A modeling study of temporal and spatial pCO2 variability on the biologically active and temperature-dominated Scotian Shelf

Continental shelves are thought to be affected disproportionately by climate change and are a large contributor to global air-sea carbon dioxide (CO2) fluxes. It is often reported that low-latitude shelves tend to act as net sources of CO2 whereas midand high-latitude shelves act as net sinks. Here, we combine a high-resolution regional model with surface water time-series and repeat transect observations from the Scotian Shelf, a mid-latitude region in the northwest North Atlantic, to 10 determine what processes are driving the temporal and spatial variability of partial pressure of CO2 (pCO2). In contrast to the global trend, the Scotian Shelf acts as a net source. Surface pCO2 undergoes a strong seasonal cycle associated with both a strong biological drawdown of Dissolved Inorganic Carbon (DIC) in spring, and pronounced effects of temperature, which ranges from 0oC in the winter to near 20oC in the summer. Throughout the summer, events with low surface-water pCO2 occur nearshore associated with coastal upwelling. This effect of upwelling on pCO2 is also in contrast to the general assumption 15 that upwelling increases surface pCO2 by delivering DIC-enriched water to the surface. Aside from these localized events, pCO2 is relatively uniform across the shelf. Our model agrees with regional observations, reproduces seasonal patterns of pCO2, and simulates annual outgassing of CO2 from the ocean of +1.9 ± 0.2 mol C m-2 yr-1 for the Scotian Shelf, net neutral CO2 flux of -0.09 ± 0.16 mol C m-2 yr-1 for the Gulf of Maine and uptake by the ocean of -0.88 ± 0.4 mol C m-2 yr-1 for the Grand Banks. 20


Introduction
The global ocean acts as a major sink of CO2 from the atmosphere (e.g., Le Quéré et al 2018;Gruber et al. 2019;Landschützer et al. 2014;Rodenbeck et al. 2015), but it has been suggested that flux density (or flux per unit area) on continental shelves is larger than in the open ocean (Chen et al. 2013;Laruelle et al. 2014). Therefore, compared to their size, continental shelves 25 are thought to disproportionately contribute to global air-sea CO2 fluxes (Laruelle et al. 2010). Additionally, they are susceptible to climate change on much shorter timescales than the open ocean (Cai et al. 2010) and are experiencing increasing impacts of human activity (Cai 2011;Doney 2010;Gruber 2015). Given their high susceptibility to negative impacts from climate change, and their potentially significant contribution to global air-sea CO2 fluxes, it is important to understand the drivers underlying inorganic carbon dynamics on continental shelves. 30 It is generally thought that continental shelves at mid to high latitudes act as net sinks of atmospheric CO2 while those at low latitudes act as net sources (e.g. Chen & Borges 2009;Cai et al. 2006;Laruelle et al. 2014;Roobaert et al. 2019). There are, however, notable deviations from this global-scale pattern. The Scotian Shelf, a mid-latitude shelf off the coast of eastern Canada, is one example with large discrepancies between independent estimates of air-sea CO2 flux (Fennel et al. 2019). Direct measurements made using a moored CARIOCA buoy on the Scotian Shelf indicate that the shelf acts as a net source of CO2 35 to the atmosphere (Shadwick et al. 2010;Shadwick et al. 2011;Shadwick & Thomas 2014). These findings are in contrast to other studies using observations from the SOCAT database, indicating that the Scotian Shelf follows the global trend and acts as a net sink of CO2 (Laruelle et al. 2014;Laruelle et al. 2015;Signorini et al. 2013). These contrasting results for the Scotian Shelf emphasize the large uncertainty inherent in shelf-wide CO2 flux estimates.
Continental shelves are highly complex and dynamic regions where many biological and physical processes modulate 40 CO2 flux (Laruelle et al. 2014;Laruelle et al. 2017;Roobaert et al. 2019). The partial pressure of CO2 (pCO2) in the ocean is one of the key factors which determines the air-sea CO2 flux. Recent global studies found that thermal controls dominate the seasonality of pCO2 but that these alone cannot describe observed pCO2 variations, particularly in temperate and high latitudes (Roobaert et al. 2019). High rates of primary production on continental shelves (Chen & Borges 2009) are another important driver of seasonal changes in pCO2. 45 Continental margins are also subject to intense horizontal transport processes, which act as additional drivers of CO2 fluxes. For example, the Continental Shelf Pump, a term first coined by Tsunogai et al. (1999) in relation to the East China Sea, describes the movement of shelf water high in dissolved inorganic carbon (DIC) across the shelf break to the subsurface open ocean leading to an influx of atmospheric CO2. This mechanism is thought to mainly occur at mid-to high-latitude shelves since it relies on winter cooling to create dense shelf water that is transported to the open ocean's subsurface layers. 50 Upwelling is another well-studied transport mechanism driving shelf-wide CO2 dynamics. The California Current system is a typical example of an upwelling system (Chavez et al. 2017;Hickey 1998;Fennel et al. 2019;Feely et al. 2008). Here, winds drive coastal upwelling, which brings DIC-rich water to the surface along the continental shelf and creates favourable conditions for CO2 outgassing to the atmosphere.
Altogether, these complex shelf dynamics lead to large spatial and temporal variability of pCO2 (Previdi et al. 2009). 55 Such large variability combined with limited data availability for many continental shelves make it difficult to accurately constrain CO2 fluxes. Limited data availability in space and time, often with seasonal biases, is a prime source of uncertainty in flux estimates that can only be overcome with more uniformly distributed sampling. To fully capture how ocean margins are reacting to perturbations caused by the steady input of anthropogenic CO2 to the atmosphere, it is important to understand the processes underlying both spatial and temporal evolution of shelf-wide pCO2. 60 Numerical models can be useful when investigating such complex interactions and constraining CO2 flux since they can interpret sparse measurements through the mechanistic representations of relevant processes. In the present study, we employ a high-resolution biogeochemical model of the northwest North Atlantic to examine the magnitude, variability and sign of the air-sea CO2 flux on the Scotian Shelf. Previous studies have evaluated our model's ability to represent the physical (Brennan et al. 2016, Rutherford & Fennel 2018) and biological (Laurent et al. 2020) dynamics of the region. Here, we focus 65 solely on the model representation of inorganic carbon dynamics, especially the spatial and temporal variability of pCO2 on a seasonal scale on the Scotian Shelf in light of new, high-resolution, shelf-wide observations.
Our overall goal is to show how both biological and transport processes work together seasonally on the Scotian Shelf to set shelf-wide surface pCO2. We additionally discuss event-based variability of the air-sea CO2 flux, and, especially, how short-term, upwelling-favourable wind events throughout the summer create spatial variability of CO2 on the Scotian Shelf. 70 To accomplish these goals, our paper: (1) discusses the seasonal cycle of pCO2 across the shelf; (2) investigates the spatial variability of pCO2, particularly during the summer months and (3) reports shelf-wide air-sea CO2 flux estimates in comparison to previously reported estimates. We discuss the importance of our findings in terms of global patterns of air-sea CO2 flux and carbon cycling.

Study Region 75
The Scotian Shelf ( Figure 1) is uniquely located at the junction of the subpolar and subtropical gyres (Loder et al. 1997;Hannah et al. 2001). Regional circulation is dominated by southward transport of the Labrador Current (Loder et al. 1998;Fratantoni & Pickart 2007). As a result, cool Arctic-derived water accumulates along the northwestern North Atlantic continental shelf separating fresh shelf waters from warmer and salty slope waters (Beardsley & Boicourt 1981;Loder et al. 1998;Fratantoni & Pickart 2007). 80 The Scotian Shelf in particular is controlled by inshore and shelf-break branches of the southwestward moving current. The shelf-break branch inhibits the movement of water across the shelf break of the Scotian Shelf (Rutherford & Fennel 2018). As a result, water moves predominantly along-shelf so that residence times in the region are relatively long, with water being retained on the Scotian Shelf for an average of 3 months before moving further southwest on the shelf (Rutherford & Fennel 2018). In terms of vertical structure, the Scotian Shelf shifts between a two-layer system in the winter, 85 when a cold, fresh layer sits over a warm, salty deep layer, and a three-layer system in the spring and summer, when a warm surface layer forms in the top 20 m above the cold intermediate layer between 20 -100 m, and the warm and salty deep layer .
The Scotian Shelf is additionally characterized by a large, shelf-wide spring bloom initiated in late-March (Ross et al. 2017;Fournier et al. 1977;Mills & Fournier 1979), when the mixed layer is still relatively deep and temperature is at its 90 coldest (Craig et al. 2015). The initiation of the spring bloom in late-March has rapid and large impacts on the observed pCO2 seasonality (Shadwick et al. 2010;Shadwick et al. 2011). Deleted: Figure 1

Physical Model Setup
We employ a biogeochemical model, based on Fennel et al. (2006), Laurent et al. (2020) that is part of the Regional Ocean Modelling System (ROMS, v.3.5;Haidvogel et al. 2008). The physical model implementation, described in more detail in Brennan et al. (2016), has 30 vertical levels and approximately 10-km horizontal resolution (240x120 horizontal grid cells), uses the GLS vertical mixing scheme (Umlauf & Burchard 2003;Warner et al. 2005), 100 atmospheric surface forcing from the European Centre for Medium-Range Weather Forecasts (ECMWF) global atmospheric reanalysis (Dee et al. 2011) and the "high-order spatial interpolation at the middle temporal level" (HSIMT) advection scheme for tracers (Wu & Zhu 2010). Physical initial and boundary conditions are defined using the regional physical ocean model of from the boundaries to a value of zero in the 11 th grid cell from the boundary. Tides are imposed from Egbert & Erofeeva (2002). Climatological river discharge is imposed for 12 major rivers and uses observed long-term monthly means from Water Survey Canada. Full details on the physical model setup and its validation can be found in Brennan et al. (2016) and Rutherford & Fennel (2018). These studies have shown that our model simulates the vertical structure and seasonal cycling of temperature and salinity on the shelf well. The model captures mesoscale features and the coastal upwelling events, and simulates the 110 volume transport of throughout the region in agreement with observation-based estimates.

Biogeochemical Module
The biogeochemical model is based on the nitrogen-cycle model with inorganic carbon component of Fennel et al. (2006) and  but was recently expanded to include 2 phytoplankton and 2 zooplankton functional groups (Laurent et al. 2021). For a detailed description and validation of the biological model, we refer to Laurent et al. (2021), who compared 115 the model output with glider transects of temperature, salinity and chlorophyll and in situ measurements of chlorophyll and nitrate. The model was evaluated on a seasonal scale for the entire model domain, mainly in the surface (top 100 m). Laurent et al. (2021) showed that the model outperforms global models for the region for all variables and that the timing of the spring bloom is well represented, but the model slightly underestimates the magnitude of the bloom and tends to overestimate nitrate throughout the year. 120  Ho et al. (2006) is used and depends on wind speed at 10 m above the sea surface and the Schmidt number. Further details of the biogeochemical model, including the carbonate chemistry equations, can be found in Laurent et al. (2017, Supporting Information). Carbon initialization, boundary conditions and climatological nudging are calculated from relationships with 130 temperature and salinity determined from bottle data for the region. DIC is nudged in an 80-grid-cell wide buffer zone along the eastern boundary, with nudging linearly decaying away from a nudging timescale of 60 days at the boundary to a value of 0 in the 81 st grid cell. At all other boundaries, a 10-grid buffer zone is used, as with temperature and salinity. Use of a wider boundary nudging zone along the eastern boundary was found to be beneficial in imposing low-frequency variability from the Labrador Sea at the northeastern boundary. The nudging zones are not used in the analysis. 135 Nitrate concentrations in rivers are prescribed from Global NEWS model output Seitzinger et al. (2005). DIC and total alkalinity (TA) in rivers were calculated by fitting a linear relationship with salinity from Gulf of St. Lawrence bottle data and extrapolating to river water salinity. The model is initialized on January 1, 1999 from Urrego-Blano and Sheng's (2012) solution for temperature and salinity. Nitrate (NO3 -) concentrations are initialized from the regional climatologies as in Laurent et al. 2020. DIC and TA initial and boundary conditions were created from observationally based relationships with 140 temperature (T) and salinity (S) using bottle data from regional cruises from 1997-2011 encompassing as far south as the Gulf of Maine and as far north as the Labrador Sea (observations from DFO's AZMP program, see: dfo-mpo.gc.ca/science/data-

Taylor Decomposition of Upwelling Events
To better understand the effects of coastal upwelling on surface pCO2, we perform a Taylor Decomposition on the model 150 output during one of the upwelling events focused on in this study, following a similar methodology to Rheuban et al. (2019) and Hauri et al. (2020). Here, we investigate into the influence of T, S, DIC and TA on pCO2 following the equation: where indicates the CO2SYS set of equations. We calculated anomalies, ∆ ! , from a reference value, !,# : The reference values for each variable were calculated as the average of that variable along the Condor transect (see Fig 1) in the upper 40 m (i.e., the part of the water column affected by the upwelling event). We decomposed ∆ ! relatively simply into perturbations related to T, S, DIC, and TA calculated as follows: We refer the reader to Rheuban et al. (2019) for a more detailed description of the Taylor Decomposition methodology.

Observational Datasets
The moored CARIOCA buoy was located at Station 2 on the Halifax Line. Station 2 (HL2; 44.3 o N, 63.3 o W) is located about 165 30 km offshore from Halifax, Nova Scotia, and occupied monthly by Bedford Institute of Oceanography. The buoy measured surface water (at approximately 1 m depth) temperature, conductivity, pCO2, salinity and Chl-a fluorescence every hour and was deployed from 2007 to 2014 with several gaps in data due to calibration and maintenance (see Table S1 in Supplement).
pCO2 was estimated using an automated spectrophotometric technique (Lemay et al. 2018). The raw pCO2 data contained high-amplitude spikes, with increases from 400 μatm to over 1000 μatm within a few hours, which were measuring artifacts 170 and did not represent pCO2 of surrounding water. These spikes were removed by binning all years of the pCO2 observations into a 365-day of year (DOY) seasonal cycle. Any points that were outside 1.5 standard deviations of the 1-month moving average pCO2 where discarded. This method removed only the extreme values and maintained much of the observed variability (see Figure 2).
The sensor-based underway system, Dal-SOOP (Arruda et al., 2020), was installed on the multipurpose platform 175 supply vessel Atlantic Condor (operated by Atlantic Towing Ltd.) and has been measuring a suite of biogeochemical parameters, including pCO2, in the surface water since May 2017. The ship transits weekly to biweekly between the Halifax Harbour (Bin 1) and the Deep Panuke gas platform off Sable Island (Bin 2) on the Scotian Shelf ( Figure 1). The Atlantic Condor pCO2 data underwent standard QA/QC procedures, which included pre-, post-deployment and regular zero-calibration of the pCO2 sensor (Pro-Oceanus Inc, Canada) and associated data corrections. The QC'd data has been deposited into the 180 Surface Ocean CO2 Atlas (SOCAT v.2020), where it was attributed an accuracy of ±10 μatm. Performance of the novel Dal-SOOP system was assessed during a 2-month transatlantic cruise in comparison with a conventional pCO2 equilibrator and showed good agreement with the latter (i.e. −5.7 ± 4.0 μatm; Arruda et al., 2020).
During the QC/QA procedure, some data collected in close proximity to Halifax, and corresponding to the outbound transects, were removed. Some of these data were biased high and attributed to prolonged ship layover in port allowing for a 185 build-up of high pCO2 within the Dal-SOOP system due to respiration. The active pumping that delivers fresh seawater to the measurement system is triggered by a GPS signal when the ship leaves the harbour; as a result, there can be a delayed response from the pCO2 sensor to the much lower pCO2 signals observed immediately outside the harbour. To account for the bias, values that were 2 standard deviations from the mean pCO2 value for the latitudinal bin closest to the Halifax Harbour were removed for some transects. Only three transects were removed.

CO2 Timeseries and Transect
Both the model and observations at the CARIOCA buoy location (see Figure 1) are shown as a seasonal cycle in Figure 2 (chlorophyll, pCO2, temperature and temperature normalized pCO2). The buoy observations show a distinct and recurring 200 seasonal cycle in pCO2. Specifically, pCO2 slightly decreases (from ~450 to 425 μatm) from day 0 to 75. In late March, at approximately day 75, there is a large (100-200 μatm) and rapid (over ~ 25 days) drop of pCO2 associated with DIC drawdown due to the spring bloom (the dashed line indicates the peak in chlorophyll and its alignment with the lowest pCO2 value). This drawdown of DIC occurs while the surface temperature is relatively constant and at its annual minimum.
Following the drop in pCO2 associated with the spring bloom, around day 100, surface water starts to warm, and this 205 warming dominates the pCO2 seasonal cycle with a maximum value of approximately 450-500 μatm reached around day 200-250 (mid to late summer). Around day 250, temperatures and pCO2 start to decrease. Also shown is the temperature-normalized pCO2 using the Takahashi et al. (2002) method for removing the thermal component of pCO2 variations. The biological drawdown of DIC is visible in the temperature-normalized pCO2 during the spring bloom starting around day 75 and a further decline throughout summer from day 150 to 250. This indicates that the overall increase in the non-normalized pCO2 in summer 210 is driven by increasing temperatures, and that biological process tend to draw down DIC during this period.
Most of the Atlantic Condor observations at this location fall within the envelope of the buoy observations' pCO2 seasonal cycle. The monthly mean SOCAT v2020 pCO2 for the entire Scotian Shelf also fall within the spread of buoy observations for most months. Exceptions include February and August when the SOCAT observations are lower than the buoy observations, and September and October when the SOCAT observations are at the low end of the buoy observations. 215 In terms of quantitative metrics, the model (year 2006) at the buoy location has an overall bias of 32.2 μatm and RMSE of 64.0 μatm compared to the buoy data. The model underestimates pCO2 throughout January and February (day 0 -80) partly because its spring bloom starts earlier than in the observations. The bloom-related minimum in pCO2 in the model is approximately 50-75 μatm higher than the buoy observations and approximately 25 -50 μatm higher than the Atlantic Condor observations. Temperature then dominates the pCO2 seasonality in the model over a similar period as in the 220 observations. During the summer (day 150-300), the model overestimates pCO2 but follows a similar cycle as the observations throughout the remainder of the year. The temperature-normalized pCO2 has similar biases (underestimation from day 0-80; overestimation from day 150-300), an RMSE of 66.5 μatm and an overall bias of 23.0 μatm for year 2006. Deleted: Figure 1 Deleted: Figure 2 A comparison of simulated pCO2 with the Atlantic Condor Transect observations along the average ship track ( Figure   1) is shown in Figure 3. Compared to the Atlantic Condor observations, the model (year 2006) has a bias of 13.9 μatm and an RMSE of 28.7 μatm. The model tends towards slightly higher pCO2 across the shelf compared to the ship data, but the bias along the ship track is about half the magnitude as that at the buoy. The seasonal cycle along the ship track (Figure 3) is similar to that at the buoy (Figure 2). The top panel of Figure 3 shows qualitatively good agreement between the model and 230 observations across the whole transect, which is reflected in the averaged pCO2 in the bottom panel. The model does a very good job at representing pCO2 throughout the winter (November through March) but does not reproduce the full spring bloom drop in pCO2 across the whole shelf throughout April as observed. The model also overestimates pCO2 throughout most of June and July. The seasonal cycle across the transect is relatively uniform throughout most of the year; however, there are some exceptions, for example, throughout July pCO2 is relatively low nearshore in both the model and observations. 235

Effects of Upwelling Events
To better understand the effect of physical events on shelf-wide pCO2, this section focuses on the cross-shelf variations in year 2006. Figure 4 shows the evolution of pCO2 along the Atlantic Condor transect throughout the year in both model ( Figure 4a) and observations ( Figure 4b). As in Figure 2 and Figure  During some of these events, temperature nearshore is about 7 o C lower than near the shelf break. These upwelling events and 255 the subsequent low pCO2 signal result in a short-term lowering of air-sea CO2 fluxes nearshore (blue) compared to farther offshore (pink) throughout the summer (at approximately half the flux value nearshore versus offshore throughout July).

9
The top panel in Figure 6 shows a snapshot of surface pCO2 from the model during one of the upwelling events (July 3, 2006; vertical dashed line in Figure 5). pCO2 is relatively uniform across most of the shelf. However, in a narrow band along the coastline, pCO2 values are nearly 100 μatm lower than the rest of the shelf. The bottom panels in Figure 6 show transects 270 of pCO2, temperature and DIC with density contours along the Atlantic Condor transect for the same time slice (July 3, 2006).
In these panels, the density gradients move upwards towards the coastline, consistent with upwelling events. This upwelling  However, in the upwelling region on July 3, temperature has the opposite effect and is acting to decrease pCO2 by ~ 50-60 μatm and DIC is acting to increase pCO2 by only ~ 5-10 μatm from the mean pCO2 in the top 40 m. The effects of alkalinity 285 (∆pCO2,TA) and salinity (∆pCO2,S) are much smaller across the shelf and in both time slices (see Supplement Figure S11).
Comparisons of ∆pCO2,T and ∆pCO2,DIC illustrate that in the upwelled region, anomalies in pCO2 from temperature are larger than those from DIC. However, if water from below 30m was upwelled, DIC would likely start to outweigh the effect of temperature on pCO2.

Regional Flux Estimates 290
The model-simulated air-sea CO2 fluxes, integrated by month and year, and averaged over the simulation from 2006-2014, for the Scotian Shelf and at the buoy location are shown in Figure 8 in comparison to the flux calculated from the CARIOCA buoy observations. The uncertainty in the model estimates is calculated as the standard deviation between years. Annually, the averaged flux between the model and observations is comparable, and the flux estimates at the buoy location are significantly larger than the shelf-wide flux estimates. The model-estimated, annually integrated flux for the Scotian Shelf shows outgassing 295 of CO2 at +1.7 ± 0.2 mmol C m -2 yr -1 . At the buoy location, just outside the upwelling region, the model estimates net outgassing of +2.3 ± 0.1 mmol C m -2 yr -1 . From the buoy observations, the annually integrated CO2 flux is estimated as net outgassing at +1.5 ± 1.4 mmol C m -2 yr -1 . Although our model-derived estimate is within the upper error-bound of the observation-based estimate, it is higher which may be due to the model's overestimation of pCO2, particularly throughout the summer months.
There are also some differences on the seasonal scale. In the model, the Scotian Shelf flux is lower in magnitude than the flux 300 Deleted: Figure 6 Deleted: Figure 6 Formatted: Font: Not Bold Deleted: Figure 4 Deleted: Figure 6 Deleted: Figure 8 at the buoy location during most of the year, and particularly from June to January. Bin 1 along the Atlantic Condor transect (Halifax Harbour/upwelling bin, Figure 1) has an annually integrated flux of +2.2 ± 0.2 mmol C m -2 yr -1 , which is comparable to the annual flux of bin 2 (Deep Panuke/shelf break bin, Figure 1) at +2.0 ± 0.2 mmol C m -2 yr -1 and the simulated flux at the buoy location. These results indicate that cross-shelf variability in air-sea CO2 fluxes is small.  Laruelle et al. (2015), and disagrees with the estimates from Signorini et al. (2013) and Vandemark et al. (2011).

Discussion
We have compared the inorganic carbon dynamics in our medium complexity biogeochemical model of the northwest North Atlantic against two different observational datasets of pCO2, one of them highly resolved in time from a CARIOCA buoy and 320 the other with high spatial resolution along a cross-shelf transect that is occupied approximately bi-weekly. The largest limitation of the model is that it is unable to capture the speed and magnitude of the DIC drawdown associated with the spring bloom throughout March and April (Figure 2 and Figure 3). The simulated pCO2 starts to decline earlier and over a longer period than in both the buoy and transect observations, and the transect shows that this timing is consistent across the whole shelf. Additionally, the model does not reach the observed pCO2 minimum during the bloom across the whole shelf. This 325 discrepancy appears to be a result of the bloom initiation occurring slightly too early and the bloom spanning a longer period of time in the model, and also because chlorophyll levels in the model do not reach the peak values that are observed ( Figure   2a). This limitation aside, the overall seasonal cycle and switch between biological-and temperature-dominated signals in pCO2 are well captured and the model simulates both the seasonal spatial and temporal variability of pCO2 across the Scotian Shelf reasonably well. 330 Notable occurrences of spatial variability of pCO2 on the Scotian Shelf occur throughout the summer months in both the model and observations. With only 1-2 clear examples of lower pCO2 within ~ 25 km of shore in the observations, we used our model to hypothesize about a possible mechanism driving this variability. In the model, we found that coastal upwelling events are driving the summertime spatial variability of pCO2 on the Scotian Shelf and could explain the variability in the observations as well. The physical dynamics of coastal upwelling is well-documented on the Scotian Shelf (Petrie et al. 1987;335 Shan et al. 2016). This upwelling only affects the nearshore region (within ~20-40 km of shore in the model, depending on the event) where water from the cold intermediate layer is transported to the surface. In the model, this creates a coastal band of Deleted: Figure 9 Deleted: Figure 2 Deleted: Figure 3 cold water at the surface that is high in DIC and low in pCO2 (Figure 6). The difference between inshore and offshore temperatures (7 o C and 15 o C, respectively) during these events has a larger influence on the pCO2 spatial variability than the DIC variations (2050 mmol C m -3 inshore and 2020 mmol C m -3 offshore; Figure 6) because the thermodynamic influence of temperature outweighs the effect of a slight increase in DIC, thus lowering pCO2 (see the Taylor decomposition in Figure 7). 345 In the example explored in the present study, the upwelled water comes from ~ 20-25 m depth that has a pCO2 approximately 100 μatm lower than the rest of the shelf. Temperature in the upwelled water is acting to lower pCO2 by ~150 μatm whereas DIC is acting to increase pCO2 by ~ 50 μatm compared to the rest of the shelf. If deeper water was being upwelled to the surface, DIC would likely start to be the dominant factor in setting pCO2 during these events (Figure 7). For the given range of DIC values (2060 to 2020 mmol C m -3 ) and a mean temperature of 11 o C, the thermodynamic effect outweighs the effect of 350 DIC differences for temperature changes larger than 4 o C. Typically, it is thought that upwelling of subsurface waters rich in DIC leads to increased surface pCO2 as is the case for the California Current System (CCS), encompassing the continental shelves off of Washington, Oregon and California, where nearshore outgassing of CO2 during upwelling events is well documented (Fennel et al. 2019, Chavez et al. 2017, Evans et al. 2015, Fiechter et al. 2014, Turi et al. 2014). There are, however, large differences between the Scotian Shelf and the typical upwelling scenario of the CCS. For instance, the size and 355 geometry of these shelves are quite different, which affects the type of water being upwelled to the surface. The California Shelf is an active margin approximately 10 km wide (Fennel et al. 2019) compared to the passive-margin Scotian Shelf with approximately 120-240 km width (Shadwick et al. 2010). As a result, the upwelling in the CCS brings DIC rich water (~2200-2250 μmol kg -1 ) from deep in the water column (below 150-200m) of the open ocean across the shelf break to the surface of the shelf (Feely et al. 2008). On the Scotian Shelf, it is only subsurface shelf water from between ~20-25 m depth that is being 360 upwelled, which is at a similar temperature to the upwelled water in the CCS (7-8 o C) but at a much lower DIC concentration (2050 mmol C m -3 ).
Our regional model shows that upwelling events could be a large contributor to setting the CO2 signal in the summer on the inner portion of the Scotian Shelf, acting to lower pCO2 here and slightly reducing outgassing compared to the outer shelf. Throughout the remainder of the year, the pCO2 distribution across the Scotian Shelf is relatively uniform (Figure 3). 365 Comparison of the inner and outer shelf pCO2 (Figure 4) shows the similar seasonality that is seen across the shelf, both in the model results and Atlantic Condor observations. Additionally, the simulated annual air-sea CO2 flux in bin 1 (upwelling bin, Figure 1) is +2.2 ± 0.2 mmol C m -2 yr -1 and similar to bin 2 (shelf break bin, Figure 1) where the annual flux is +2.0 ± 0.2 mmol C m -2 yr -1 . For comparison, the annual flux for the entire shelf flux is +1.7 ± 0.2 mmol C m -2 yr -1 and the flux at the CARIOCA buoy is 2.3 ± 0.1 mmol C m -2 yr -1 . Our results indicate that the short-term upwelling events in the summer do not 370 significantly affect the shelf-wide fluxes on an annual scale. Overall, the location of the CARIOCA buoy slightly overestimates shelf-wide fluxes but is fairly representative of the shelf-wide pCO2 dynamics overall. According to the model, the Scotian Shelf acts as a net source of CO2 to the atmosphere (+1.7 ± 0.2 mol C m -2 yr -1 ), the Gulf of Maine is a net sink of CO2 (-0.5 ± 0.2 mol C m -2 yr -1 ) and the Grand Banks act as a net sink of CO2 (-1.3 ± 0.3 mol C m -2 yr -1 ). These results are in agreement with Shadwick et al. (2014) for the Scotian Shelf, andLaruelle et al. (2014, 2015) 375 Deleted: Figure 6 Deleted: Figure 6 Deleted: Figure 3 for the Gulf of Maine. Our results disagree, however, with results from other global (Laruelle et al. 2014) and regional studies (Laruelle et al. 2015;Signorini et al. 2013;Vandemark et al. 2011). The discrepancy in reported air-sea CO2 flux between 380 these studies is partly a result of how each study defines the area of the Scotian Shelf and Gulf of Maine. For example, Laruelle et al. (2015) calculates one estimate for both the Scotian Shelf and Gulf of Maine. The shelves of eastern North America are diverse, particularly in width and circulation features, and defining them as a single region is not representative. Additionally, the Scotian Shelf waters are strongly influenced by cold, carbon-rich Labrador Sea water, which is not the dominant endmember south of the Gulf of Maine (Loder et al. 1998, Rutherford & Fennel 2018Fennel et al. 2019). Calculating a single 385 flux estimate for the entirety of this dynamically diverse region is problematic and will yield a different estimate than when considering smaller and more specific regions. However, this only partially explains the difference in flux estimates.
Another reason is that the global SOCAT database was missing important regional data until recently. Signorini et al. (2013) used data from version 1.5 and Laruelle et al. (2014Laruelle et al. ( , 2015 used data from version 2.0 of the SOCAT database. Neither of the observational datasets used in the present study were included in SOCAT versions 1.5 and 2.0. Figure 10 illustrates the 390 difference between different SOCAT versions for seasonal pCO2 on the Scotian Shelf. SOCAT v2020 has consistently higher average pCO2 values than v1.5 and v2, with at least double the number of years and a much larger number of observations going into each monthly average (on the order of 1000 to 10000 measurements in v2020 versus 100 to 1000 in v1.5 and v2).
We believe that flux estimates using the updated SOCAT v2020 will agree better with our estimates and those of Shadwick et al. (2014) since SOCAT v2020 includes more observations with higher spatial and temporal resolution to better capture the 395 distinct seasonal cycle here. Our study, however, only focuses on the recent seasonality of pCO2, making it difficult to distinguish if earlier SOCAT versions miss the regional dynamics solely due to low resolution of observations, or if the estimates from the different SOCAT versions are reflective of a shift in the behaviour of the shelf system. More work should therefore be done to better understand how variability on longer timescales could be affecting regional pCO2 and if that variability could also be a reason for the disagreement between the different SOCAT version. 400 In the present study, we have synthesized and compared our model simulations with high-resolution observations to highlight the dependence of Scotian Shelf pCO2 seasonality on: (1) biological drawdown of DIC during the spring bloom, (2) temperature effects throughout the summer months, and (3) wind-driven coastal upwelling events. In Figure 2d, the temperature-normalized pCO2 shows the non-thermal pCO2 signal, which distinguishes the influence of biological and transport processes on pCO2 (Takahashi et al. 2002). There is a clear decrease of pCO2 associated with the spring bloom. The 405 simulated decrease in pCO2 is smaller than in the observations, likely due to the bloom occurring too early and over a more extended period in the model than the observations. In summer, temperature-normalized pCO2 continues to decrease rather than follow the increasing temperature signal of non-normalized pCO2. Previous studies have noted that, in summer, the thermodynamic signal in pCO2 outweighs the influence of biological activity (Shadwick et al. 2011;Shadwick & Thomas 2014), which could explain the differences in seasonality between pCO2 and temperature-normalized pCO2 in the present 410 study. We believe this thermodynamic influence is an important factor driving the net outgassing observed on the Scotian Shelf, particularly when combined with the delivery of DIC-rich water from the Labrador Sea.  Understanding what processes presently control CO2 dynamics is important for projecting how the region will be affected by changes in climate. Previous studies have suggested that the frequency and intensity of coastal upwelling could increase (e.g., Xiu et al. 2018). In the case of the Scotian Shelf, increased upwelling would lead to less outgassing or even net ingassing during summer along the coast of Nova Scotia. Climate change could therefore disproportionately affect the nearshore region here and lead to an intensification of spatial gradients. Such an upwelling signal would be in addition to the 420 effect of increasing atmospheric CO2, which may be driving the entire Scotian Shelf towards a more neutral system with less outgassing. The effect of the thermal control on Scotian Shelf pCO2 is also an important aspect to consider. As temperatures continue to rise, summer pCO2 values will also likely increase, potentially offsetting some of the effect of increased atmospheric CO2 but also affecting production and respiration rates. Of course, none of these factors act independently and will instead combine to alter both the seasonal and spatial patterns of pCO2 in the region, making the overall outcome of 425 climate-related perturbations on the Scotian Shelf difficult to predict. However, the implementation of a regional model that resolves current conditions well, as in the present study, is an important step towards projecting future climate-related changes in the region.

Conclusions
In this study, we have validated surface pCO2 fields on a seasonal scale from a medium-complexity regional biogeochemical 430 model for the northwest North Atlantic shelf region against pCO2 observations from a CARIOCA buoy and repeated crossshelf transects from a ship of opportunity that crosses the Scotian Shelf. Except for the strength and speed of the pCO2 drawdown associated with the spring bloom, the model simulations represent the observed spatial and temporal variability of pCO2 on the Scotian Shelf well. Contrary to most coastal upwelling systems, upwelling events in summer are acting to lower pCO2 within ~25 km of the coastline, as cold, carbon-enriched intermediate layer water is brought to the surface. The lowering 435 of surface pCO2 during these events occurs because the temperature effect leading to a lowering of pCO2 overwhelms the increase in pCO2 associated with DIC enrichment. We found pCO2 to be relatively uniform across the shelf, with the exception of a narrow band near shore impacted by summer upwelling events. Overall, the Scotian Shelf acts as a net source of CO2 (+1.7 ± 0.2 mol C m -2 yr -1 ), the Gulf of Maine is a net sink of CO2 (-0.5 ± 0.2 mol C m -2 yr -1 ) and Grand Banks acts as a net sink of CO2 (-1.3 ± 0.3 mol C m -2 yr -1 ) in our simulation. Combination of the model simulation and the highly resolved 440 observational data sets emphasizes that the seasonal cycle of pCO2 is driven by strong biological drawdown of DIC in early spring and a dominant thermal control throughout the summer months. Except for the short spring bloom period, surface pCO2 is oversaturated with respect to atmospheric values, which results in net outgassing. Ongoing changes in climate and carbon cycling will likely alter both the seasonal and spatial patterns of pCO2 on the Scotian Shelf.