Carbon cycling in the North American coastal ocean: A synthesis

. A quantiﬁcation of carbon ﬂuxes in the coastal ocean and across its boundaries with the atmosphere, land, and the open ocean is important for assessing the current state and projecting future trends in ocean carbon uptake and coastal ocean acidiﬁcation, but is currently a missing component of global carbon budgeting. This synthesis reviews recent progress in characterizing these carbon ﬂuxes for the North American coastal ocean. Several observing networks and high-resolution regional models are now available. Recent efforts have focused primarily on quantifying net air-sea exchange of carbon dioxide 5 (CO 2 ). Some studies have estimated other key ﬂuxes, such as the exchange of organic and inorganic carbon between shelves and the open ocean. Available estimates of air-sea CO 2 ﬂux, informed by more than a decade of observations, indicate that the North American Exclusive Economic Zone (EEZ)

Carbon export flux from coastal waters to the deep ocean cannot be quantified easily or accurately through direct observation, especially considering the 3-dimensional nature of exchanges between the coastal and open ocean (Frischknecht et al., 2018).
Thus, the only available estimates of such export are indirect, using mass balances of POC and dissolved oxygen (Hales et al., 2006), mass balances of DOC (Barrón and Duarte, 2015;Vlahos et al., 2002), mass balances of TOC and DIC (Najjar et al., 2018), or model estimates (Izett and Fennel, 2018a, b;Bourgeois et al., 2016;Fennel and Wilkin, 2009;Mannino et al., 2016;5 Xue et al., 2013;Frischknecht et al., 2018). If the total carbon inventory in a coastal system can be considered constant over a sufficiently long timescale (i.e., on the order of years), inferring carbon export is possible from using the sum of all other exchange fluxes across the system's interfaces over that same period. Export to the open ocean must balance the influx of carbon from land and wetlands, its net exchange across the air-sea interface, lateral exchange caused by advection, and any removal through permanent sediment burial. The accuracy of the inferred export flux directly depends on the accuracy of the 10 other flux estimates and of the assumption of a constant carbon inventory. Quantifying internal transformation processes (e.g., respiration, primary and secondary production) does not directly enter this budgeting approach but can elucidate the processes that drive fluxes across interfaces.
Current estimates of carbon fluxes across coastal interfaces come with significant uncertainties . These uncertainties are caused by a combination of small-scale temporal and spatial variability, which is undersampled by currently 15 available means of direct observation, and regional heterogeneity, which makes scaling up observations from one region to larger areas difficult. Contributing to variability in regional carbon budgets and export are geographical differences arising from variations in shelf width, upwelling strength, the presence or absence of large rivers, seasonal ice cover, and latitude.
Section 3 describes the regional characteristics of North American coastal waters and how these characteristics influence carbon dynamics. 20 The motivation for quantifying permanent burial of organic carbon and export of carbon from coastal waters to the deep ocean is that both processes remove CO 2 from the atmospheric reservoir. A more relevant but harder to obtain quantity in this context is the burial or export of anthropogenic carbon, i.e. the carbon that was added to the atmosphere by anthropogenic activities. Present-day carbon fluxes represent a superposition of the anthropogenic flux component and the natural background flux (see McNeil and Matear, 2013, for further details). Only total fluxes-the sum of anthropogenic and background fluxes- 25 can be observed directly. Distinction between anthropogenic fluxes and the natural background is difficult to assess for coastal ocean fluxes and has to rely on process-based arguments and models  .

Review of coastal carbon fluxes around North America
In this section we briefly describe the bathymetric and hydrographical features of the four major North American coastal margins (the Atlantic coast, the Pacific coast, the coast of the northern Gulf of Mexico, and the Arctic coast), followed by 30 a review of available carbon flux estimates for each. Where multiple flux estimates are available for the same region it is important to keep in mind that their spatial footprints and time windows do not necessarily match exactly.  Table S1.

Atlantic coast
The North American Atlantic coast borders on a wide, geologically passive-margin shelf that extends from the southern tip of Florida to the continental shelf of the Labrador Sea ( Figure 1). The shelf is several hundred kilometers wide in the north (Labrador shelf and Grand Banks) but narrows progressively toward the south in the Mid Atlantic Bight (MAB), which is between Cape Cod and Cape Hatteras, and the South Atlantic Bight (SAB), which is south of Cape Hatteras. The SAB shelf width measures only several tens of kilometers. Two major semi-enclosed bodies of water are the Gulf of Maine (GOM) and  (Loder, 1998). Because shelf waters north of Cape Hatteras are sourced from the Labrador Sea, they are relatively cold, fresh, and carbon rich, while slope waters (those located between the shelf break and the northern wall of the 10 Gulf Stream) are a mixture of Labrador Current and Gulf Stream water. South of Cape Hatteras, exchange between the shelf and open ocean across the shelf break is impeded by the presence of the Gulf Stream and occurs via baroclinic instabilities in its northern wall (Lee et al., 1991). In the MAB and on the Scotian Shelf, cross-shelf exchange is hindered by shelf-break jets and fronts (Rutherford and Fennel, 2018).
Air-sea fluxes of CO 2 exhibit a large-scale latitudinal gradient along the Atlantic coast ( Figure 2) and significant seasonal 15 and interannual variability ( Figure 3). Discrepancies in independent estimates of net air-sea flux are largest for the Scotian Shelf and the Gulf of Maine ( Figure 2). For the Scotian Shelf, Shadwick et al. (2010), combining in situ and satellite observations, reported a large source of CO 2 to the atmosphere of 8.3 ± 6.6 g C m −2 yr −1 . In contrast, Signorini et al. (2013) estimated a relatively large sink of atmospheric CO 2 , 14 ± 3.2 g C m −2 yr −1 , when using in situ data alone and a much smaller uptake, 5.0 ± 4.3 g C m −2 yr −1 , from a combination of in situ and satellite observations. The open GOM (excluding the tidally mixed 20 Georges Bank and Nantucket Shoals) was reported as a weak net source of 4.6 ± 3.1 g C m −2 yr −1 by Vandemark et al.
The shallow, tidally mixed Georges Bank and Nantucket Shoals are thought to be sinks, however (Table S1).
The MAB and SAB are consistently estimated to be net sinks. Observation-based estimates for the MAB sink are 13 ± 8.3 g C m −2 yr −1 (DeGrandpre et al., 2002) and 13 ± 3.2 g C m −2 yr −1 (Signorini et al., 2013). Estimates for the SAB sink 25 are 5.8 ± 2.5 g C m −2 yr −1 (Jiang et al., 2008) and 8.2 ± 2.9 g C m −2 yr −1 (Signorini et al., 2013). The transition from neutral or occasional net source on the Scotian Shelf and in the GOM to net sink in the MAB arises because the properties of shelf water are modified during its southwestward flow by air-sea exchange, inflows of riverine and estuarine waters (Salisbury et al., 2008(Salisbury et al., , 2009, and exchange with the open North Atlantic across the shelf break (Cai et al., 2010b;Wang et al., 2013).
The cold, carbon-rich water on the Scotian Shelf carries a pronounced signature of its Labrador Sea origin. The GOM, which 30 is deeper than the Scotian Shelf and the MAB and connected to the open North Atlantic through a relatively deep channel, is characterized by a mixture of cold, carbon-rich shelf waters entering from the Scotian Shelf and warmer, saltier slope waters.
Shelf water in the MAB is sourced from the GOM and thus is a mixture of Scotian Shelf and slope water.
Shelf water in the SAB is distinct from that in the MAB and has almost no trace of Labrador Current water; instead, its characteristics are similar to those of the Gulf Stream, but its carbon signature is modified by significant organic and inorganic 35 carbon and alkalinity inputs from coastal marshes (Cai et al., 2003;Jiang et al., 2013;Wang and Cai, 2004;Wang et al., 2005). Herrmann et al. (2015) estimated that 59% of the 3.4 Tg C yr −1 of organic carbon export from U.S. East Coast estuaries occurs in the SAB. The subsequent respiration of this organic matter and direct outgassing of marsh-derived carbon make the nearshore regions a significant CO 2 source almost year-round. Despite the carbon inputs from marshes, uptake of CO 2 on the mid-and outer shelf regions during the winter months is large enough to balance CO 2 outgassing in the other seasons and on 5 the inner shelf, making the SAB an overall weak sink (Jiang et al., 2008).
The seasonality of CO 2 dynamics along the Atlantic coast varies with latitude. North of Cape Hatteras, CO 2 dynamics are characterized by strong seasonality with solubility-driven uptake by cooling in winter and biologically driven uptake in spring followed by outgassing in summer and fall due to warming and respiration of organic matter (Figure 3d; DeGrandpre et al., 2002;Shadwick et al., 2010;Signorini et al., 2013;Vandemark et al., 2011;Wang et al., 2013). South of Cape Hatteras, seasonal 10 phytoplankton blooms do not occur regularly and biologically driven CO 2 uptake is less pronounced than that further north ( Figure 3e; Reimer et al., 2017;Wang et al., 2013), although sporadic phytoplankton blooms do occur because of intrusions of high-nutrient subsurface Gulf Stream water (Wang et al., 2005. The influence of riverine inputs is small and localized in the SAB (Cai and Wang, 1998;Wang and Cai, 2004;Wang et al., 2005;Xue et al., 2016a;Reimer et al., 2017). An exception to this north-south importance in riverine input is the GOM, where riverine inputs of carbon and nutrients are relatively small. 15 Nevertheless, even here, these inputs can cause local phytoplankton blooms, CO 2 drawdown, and low-pH conditions (Salisbury et al., 2008(Salisbury et al., , 2009).
Regional biogeochemical models reproduce the large-scale patterns of air-sea CO 2 flux with oceanic uptake increasing   ; South Atlantic Bight (Gray's Reef mooring; Sutton et al., 2011) attributed them to gas hydrate degradation. Estimated CH 4 efflux from the sediment in this region ranges from 1.5 × 10 −5 to 1.8 × 10 −4 Tg yr −1 , where the uncertainty range reflects different assumptions underlying the conversion from CH 4 plume observations to seepage rates. The fraction of the released CH 4 that escapes to the atmosphere remains uncertain (Phrampus and Hornbach, 2012).

5
The North American Pacific coast extends from Panama to the Gulf of Alaska and is an active margin with varying shelf widths ( Figure 1). The continental shelf is narrow along the coasts of California, Oregon, and Washington, on the order of 10 km, but widening significantly in the Gulf of Alaska, where shelves extend up to 200 km offshore. In the Gulf of Alaska, freshwater and tidal influences strongly affect cross-shelf exchange, and the shelf is dominated by a downwelling circulation.
The region from Vancouver Island to Baja California is a classic eastern boundary current upwelling region-the California flow in the California Current and by coastal jets and their associated eddies and fronts that extend offshore, particularly off the coasts of Baja California, California, Washington, and Oregon (Huyer, 1983).
The northern California Current System experiences strong freshwater influences and seasonality in wind forcing that diminish toward the south. In addition to the Columbia River and the Fraser River, a variety of small mountainous rivers, with highly variable discharge, supply freshwater. The Central American Isthmus runs from Panama to the southern tip of Baja California 5 and experiences intense and persistent wind events, large eddies, and high waves that combine to produce upwelling and strong nearshore mixing (Chapa-Balcorta et al., 2015;Franco et al., 2014). In addition to alongshore winds, strong seasonal wind jets that pass through the Central American cordillera create upwelling hotspots and drive production during boreal winter months in the gulfs of Tehuantepec, Papagayo, and Panama (Brown et al., 2015;Chapa-Balcorta et al., 2015;Chelton et al., 2000a, b;Gaxiola-Castro and Muller-Karger, 1998). The California Current brings water from the North Pacific southward into 10 the southern California and Central American Isthmus regions, while the California Undercurrent transports equatorial waters northward in the subsurface (Hickey, 1998).
The net exchange of CO 2 with the atmosphere along the Pacific coast is characterized by strong spatial and temporal variation and reflects complex interactions between biological uptake of nutrients and degassing of nutrient-and carbon-rich upwelled waters (Figure 3b, c). A growing number of coastal air-sea flux studies have used extrapolation techniques to estimate fluxes 15 across the coastal oceans on regional to continental scales ( Figure 2). Observation-based studies of air-sea CO 2 flux suggest that estimates for the coastal ocean from Baja California to the Gulf of Alaska range from a weak (Coronado-Álvarez et al., 2017) to moderate sink of atmospheric CO 2 over this broad longitudinal range. Central California coastal waters have long been understood to have near-neutral air-sea CO 2 exchange because of their large and counter-balancing periods of efflux during upwelling conditions and influx during periods of relaxation and high primary productivity; this pattern is strongly modulated 20 by El Niño-La Niña conditions (Friederich et al., 2002). Hales et al. (2005) used seasonal data to estimate an uptake of 88 g C m −2 yr −1 by Oregon coastal waters. In a follow-up analysis with greater temporal coverage, Evans et al. (2011) showed how large flux events can significantly alter the estimation of net exchanges for the Oregon shelf. After capturing a large and short-lived efflux event, their annual estimate was outgassing of 3.1 ± 82 g C m −2 yr −1 for this same region. The disparity illustrates the importance of basing regional flux estimates on   Figure 2 because the area-normalized flux is not available from that study.) Both models simulate strong outgassing within the first 100 km of shore, driven by intense upwelling of nutrient-and carbon-rich water, compensated by biologically driven CO 2 uptake from the atmosphere as upwelled nutrients are consumed by photosynthesis during subsequent offshore advection within several hundreds of kilometers of the coast. The 5 disagreement in mean simulated fluxes between the two models may result partly from different choices of averaging region and period and differences in model forcing, such as the climatological forcing in Turi et al. (2014) versus realistic variability in Fiechter et al. (2014). Notably, observations for the Oregon shelf by Evans et al. (2015a) showed intense summer upwelling that led to strong outgassing with pronounced variability in air-sea fluxes but only weak stimulation of primary production. They hypothesized that nutrient-rich waters might be subducted offshore at convergent surface temperature fronts before nutrients 10 are fully consumed by primary producers.
Cross-shelf exchange of carbon occurs in the California Current System mostly in response to wind-driven circulation and eddies, but river plumes and tides also have been shown to increase offshore transport in the northern part of the system (Barth et al., 2002;Hales et al., 2006). Uncertainties in published estimates are high, ranging from very small (Ianson and Allen, 2002; Pennington et al., 2010) to very high fractions of primary production (Hales et al., 2005;Turi et al., 2014), again as a result 15 of the region's large spatial and temporal variability. Frischknecht et al. (2018) showed that about 30% of the organic matter produced within 100 km of shore is laterally advected toward the open ocean.
Less is known about the air-sea flux of CH 4 along the Pacific margin. Recent studies inventoried sedimentary sources of CH 4 hydrates, derived from terrestrial and coastal primary production, and suggested that extensive deposits along the Cascadia margin are beginning to destabilize because of warming (Hautala et al., 2014;Johnson et al., 2015). 20

Gulf of Mexico
The Gulf of Mexico is a semi-enclosed marginal sea at the southern coast of the conterminous United States. The passive margin shelves of its northern portion are relatively wide (up to 250 km west of Florida) but, in contrast to shelf waters of the Atlantic coast, those of the Gulf of Mexico are not separated from open-ocean waters by shelf-break fronts or currents. Ocean water enters the Gulf mainly through the Yucatan Channel, where it forms the northeastward meandering Loop Current, which 25 sheds anticyclonic eddies and exits the Gulf through the Florida Straits (Muller-Karger et al., 2015;Rivas et al., 2005). While shelf circulation is influenced primarily by local wind and buoyancy forcing, outer-shelf regions are at times influenced by Loop Current eddies that impinge on and interact with the shelf (Lohrenz and Verity, 2004). Riverine input is substantial in the northern Gulf of Mexico, where the Mississippi-Atchafalaya river system delivers large loads of freshwater, nutrients, and sediments.

30
Estimates of air-sea CO 2 flux are available from observations and model simulations ( Figure 2). Observational estimates indicate that the Gulf of Mexico, as a whole, is a weak net sink of atmospheric CO 2 with an annual average of 2.3 ± 1.0 g C m −2 yr −1 (Robbins et al., 2014). Smaller shelf regions within the Gulf differ markedly from this mean flux. The West Florida Shelf and western Gulf shelf act as sources to the atmosphere, with estimated annual average fluxes of 4.4 ± 1.3 and 2.2 ± 0.6 g C m −2 yr −1 , respectively; the northern Gulf acts as a sink, with an estimated flux of 5.3 ± 4.4 g C m −2 yr −1 ; and the Mexican shelf is almost neutral, with an estimated uptake flux of 1.1 ± 0.6 g C m −2 yr −1 (Robbins et al., 2014). A more recent estimate for the West Florida Shelf is 4.4 ± 18.0 g C m −2 yr −1 (Robbins et al., 2018). Huang et al. (2015) estimated a larger uptake on the northern Gulf shelf of 11 ± 44 g C m −2 yr −1 (i.e., about twice the estimate of Robbins et al. (2014)) and reported a much larger uncertainty. In an analysis that combines satellite and in situ observations, Lohrenz et al. (2018) estimated a similar uptake for 5 the northern Gulf of Mexico of 13 ± 3.6 g C m −2 yr −1 . The overall air-sea carbon exchanges in the Gulf vary significantly from year to year because of interannual variability in wind, temperature, and precipitation (Muller-Karger et al., 2015).
Model-simulated air-sea CO 2 fluxes by Xue et al. (2016b) agree relatively well with the estimates of Robbins et al. (2014), reproducing the same spatial pattern though their simulated Gulf-wide uptake of 8.5 ± 6.5 g C m −2 yr −1 is larger. This discrepancy results largely from a greater simulated sink in the open Gulf. Also, the uncertainty estimates of the model- They estimated ocean-atmosphere fluxes from seep plumes of 1,150 ± 790 to 38,000 ± 21,000 g CH 4 m −2 d −1 compared with 2.2 ± 2.0 to 41 ± 8.2 g CH 4 m −2 d −1 for background sites. Subsequent acoustic analyses of bubble plume characteristics question the finding that CH 4 bubbles make their way to the surface (Weber et al., 2014), and the fate of CH 4 emissions from seeps and their overall contribution to atmospheric CH 4 remain uncertain.

Arctic coast
The North American Arctic coastal ocean comprises broad (∼300 km) shallow shelves in the Bering and Chukchi seas, the narrower (<100-km) Beaufort Sea shelf, Hudson Bay, and the extensive Canadian Arctic shelf ( Figure 1). Shelf water enters these regions from the North Pacific through the Bering Strait and follows a large-scale pathway via the Chukchi and Beaufort seas onto the Canadian Arctic shelf and, ultimately, the North Atlantic (Carmack et al., 2006(Carmack et al., , 2015. Hudson Bay receives 25 significant inputs of freshwater (Déry et al., 2005). Except for the southernmost Bering Sea, most of the coastal Arctic is covered with sea ice from about October to June. Areas of persistent multiyear sea ice at the northernmost extent of the Canadian Arctic shelf are rapidly declining (Stroeve et al., 2012).
Coastal waters in the Arctic have been described consistently as a net sink for atmospheric CO 2 (Figure 2). An observationbased estimate for uptake in the Bering Sea is 9.6 g C m −2 yr −1 by Cross et al. (2014b). Estimates for the Chuckchi Sea 30 range from 15 g C m −2 yr −1 (Evans et al., 2015b) to 175 ± 44 g C m −2 yr −1 (Bates, 2006). For the Beaufort Sea, estimates range from from 4.4 g C m −2 yr −1 (Evans et al., 2015b) to 44 ± 28 g C m −2 yr −1 (Else et al., 2013). In Hudson Bay, Else et al. (2008) estimated an uptake of 3.2±1.8 g C m −2 yr −1 . The consistently observed uptake is thought to be caused by low surface water pCO 2 relative to the atmosphere during ice-free months (see, e.g., Figure. 3a). These low levels are set by a combination of low water temperatures and seasonally high rates of both ice-associated and open-water primary production (Cai et al., 2010a(Cai et al., , 2014Steiner et al., 2014), as well as by limited gas exchange through sea ice relative to open water during winter (Butterworth and Miller, 2016;van der Loeff et al., 2014).
In recent years, sea ice growth and decay has been shown to significantly affect the air-sea CO 2 flux (Rysgaard et al., 2007(Rysgaard et al., , 2009). During sea ice formation, brine rejection forms dense high-saline water that is enriched in DIC relative to alkalinity 5 (Dieckmann et al., 2008;Rysgaard et al., 2013). During sea ice decay, an excess of alkalinity relative to DIC is released in meltwater. The sinking of dense, carbon-rich brine provides a pathway for carbon sequestration as suggested by Rysgaard et al. (2013), although modeling studies suggest that carbon export through this process is relatively small (Moreau et al., 2016(Moreau et al., , 2015. With regard to Arctic CH 4 fluxes, much more is known about the emission potential, distribution, and functioning of terres-10 trial sources (McGuire et al., 2009); knowledge of marine CH 4 sources is developing slowly due to sparse observations and the logistical challenges of Arctic marine research. The largest marine CH 4 source in the Arctic is dissociation of gas hydrates stored in continental margin sediments (Parmentier et al., 2013). As sea ice cover continues to retreat and ocean waters warm, CH 4 hydrate stability is expected to decrease with potentially large and long-term implications. An additional potential marine CH 4 source, unique to polar settings, is release from subsea permafrost layers, with fluxes from thawed sediments reported to 15 be orders of magnitude higher than fluxes from adjacent frozen sediments (Shakhova et al., 2015).

Summary estimate of CO 2 uptake and a carbon budget for the North American EEZ
Despite the variability in regional estimates discussed above and summarized in Figure 2 and Table S1, North American coastal waters clearly act as a net sink of atmospheric carbon. However, because of the local footprint of some studies, discrepancies in temporal and spatial coverage among studies, and gaps in space and time it is difficult to combine these various regional 20 estimates into one summary estimate of CO 2 uptake for the North American EEZwith any confidence. In order to arrive at such a summary estimate, we draw on the regional information compiled in the previous sections in combination with two global approaches: the observation-based synthesis of Laruelle et al. (2014) (Table 1). Although there are 30 significant regional discrepancies between the two estimates for the eastern tropical Pacific Ocean, the Gulf of Mexico, the Florida Upwelling region (actually covering the eastern United States including the SAB, MAB and GOM), the Labrador Sea, and the Canadian Arctic shelf, the total flux estimates for North America are in close agreement. This builds some confidence in the global model estimates. Next, we use the global model to obtain CO 2 flux estimates for the North American EEZ, in a finer grained decomposition ( Figure 4, Table 2). First, we compare the regional area-specific flux estimates from Figure 2 and Table 2 with the global model (see top three bar graphs in Figure 4). For the Atlantic coast there is excellent agreement for the SAB, but the global model simulates a higher CO 2 uptake than the regional studies in the MAB, GOM and the Scotian Shelf. This discrepancy is likely due to the fact that the global model is too coarse to accurately simulate the shelf-break current systems and thus significantly 5 underestimates the water residence times on the shelf (Rutherford and Fennel, 2018). Area-specific CO 2 flux (g C m -2 y -1 ) Ocean uptake, Outgassing Total CO 2 flux (Tg C y -1 ) Total flux, Anthropogenic flux component For the Labrador Shelf, the region with the largest flux estimate, there are no regional estimates available. For the two next largest, the Gulf of Alaska and the Bering Sea, the global model agrees well with the regional estimates. Thus, despite of some regional discrepancies, it seems justifiable to use the global model estimates for a North American coastal carbon budget. The flux contributions are 104 Tg C yr −1 , 62 Tg C yr −1 , and -3.7 Tg C yr −1 , respectively, with area contributions of 51%, 25%, and 24% to the total area of the North American EEZ, respectively.
Next, we construct a carbon budget for the North American EEZ by combining the atmospheric CO 2 uptake estimate with estimates of carbon transport from land and carbon burial in ocean sediments. We assume 160 Tg C yr −1 as the best estimate of net uptake by coastal waters of North America, excluding tidal wetlands and estuaries. Unfortunately, there are no formal 5 error estimates for this uptake. Instead, we estimate an error by noting first that the Bourgeois et al. (2016) model is in good agreement with the observation-based estimates for the MARCATS regions of North America, and furthermore, that the error estimate for the uptake by continental shelves globally is about 25%, with the North American MARCATS regions having mainly "fair" data quality (Laruelle et al., 2014). Hence, assuming an error of ±50% for the uptake by North American EEZ waters seems reasonable.  Table 2, with a total flux of 120 Tg C yr −1 . We consider these fluxes to be an upper bound because they are substantially larger than other estimates. The Dunne et al. (2007) global estimates of organic carbon burial in waters shallower 15 than 2,000 m are 19±9 g C m −2 yr −1 , much larger than the estimates of 6 and 1 g C m −2 yr −1 by Chen (2004) Dunne et al. (2007). Finally, Dunne et al. (2007) estimated the organic carbon burial in Hudson Bay to be 19 g C m −2 yr −1 , compared to a mean estimate of 1.5±0.7 g C m −2 yr −1 of burial from sediment cores (Kuzyk et al., 2009). Given these results, we consider the estimates of Dunne et al. (2007) to be an upper bound and assume that a reasonable lower bound is about an order of magnitude smaller, thus placing the organic carbon burial estimate at 65±55 Tg C yr −1 .
If these estimates of net air-sea flux, carbon burial, and carbon input from land are accurate, then the residual must be 5 balanced by an increase in carbon inventory in coastal waters and a net transfer of carbon from coastal to open-ocean waters ( Figure 5). The rate of carbon accumulation in the North American EEZ from the model of Bourgeois et al. (2016) is 50 Tg C yr −1 (Figure 5). Here again, we assume an uncertainty of ±50%. The residual of 151±105 Tg C yr −1 is the inferred export of carbon to the open ocean ( Figure 5). The fact that the error in this residual is large in absolute and relative terms emphasizes the need for more accurate quantification of the terms in the coastal carbon budget. The challenge, however, is that 10 many of these terms are small compared to internal carbon cycling in coastal waters, which is dominated by primary production and decomposition. Two separate estimates of primary production ( Table 2) are in broad agreement and reveal that the fluxes in Figure 5 (on the order of 10 to 100 Tg C yr −1 ) are just a few percent of primary production (on the order of 1000 Tg C yr −1 , see Total in Table 2). This also indicates that small changes in carbon cycling in coastal waters can result in large changes in atmospheric uptake and transport to the open ocean. A direct effect of increasing atmospheric CO 2 will be an increase in net uptake by the coastal ocean, but this is tempered by a decrease in the ocean's buffer capacity as DIC increases. More indirect effects include changes in climate forcings (i.e., 25 surface heat fluxes, winds, and freshwater input).
Ocean warming reduces the solubility of gases and thus directly affects gas concentrations near the surface; this likely will decrease the net air-sea flux of CO 2 by reducing the undersaturation of CO 2 (see Cahill et al., 2016, for the North American Atlantic coast). Surface warming may also strengthen vertical stratification and thus impede vertical mixing, which will affect upward diffusion of nutrients and DIC. Enhanced stratification, therefore, could lead to decreases in both biologically driven 30 carbon uptake and CO 2 outgassing. However, model projections for the northern Gulf of Mexico show that the direct effect of increasing atmospheric CO 2 overwhelms the other more secondary effects (Laurent et al., 2018). Along the Pacific coast, surface warming will increase the horizontal gradient between cold, freshly upwelled source waters and warm, offshore surface water, leading to a greater tendency for the subduction of upwelled water at offshore surface temperature fronts during periods of persistent and strong upwelling-favorable winds. The cumulative effect of these processes for the Pacific coast may be greater and more persistent CO 2 outgassing nearshore and lower productivity offshore as upwelled nitrate is exported before it can be used by the phytoplankton community (Evans et al., 2015a). In the Arctic, warming leads to reductions in ice cover, which increases air-sea gas exchange (Bates and Mathis, 2009), and the melting of permafrost, which leads to the release of large quantities of CH 4 to the atmosphere, from both the land surface and the coastal ocean (Crabeck et al., 2014;Parmentier et al., 2013).

5
Changes in wind stress directly affect air-sea CO 2 fluxes via the gas transfer velocity, which is a function of wind speed, but also indirectly through changes in ocean circulation (Bakun, 1990). For the North American Atlantic coast, changes in wind stress were shown to significantly modify air-sea fluxes (Wanninkhof and Trinanes, 2017;Cahill et al., 2016;Previdi et al., 2009). Along the North American Pacific coast, upwelling-favorable winds have intensified in recent years, especially in the northern parts of the upwelling regimes (García-Reyes et al., 2015;Rykaczewski and Checkley, 2008;Rykaczewski et al., 10 2015;Sydeman et al., 2014). This has led to a shoaling of nutrient-rich subsurface waters (Aksnesa and Ohman, 2009;Bograd et al., 2015), increased productivity (Chavez et al., 2011(Chavez et al., , 2017Jacox et al., 2015;Kahru et al., 2015), higher DIC delivery to the surface (Turi et al., 2016), and declining oxygen levels (Crawford and Peña, 2016;Peterson et al., 2013;Bograd et al., 2015). In the coastal Arctic, late-season air-sea CO 2 fluxes may become increasingly directed toward the atmosphere as Arctic low-pressure systems with storm-force winds occur more often over open water, thus ventilating CO 2 respired from the high 15 organic carbon loading of the shallow shelf (Evans et al., 2015b;Hauri et al., 2013;Steiner et al., 2013).
In a study that directly assesses changes in coastal carbon uptake, Laruelle et al. (2018) investigated trends in the airsea pCO 2 gradient (∆pCO 2 = atmospheric pCO 2 − ocean pCO 2 ), which imply a strengthening or weakening of the net CO 2 uptake by shelf systems. An increasing ∆pCO 2 implies that ocean pCO 2 rises more slowly than atmospheric pCO 2 and corresponds to increased net uptake and potentially increased cross-shelf export. In their observation-based analysis of decadal 20 trends in shelf pCO 2 , Laruelle et al. (2018) found that coastal waters lag compared to the rise in atmospheric CO 2 in most regions. In the MAB, the Labrador Shelf, the Vancouver Shelf, and the SAB ∆pCO 2 has increased by 1.9 ± 3.1 µatm yr −1 , 0.68 ± 0.61 µatm yr −1 , 0.83 ± 1.7 µatm yr −1 , and 0.51 ± 0.74 µatm yr −1 , respectively, implying that surface ocean pCO 2 does not increase or increases at a slower rate than atmospheric CO 2 . The only North American coastal region that exhibits a negative trend is the Bering Sea, with -1.1 ± 0.74 µatm yr −1 , meaning that surface ocean pCO 2 increases at a faster rate 25 than in the atmosphere. Laruelle et al. (2018) concluded that the lag in coastal ocean pCO 2 increase compared to that in the atmosphere in most regions indicates an enhancement in the coastal uptake and export of atmospheric CO 2 , although they did not investigate alternative explanations.

Acidification trends
Increasing atmospheric CO 2 emissions lead to rising atmospheric CO 2 levels ( Figure 6) and a net ocean uptake of CO 2 . Since 30 about 1750, the ocean has absorbed 27% of anthropogenic CO 2 emissions to the atmosphere from fossil fuel burning, cement production, and land-use changes (Canadell et al., 2007;Le Quéré et al., 2015;Sabine and Tanhua, 2010). As a result of this uptake, the surface ocean pCO 2 has increased ( Figure 6) and oceanic pH, carbonate ion concentration, and carbonate saturation state have decreased (Caldeira and Wickett, 2003;Feely et al., 2004Feely et al., , 2009Orr et al., 2005). Commonly called is calculated using CO2SYS with Mehrbach refit coefficients for the dissociation constants of H2CO3 and HCO − 3 , and Dickson's dissociation constant for HSO − 4 . Data sources: Mauna Loa, www.esrl.noaa.gov/gmd/ccgg/trends/data.html; HOT, hahana.soest.hawaii.edu/hot/hotdogs/interface.html "ocean acidification," this suite of chemical changes is defined more precisely as "any reduction in the pH of the ocean over an extended period, typically decades or longer, which is caused primarily by uptake of CO 2 from the atmosphere but can also be caused by other chemical additions or subtractions from the ocean" (IPCC, 2011, p. 37). In addition to uptake of CO 2 from the atmosphere, variations in DIC concentrations and thus pH can be caused by ocean transport processes, and biological production and respiration. Ocean acidification can significantly affect growth, metabolism, and life cycles of marine organisms 5 (Fabry et al., 2008;Gattuso and Hansson, 2011;Somero et al., 2015) and most directly affects marine calcifiers, organisms that precipitate CaCO 3 to form internal or external body structures. When the carbonate saturation state decreases below the equilibrium point for carbonate precipitation/dissolution, conditions are said to be corrosive, or damaging, to marine calcifiers.
Ocean acidification makes it more difficult for calcifying organisms to form shells or skeletons, perform metabolic functions and survive. Early life stages are particularly vulnerable as shown by recent large-scale die-offs of oyster larvae in the coastal 10 Pacific where increased energetic expenses under low pH have led to compromised development of essential functions and insufficient initial shell formation in oyster larvae .
Acidification trends in open-ocean surface waters tend to occur at a rate that is commensurate with the rate of the increase in atmospheric CO 2 (see trends of atmospheric CO 2 in comparison to surface ocean pCO 2 at the Hawaii Ocean Time Series in Figure 6). Acidification in coastal waters is more variable and often event-driven because coastal waters have greater seasonality 15 ( Figure 3) and are more susceptible to changes in circulation, such as upwelling. Along the Pacific coast, climate-driven changes in upwelling circulation result in coastal acidification events (Chavez et al., 2017). As mentioned in Section 3.2, upwellingfavorable winds along this coast have intensified over recent years, especially in the northern parts of the upwelling regime (García-Reyes et al., 2015;Rykaczewski and Checkley, 2008;Rykaczewski et al., 2015;Sydeman et al., 2014). Intensified upwelling supplies deep water to the shelf that is rich in DIC and nutrients but poor in oxygen. Ocean acidification and hypoxia are thus strongly linked ecosystem stressors because low-oxygen, high-CO 2 conditions derive from the microbial respiration of organic matter (Feely et al., 2008(Feely et al., , 2016(Feely et al., , 2018. In the northern California Current System, pCO 2 , pH, and aragonite saturation reach levels known to be harmful to ecologically and economically important species during the summer upwelling season (Barton et al., 2012(Barton et al., , 2015Bednaršek et al., 2014Bednaršek et al., , 2016Bednaršek et al., , 2017Feely et al., 2008Feely et al., , 2016Harris et al., 2013;Sutton et al., 2016).

5
In the Gulf of Alaska, aragonite saturation drops to near saturation values during the winter months when deep mixing occurs and surface ocean pCO 2 exceeds atmospheric pCO 2 (Evans and Mathis, 2013). Along the Pacific coast, 50% of shelf waters are projected to experience year-long undersaturation by 2050 (Gruber et al., 2012;Hauri et al., 2013;Turi et al., 2016).
Polar regions are naturally prone to low pH values (Orr et al., 2005;Steinacher et al., 2009) and thus closer to critical acidification thresholds than lower-latitude waters. The low pH levels result from the naturally high ratio of DIC to alkalinity 10 in polar waters due to their low temperatures, multiple sources of freshwater (e.g., riverine, glacial melt, and sea ice melt), and high respiratory DIC content. In addition to the naturally low pH, the rate of acidification is relatively high in polar waters (Feely et al., 2009;Mathis et al., 2015;Sutton et al., 2016)  Yamamoto- Kawai et al., 2013). Model projections based on the IPCC high-CO 2 emissions scenario RCP 8.5, suggest the Beaufort Sea surface water will become undersaturated with respect to aragonite around 2025 (Steinacher et al., 2009;Steiner et al., 2014). As these conditions intensify, negative impacts for calcifying marine organisms are expected to become a critical 20 issue reshaping ecosystems and fisheries across the Arctic (Moore and Stabeno, 2015).
In contrast, surface aragonite saturation states typically range from 3.6 to 4.5 and are thus well above the dissolution threshold in the northern Gulf of Mexico Wanninkhof et al., 2015). Here excessive nutrient inputs from the Mississippi River result in hypoxia and eutrophication-induced acidification of near-bottom waters (Cai et al., 2011;Laurent et al., 2017).
Similar to the California Current System, low-oxygen and high-CO 2 conditions coincide and derive from microbial respiration 25 of organic matter (Cai et al., 2011;Laurent et al., 2017;Feely et al., 2018). Currently, aragonite saturation states are around 2 in hypoxic bottom waters and thus well above the saturation threshold. Projections suggest that aragonite saturation states of these near-bottom waters will drop below the saturation threshold near the end of this century (Laurent et al., 2018).
Along the Atlantic coast, the northern regions (the Mid Atlantic Bight and Gulf of Maine) have, on average, lower pH and lower aragonite saturation states than more southern coastal regions (i.e. the South Atlantic Bight)  Tremendous progress has been made during the last decade in improving understanding and constraining rates of carbon cycling in coastal waters because of a greatly expanded suite of observations, process studies, and models. However, quantification of many coastal carbon fluxes remains difficult. One of the challenges is that carbon is constantly exchanged across a multitude of interfaces, i.e. the sea surface as well as the interfaces between land and coastal ocean, coastal and open ocean waters, and 5 water and sediment. Furthermore, net exchange fluxes and trends are relatively small signals masked by a large and fluctuating background. At present, most of these fluxes are not quantified well enough to derive well-constrained carbon budgets for North American coastal waters or to project how those fluxes will change in the future due to various drivers.
This synthesis focused primarily on the role of ocean margins in sequestering atmospheric CO 2 and coastal ocean acidification. In the coastal ocean, a net removal of carbon from direct interaction with the atmosphere can occur by export of dissolved 10 or particulate carbon to the deep ocean or by permanent burial in sediments. Neither of these is easily observed or well quantified. The best-observed flux is gas exchange across the air-sea interface, although extracting the small net flux and its trend from a variable background with large-amplitude seasonal fluctuations is difficult. Ultimately, the uptake of anthropogenic carbon is the relevant quantity for assessing the contribution of ocean margins to the total ocean uptake of anthropogenic carbon; however, the separation of anthropogenic fluxes from the natural background is thus far elusive for coastal waters. The only 15 available estimates are from a global modeling study (Bourgeois et al., 2016).
Estimates of air-sea CO 2 fluxes currently provide the best evidence for the contribution of coastal waters to overall carbon uptake by the ocean. Our synthesis of regional studies shows that, overall, the North American coastal waters act as a sink of atmospheric CO 2 . The limited temporal and spatial footprint of many of these studies and gaps in spatial coverage (e.g., no estimates exist for the Labrador Sea and Canadian Arctic shelf) prevented us from combining these regional studies into a 20 summary estimate for the North American EEZ. Instead, we compared the regional studies to the global model of Bourgeois et al. (2016) for a fine-grained decomposition of the North American EEZ. The reasonable agreement between the regional studies and estimates from the global model builds some confidence in the model estimate of 160 Tg C yr −1 of uptake for the North American EEZ (with an anthropogenic flux contribution of 59 Tg C yr −1 ). The Labrador Sea, Gulf of Alaska, and Bering Sea are the biggest contributors making up more than half of the total uptake (and almost half of the anthropogenic uptake), 25 while accounting for less than one third of the surface area of the EEZ. The Arctic and sub-Arctic, mid-latitude Atlantic and mid-latitude Pacific account for 104 Tg C yr −1 , 62 Tg C yr −1 , and -3.7 Tg C yr −1 , respectively, while making up 51%, 25% and 24% of the total area, respectively. The comparatively large uptake along the mid-latitude Atlantic simulated in the global model of Bourgeois et al. (2016) is larger than regional estimates (e.g., Signorini et al., 2013) likely because it significantly underestimates the residence times of coastal waters (Rutherford and Fennel, 2018). Nevertheless, we used this flux estimate, 30 assuming an uncertainty of 50%, to construct a first carbon budget for the North American EEZ. The estimated uptake of atmospheric carbon of 160 ± 80 Tg C yr −1 combined with an input from land of 106 ± 30 Tg C yr −1 (Windham-Myers et al., 2018) minus an estimated burial of 65 ± 55 Tg C yr −1 leads to a large net addition of carbon. This term has to be balanced by carbon storage in waters of the EEZ and export to the open ocean. The estimated carbon storage of 50 ± 25 Tg C yr −1 from the global model leads to a carbon export of 151 ± 105 Tg C yr −1 to the open ocean. The estimated uptake of atmospheric carbon in the North American EEZ amounts to 6.4% of the global ocean uptake of atmospheric CO 2 of 2,500 Tg C yr −1 (Quéré et al., 2018). Given that the North American EEZ is about 4% of the global ocean surface area, its uptake of CO 2 is about 50% more efficient than the global average.
Coastal waters contribute significantly to the carbon budget of North America. According to Birdsey et al. (2018), the coastal 5 carbon sink estimated here is about one quarter of the net carbon sink on land in North America (606 Tg C yr −1 = the net uptake by ecosystems and tidal wetlands minus emissions from harvested wood, inland waters and estuaries). Coastal waters of North America are also a key component of aquatic carbon fluxes; specifically, much of the large net outgassing of CO 2 from inland waters, estimated at 247 Tg C yr −1 (Butman et al., 2018), is offset by coastal ocean uptake.
Several drivers influence secular trends in coastal carbon fluxes and will continue to do so in the future. These drivers include 10 the direct effect of rising atmospheric CO 2 levels and indirect effects due to changes in atmosphere-ocean interactions (e.g., wind forcing and heat fluxes), changes in the hydrological cycle, and anthropogenic perturbations of global nutrient cycling (particularly, the nitrogen cycle). The direct effect of rising atmospheric CO 2 levels alone will likely amplify coastal CO 2 uptake and carbon export, at least until the ocean's buffering capacity is significantly reduced, but the extent of this increase will depend on the rate of the atmospheric CO 2 rise, the residence time of shelf waters, and the carbon content of open-ocean 15 source waters supplied to coastal regions. Indeed, there is observational evidence for a strengthening of coastal ocean carbon export from increasing trends in the air-sea pCO 2 gradient for the Labrador Sea, Vancouver Shelf, and South Atlantic Bight (Laruelle et al., 2018). Several indirect effects such the enhanced subduction of upwelled water along the Pacific margin (Evans et al., 2015a), increased stratification on the shelf (Laurent et al., 2018), and decreased CO 2 solubility due to warming (Cahill et al., 2016) will partly counteract a strengthening of the coastal CO 2 uptake. On Arctic shelves, significant releases of the 20 potent greenhouse gas CH 4 will result from melting permafrost (Parmentier et al., 2013).
A major concern is coastal acidification, which can affect the growth, metabolism, and life cycles of many marine organisms, specifically calcifiers, and can trigger cascading ecosystem-scale effects. Most vulnerable are those organisms that precipitate aragonite, one of the more soluble forms of biogenic CaCO 3 in the ocean. The Arctic Ocean is naturally closer to critical acidification thresholds with subsurface waters in the Beaufort Sea already routinely below aragonite saturation (favoring 25 dissolution) Miller et al., 2014). Model projections show undersaturation in surface waters in the Beaufort Sea by 2025 (Steiner et al., 2014). Along the Pacific coast, atmospheric CO 2 uptake in combination with intensified upwelling that brings low-pH, low-oxygen water onto the shelves leads to aragonite levels below the saturation threshold in large portions of the subsurface waters (Feely et al., 2016(Feely et al., , 2018Chavez et al., 2017). Here half of the shelf is projected to be undersaturated by 2050 (Gruber et al., 2012;Hauri et al., 2013;Turi et al., 2016). In the northern Gulf of Mexico, aragonite saturation states 30 are currently well above the dissolution threshold despite the eutrophication-induced acidification occurring in bottom waters due to Mississippi River inputs of nutrients and freshwater (Cai et al., 2011;Laurent et al., 2017). Here undersaturation is projected to occur by 2100 (Laurent et al., 2018).
Given the importance of coastal margins, both in contributing to carbon budgets and in the societal benefits they provide, further efforts to improve assessments of the carbon cycle in these regions are paramount. Critical needs are maintaining and 35 expanding existing coastal observing programs, continued national and international coordination, and integration of observations, modeling capabilities, and stakeholder needs.
Data availability. All data sets used in this synthesis are publicly available at the sources indicated.
Competing interests. The authors are not aware of any competing interests.