotrophy).
Interannual variability of NCP was also substantial (Figure 9), due to large interannual variations in DIN transport into the MAB. Riverine fluxes along the U.S. East Coast vary substantially from year to year in response to variability in rates of precipitation and evapotranspiration (Yang, Tian, Friedrichs, Hopkinson, et al., 2015; Yang, Tian, Friedrichs, Liu, et al., 2015). Although the interannual variability of riverine DIN inputs may be significant, that is, nearly a factor of 2 different between 2004 (0.15 Tg N/year) and 2007 (0.08 Tg N/year; Table 3), the interannual difference (Δ = 0.07 Tg N/year) is much less than the interannual difference of DIN entering along the shelf from the northeast (Δ = 0.56 Tg N/year) and flowing across the continental slope (Δ = 1.00 Tg N/year). However, the interannual variability of the sum of these two fluxes (across slope + along shelf) is smaller than either individually, since in each year the large volume flux entering from the northeast (5-year mean = 80 mSv; Table 1) is nearly balanced by the large volume flux leaving the MAB across the continental slope (5-year mean = 87 mSv; Table 1). Nevertheless, the interannual variability of the net advective flux to the MAB (across slope + along shelf) is still larger than that entering from the estuaries and rivers both in terms of DIN inputs and water volume flux.
The large interannual variability in along-shelf and across-shelf transport in the MAB is well known (e.g., He et al., 2011) and is ultimately responsible for the interannual variability in NCP (Figure 9). Somewhat surprisingly, NCP is highest in the year that the water volume flux entering from the estuaries and rivers is lowest (2007), indicating that the main source of DIN for the high NCP is not terrestrial. For the years studied here (2004–2008), low estuarine and riverine input of water to the MAB cooccurs with the greatest input of water across the continental slope in the southern MAB (Figure 6e and Table 1). Specifically, NCP is highest in the years when the Gulf Stream is situated closest to the MAB (2007 and 2008), remaining along the continental slope longer relative to other years before heading out to the open ocean (Figures 8e and 8f) and thus providing greater opportunity for onshore intrusions of offshore waters through warm core rings that develop from Gulf Stream meanders and other shelf break exchange processes (Gawarkiewicz et al., 1996; Kang & Curchitser, 2015; Lentz, 2003; W. G. Zhang et al., 2013; W. G. Zhang & Gawarkiewicz, 2015; S. Zhang et al., 2018). The unusually close proximity of the warm Gulf Stream waters to the continental slope in these particular years (2007–2008) is evident from satellite-derived sea surface temperatures showing unusually warm MAB waters (Figures 11a and 11c), in close agreement with the model results (Figures 11b and 11d). The literature also provides other examples of onshore excursions of the Gulf Stream at the southern end of the MAB. For example, using sea surface temperatures and hydrographic data, Churchill and Cornillon (1991) tracked such excursions and documented the presence of Gulf Stream water on the continental shelf 80–300 km north of Cape Hatteras. Similarly, Biscaye et al. (1994) show a large northward excursion of the Gulf Stream occurring during the SEEP-II experiment in 1988–1989. Overall, our results showing that broad-scale circulation patterns are significant drivers of ecosystem variability are consistent with other studies that have demonstrated that fluctuations in water masses and circulation patterns drive shifts in ecosystem states (Nye et al., 2014). However, a longer model simulation would be required to determine whether there is a significant inverse correlation between water transported into the MAB from the rivers and estuaries, and water entering the southern MAB across the continental slope.
4.3 Future Modeling Efforts
The advantages of the modeling system used in this study relative to those used in earlier model-based analyses of MAB nitrogen budgets are as follows: (1) linkages with a sophisticated mechanistic terrestrial model (DLEM; Tian et al., 2015; Yang, Tian, Friedrichs, Hopkinson, et al., 2015); Yang, Tian, Friedrichs, Liu, et al., 2015 that provides riverine inputs of freshwater and nitrogen, (2) linkages with the CMAQ model that provides atmospheric inputs of nitrogen (St-Laurent et al., 2017), and (3) a more realistic biogeochemical model that includes multiple phytoplankton and zooplankton size classes (Xiao & Friedrichs, 2014a) as well as dissolved organic matter, resuspension, and burial (Druon et al., 2010). However, this modeling system can be improved by better representation of the estuaries and inclusion of data assimilation, each of which is discussed below.
Estuaries are key regions of biogeochemical transformations of inorganic and organic nitrogen. Although the regional model used here resolves the U.S. East Coast estuaries to some degree (Figure 1), the horizontal resolution of the grid prohibits adequate resolution of the complex coastline and critical bathymetric gradients inside these estuaries, which are critical for correctly representing both physical and biogeochemical processes (Ye et al., 2018). For example, the lack of a deep mainstem trench in the Chesapeake Bay prevents the development of hypoxia and hence water column denitrification, which is a key biogeochemical transformation process in the estuary (Da et al., 2018; Feng et al., 2015; Irby et al., 2018, 2016; Wiggert et al., 2017). The version of the USECoS biogeochemical model used here also neglects inorganic sediment inputs from the rivers, which can significantly affect light attenuation and therefore estuarine productivity. Specifically, better spatial resolution of the estuarine physical and biogeochemical processes will alleviate overestimating the input of inorganic nitrogen to the MAB and underestimating the input of organic nitrogen. Using a higher-resolution Chesapeake Bay hydrodynamic-biogeochemical model, Feng et al. (2015) found an order of magnitude more DIN entering the Chesapeake Bay via rivers than exiting through the bay mouth to the MAB. They contrasted their results with those of a lower-resolution model without estuarine-specific biogeochemical processes such as water column denitrification and light attenuation due to inorganic suspended sediment. The lower-resolution model quickly transported riverine nutrients directly to the shelf, so there was no time for extensive remineralization of the nutrients within the estuarine system. As a result, riverine inputs of DIN and coastal exports of DIN were much more similar in magnitude for the low-resolution model. Results from our analyses suggest that the modeling system used here is likewise overestimating the input of DIN to the continental shelf, likely again because of underestimating estuarine residence times and neglecting critical estuarine processes. Ongoing work is focused on implementing a nested higher-resolution grid in the east coast estuaries, implementing estuarine-specific processes such as water column denitrification and inorganic sediment inputs, and examining the impact on the nitrogen budget.
Assimilation of biogeochemical observations, either in situ or remotely sensed, provides an approach for improving simulations of nutrient and carbon budgets (Friedrichs & Kaufman, 2018; Hofmann & Friedrichs, 2001; Kaufman et al., 2018; Xiao & Friedrichs, 2014a, 2014b). Perhaps more importantly, assimilation of physical data using a variational data assimilation algorithm will help ensure that the space and time structure of the MAB circulation is consistent with available observations and includes the details of the regional circulation (Levin et al., 2018). In particular, assimilating physical data such as temperature, salinity, and sea surface height anomalies will improve the realism of the model simulations by adjusting the positions of the Gulf Stream and associated mesoscale eddies (Kang & Curchitser, 2013) closer to their observed locations. Since a key result of this paper is that nitrogen fluxes are highly dependent on the MAB circulation field, it is important to accurately represent the circulation of the MAB when attempting to quantify these fluxes.
5 Summary and Conclusions
Continental shelves are characterized by complex physical and biological processes occurring at a variety of temporal and spatial scales, making it difficult to acquire enough observations to fully resolve area-integrated annual nitrogen fluxes for shelf regions. Here a 3-D coupled biogeochemical-circulation model was applied to the MAB to investigate the nitrogen budget of this region and examine the spatiotemporal variability of key nitrogen fluxes. NCP was characterized by strong spatial, seasonal, and interannual variability, especially when compared to other biogeochemical nitrogen fluxes such as burial and denitrification. The strong interannual variability of NCP was not directly due to interannual variability of terrestrial inorganic nitrogen inputs but rather was due to interannual variability in oceanic circulation patterns, which caused significant changes in the along-shelf and across-slope horizontal fluxes transporting water into and out of the MAB. Typically, horizontal water transport and inorganic nitrogen flux entered the MAB from the northeast and exited the MAB along the continental slope. However, when a meander caused the Gulf Stream to remain along the continental slope longer before veering out to the open ocean, intrusions of deep high nutrient water onto the shelf stimulated productivity and increased NCP. In this way oceanic circulation patterns were found to drive NCP variability to a greater degree than interannually varying terrestrial inputs of inorganic nitrogen.
The modeling system used here has been evaluated with in situ and satellite data (St-Laurent et al., 2017) and includes substantial improvements over earlier MAB models (Cahill et al., 2016) including linkages with a sophisticated terrestrial model, an air-quality model, and a more realistic biogeochemical model. However, significant improvements are already underway, including an increased resolution in the estuaries through nesting, assimilation of physical data to improve the realism of circulation patterns, and longer simulations to further test the results found here for the 5-year analysis. Overall, the strong spatiotemporal variability in nitrogen fluxes reported here highlights the importance of the breadth of data coverage throughout all seasons and across multiple years in order to correctly assess the current status and future changes in the MAB nitrogen budget.
Acknowledgments
This work was supported by NASA Headquarters under the NASA Earth and Space Science Fellowship Program (NNX10AN50H) and the NASA Interdisciplinary Science Program (NNX11AD47G and NNX14AF93G). D.A. Narvaez is partially funded by COPAS Sur-Austral CONICYT PIA APOYO CCTE AFB170006 and FONDECYT 1116109. This work was performed using High Performance Computing facilities at the College of William & Mary, which were provided by contributions from the National Science Foundation, the Commonwealth of Virginia Equipment Trust Fund and the Office of Naval Research. We would also like to thank the two anonymous reviewers for their very helpful comments and suggestions. Model output is publicly available through W&M's ScholarWorks at (https://doi.org/10.25773/2f36-pn56). This paper is contribution 3793 of the Virginia Institute of Marine Science, College of William and Mary.
