Seasonal variability of net sea-air CO2 fluxes in a coastal region of the northern Antarctic Peninsula

We show an annual overview of the sea-air CO2 exchanges and primary drivers in the Gerlache Strait, a hotspot for climate change that is ecologically important in the northern Antarctic Peninsula. In autumn and winter, episodic upwelling events increase the remineralized carbon in the sea surface, leading the region to act as a moderate or strong CO2 source to the atmosphere of up to 40 mmol m–2 day–1. During summer and late spring, photosynthesis decreases the CO2 partial pressure in the surface seawater, enhancing ocean CO2 uptake, which reaches values higher than − 40 mmol m–2 day–1. Thus, autumn/winter CO2 outgassing is nearly balanced by an only 4-month period of intense ocean CO2 ingassing during summer/spring. Hence, the estimated annual net sea-air CO2 flux from 2002 to 2017 was 1.24 ± 4.33 mmol m–2 day–1, opposing the common CO2 sink behaviour observed in other coastal regions around Antarctica. The main drivers of changes in the surface CO2 system in this region were total dissolved inorganic carbon and total alkalinity, revealing dominant influences of both physical and biological processes. These findings demonstrate the importance of Antarctica coastal zones as summer carbon sinks and emphasize the need to better understand local/regional seasonal sensitivity to the net CO2 flux effect on the Southern Ocean carbon cycle, especially considering the impacts caused by climate change.

. Location of the (a) western and northern Antarctic Peninsula and the (b) Gerlache Strait, with a simplified surface circulation pattern (red arrows) that is strongly influenced by the Bellingshausen Sea. The surface circulation in (b) was based on Savidge and Amft 98 . The dashed red arrows represent the modified Circumpolar Deep Water intrusions into the strait, which were identified by Smith et al. 42 , Prézelin et al . 36 and García et al. 43 . The green square depicts the U.S. Palmer Station location (64.8°S, 64.1°W), from which we extracted atmospheric data. The colour shading represents the bottom bathymetry. These maps were generated by using the software Ocean Data View (v. 5.3.0, https ://odv.awi.de) 100 .

Results
Hydrographic properties and the carbonate system. Negative sea surface temperatures were recorded from April to November (Fig. 2a), and the lowest values in summer were observed in the northernmost part of the strait, where the highest salinities were recorded ( Figure S4). The opposite temperature distribution pattern occurred during spring, when the lowest temperatures were recorded at the southern end of the strait. At the connection between the central basin of the Gerlache Strait and the Bellingshausen Sea (i.e., Schollaert Channel), higher temperatures were associated with lower salinity ( Figure S4). On the other hand, the spatial distributions of temperature and salinity in autumn and winter were more homogeneous than those in summer and spring. The carbonate system properties also demonstrated distinct spatial distribution patterns among seasons ( Figures S5-S8). The seasonal variabilities of total alkalinity (A T ) and total dissolved inorganic carbon (C T ) followed that of seawater CO 2 partial pressure (pCO 2 sw ) and were inverse to those of pH and the calcite and aragonite saturation states (Ω Ca and Ω Ar , respectively) throughout the year. A T was higher than C T from December to March and was lower than C T during the rest of the year (Fig. 2d). This seasonal pattern was also observed for CO 2 saturation relative to the atmosphere; i.e., the difference (∆pCO 2 ) between pCO 2 sw and the CO 2 partial pressure in the atmosphere (pCO 2 atm ) was positive from April to November and negative from December to March (Fig. 2b). Minimum pH values (total scale) of 7.99 ± 0.02 were observed in winter, while in the other seasons, they were equal to or greater than 8.00 (Fig. 2c). Undersaturated carbonate calcium conditions (i.e., Ω less than 1) were not observed for either species during the seasonal cycle (Fig. 2c), although the lowest surface values of Ω Ca and Ω Ar were recorded in winter, on average.
In summer, virtually all processes exerted some influence on the surface CO 2 system, as shown by the wide dispersion of the salinity-normalized A T and C T (nA T and nC T , respectively; Fig. 3a). In general, carbonate dissolution seems to exert a greater influence in autumn and winter than in spring and summer, although sea ice growth also acts to control A T and C T in winter. Carbonate dissolution/calcification processes were observed to play a role in changing the A T and C T surface distributions in spring, although sea ice growth and melting processes are also expected to exert an influence, mainly during October and November, in association with low temperatures (Fig. 3d) and high pCO 2 sw . On the other hand, high temperatures (> 0 °C) in spring were associated with an increased influence of photosynthesis on the A T and C T (Fig. 3d).

Drivers of pco 2
sw seasonal changes. C T had the dominant effect on changes in pCO 2 sw throughout the year. A T and temperature were secondary drivers of these changes, while salinity had a minor influence on surface pCO 2 sw (Fig. 4). In summer and spring, there was a considerable decrease in pCO 2 sw , mainly due to the C T drawdown. This decrease was compensated by the increasing effect on pCO 2 sw of the reduction in A T and the increase in temperature. In winter and autumn, the considerable increase in pCO 2 sw was driven by the increase in C T and partially compensated for by the increase in A T and decrease in temperature. net sea-air co 2 fluxes (FCO 2 ). FCO 2 exhibited distinct seasonality throughout the year, with the region swinging from a strong CO 2 sink (FCO 2 < − 12 mmol m -2 day -1 ) in summer to a strong CO 2 source (FCO 2 > 12 mmol m -2 day -1 ) in winter (Fig. 5). During autumn and spring, the behaviour of the region oscillated between the major situations normally observed during winter and summer, resulting in a moderate FCO 2 . Despite this well-marked seasonality, the region was an annual weak CO 2 source from 2002 to 2017, with an average estimated FCO 2 of 1.24 ± 4.33 mmol m -2 day -1 . Notably, with high spatial and temporal variability, this net near-equilibrium condition was achieved because the region switched from a moderate to strong CO 2 ocean sink from December to March to a moderate to strong CO 2 source to the atmosphere throughout the rest of the year (Fig. 5). Months with the most intense CO 2 uptake levels (< − 12 mmol m -2 day -1 ) have occurred more frequently since 2011, with the peak in January and February of 2016. On the other hand, months with the maximum CO 2 outgassing (> 12 mmol m -2 day -1 ) seem to have become less frequent since 2009 (Fig. 5).
Considering all seasons between 2002 and 2017, high seasonal variability in FCO 2 magnitude was identified (Fig. 6). However, the behaviour of the Gerlache Strait as a CO 2 sink or source remained almost consistent within each season, as observed in summer (Fig. 6b) and winter (Fig. 6d). Only two particular exceptions occurred in the autumns of 2011 and 2014, when the region was a weak CO 2 sink (Fig. 6c). Exceptions were also identified in spring, when the region behaved as a strong CO 2 source in 2008 and a particularly strong CO 2 sink in 2010 (Fig. 6e). Although the specific episodes in autumn did not appear to influence the average annual FCO 2 , the unusual spring FCO 2 magnitudes coincided with increases in the average annual FCO 2 in the respective years (Fig. 6a). The Gerlache Strait acted as an absolute annual CO 2 source of 4.4 ± 2.8 mmol m -2 day -1 from 2002 to 2009 and has become predominantly a net annual CO 2 sink of − 2.0 ± 3.0 mmol m -2 day -1 since 2010 (Fig. 6a).
A seasonal pattern in the spatial distribution of FCO 2 along the Gerlache Strait was also identified. This pattern was characterized by a more homogeneous spatial distribution in autumn and winter (Fig. 7b,c) than in summer and spring (Fig. 7a,d). Moreover, the northernmost part of the strait, north of 64°S, had a higher annual FCO 2 (3 ± 8 mmol m -2 day -1 ) than the southernmost part of the strait, south of 65°S. In the southernmost part, there was an annual CO 2 uptake of − 7 ± 16 mmol m -2 day -1 .

Discussion
Seasonal changes in sea-air CO 2 fluxes. In late spring and summer, the Gerlache Strait is a CO 2 sink, with rates ranging from − 13 ± 12 mmol m -2 day -1 in January to − 5 ± 9 mmol m -2 day -1 in March (Fig. 5). This strong CO 2 uptake is driven by an increase in biological activity coupled with meltwater input (Fig. 8a)   www.nature.com/scientificreports/ from December until late summer (Fig. 5), when sea ice formation becomes gradually more intense 9,10 . This is revealed by the substantial C T drawdown (Fig. 4), which characterizes the influence of photosynthesis on the surface water 3,55,56 , associated with a slight decrease in A T as a result of further respiration (Fig. 3b). Phytoplankton growth is favoured by the increased stability of the nutrient-rich shallower mixed layer in summer and late spring (Fig. 8a), mainly due to meltwater input 14,46,53,54,57 . This is more evident in the southernmost part of the strait, where intrusions of warmer mCDW would likely lead to sea ice melting 36 and the higher percentage of meteoric water ( Figure S9) than in the northernmost region, which is comparatively ice-free ( Figure S9). Hence, this could potentially account for the greater CO 2 uptake in the southern region than in the northern region (Fig. 7). Nevertheless, the spatial variability of the carbonate system parameters is clearly greater in spring and summer than in autumn and winter. Therefore, it is likely that other oceanographic processes simultaneously have roles in changing the surface nA T and nC T . In fact, during early spring, the carbonate dissolution/precipitation and sea ice growth/melt associated with low temperatures (Fig. 3d) seem to influence the carbonate system due to the increase in C T that is rejected through the sea ice brine. However, the impact of each of these processes, and even the presence of other involved processes, is not yet well understood. The dominant processes in spring (i.e., carbonate dissolution/precipitation or photosynthesis/respiration), as well as during other seasons, can also exhibit interannual variability. For example, during summer, there is variability in CO 2 uptake oscillating between 2 and 4 years, by which FCO 2 in the region alternates between strong CO 2 sink and near-equilibrium conditions 15 . This variability is associated with both intense biological activity and the intrusion of local upwelled CO 2 -rich waters (e.g., mCDW). In addition, it is linked to the influence of modes of climate variability, such as ENSO, which decreases the wind intensity, leading to favourable conditions for phytoplankton blooms 14 . This explains why the most intense CO 2  Figure S4) total alkalinity and total dissolved inorganic carbon (nA T and nC T , respectively) dispersal diagram for the (a) summer, (b) autumn, (c) winter, and (d) spring. nA T and nC T were calculated for non-zero salinities following Friis et al. 99 . Arrows represent the nA T :nC T ratio that characterizes the physical-biogeochemical processes that affect nA T and nC T (adapted from Zeebe 56 ). The theoretical arrow representing the sea ice growth and melt processes was based on the threshold values for A T and C T described in Rysgaard et al. 64 . More details about the normalization of A T and C T as well as sea ice growth and melt processes are provided in the Supplementary Material. Note that the magnitudes of the axes are different among subplots.
Scientific RepoRtS | (2020) 10:14875 | https://doi.org/10.1038/s41598-020-71814-0 www.nature.com/scientificreports/  www.nature.com/scientificreports/ uptake was recorded in 2016 ( Fig. 5), as this was the year with the most extreme ENSO since 1998 58 , which was associated with biogeochemical changes along the water column 41 . Therefore, the same mechanism underlying the shift in the dominant physical processes may occur in other seasons of the year. This would likely explain why the region was an exceptionally strong CO 2 source in spring 2008 but a strong CO 2 sink in spring 2010 (Fig. 6e).
In autumn, the region becomes a moderate CO 2 source to the atmosphere, with the maximum magnitude in August (14 ± 7 mmol m -2 day -1 ). Such behaviour is due to a significant increase in C T , which leads to an increase in pCO 2 sw . This is further partially offset by the effect that the increase in A T has on pCO 2 sw (Fig. 4), implicating the upwelling process as a likely cause. In fact, more intense short-term irregular intrusions of mCDW 44,59,60 coupled to the deeper mixed layer, which lead to intensified vertical mixing in the winter 61 , are likely to carry CO 2 -rich waters to the surface layer of the strait (Fig. 8b). Indeed, this has been the process most observed in other Southern Ocean coastal regions 8,11,24 . On the western Antarctic Peninsula shelf, for example, there is no evidence of inorganic macronutrient regeneration in late summer, revealing that the increase in C T must be more associated with upwelling and/or advection processes 18 . Although these mCDW intrusions can occur throughout the year and through virtually all connections of the Gerlache Strait 36,42,43 , they are expected to be more intense in winter 61 and at the southern end of the strait 62 . In addition, the rejection of C T through sea ice brine 63,64 is an important process (Fig. 8b). Despite occurring more intensely in winter than in other seasons, this process should also contribute to CO 2 release in autumn, as it was also dominant in controlling A T and C T (Fig. 3c). The increase in C T due to ice growth, first shown in a laboratory experiment 63 , occurs in both Arctic and Antarctic regions, where there is an intense sea ice dynamic 64 . Hence, the increase in C T leads to high pCO 2 sw values but is also related to decreases in Ω Ca and Ω Ar 65 . Thus, these conditions contribute to maintaining a relatively low pH (≤ 8.00) until mid-spring, when sea ice begins to melt and both C T and pCO 2 sw decrease towards the summer season. Although the spatial distribution of FCO 2 is more homogeneous in autumn and winter than in other seasons (Fig. 7), there is intense interannual variability in these fluxes (Fig. 6). It is not yet clear what drives this variability, but it has been linked to sea ice cover variability in other Antarctic regions 8,13,66 . This link makes sense due to the good correlation (r 2 = 0.73; p = 0.0006; n = 12) of the FCO 2 seasonal cycle with the sea ice cover seasonality in www.nature.com/scientificreports/ the Gerlache Strait, mainly in the months when it acts as a CO 2 source (r 2 = 0.93; p = 0.0136; n = 7) ( Figure S10). Despite the strong CO 2 outgassing during these periods, sea ice cover constrains sea-air CO 2 exchanges 8,26 , leading to the conclusion that this CO 2 outgassing could be even more intense under sea ice-free conditions, as observed in the Arctic Ocean 67 . Hence, the FCO 2 dynamics in sea ice-covered periods may be more sensitive than previously thought.
Seasonality of the carbonate system and acidification process. The  . However, we did not find the calcium carbonate in the surface of the Gerlache Strait to be in a subsaturated state, even in winter when there was high pCO 2 sw ; this was also the case in Ryder Bay 18,19 , a region located farther south on the western Antarctic Peninsula shelf, which is under dynamic conditions similar to those of the Gerlache Strait. In summer, carbonate mineral supersaturation is associated with regions where there is strong CO 2 uptake, such as in the southernmost portion of the strait, where meteoric water input is most intense ( Figure S9) and salinity is relatively low (Figures S4 and S7). This reveals that the intense pCO 2 sw drawdown caused by biological activity outweighs the increase in pCO 2 sw by the effect of carbonate precipitation 18 , and carbonate dissolution is minimized due to the biological uptake of C T . Nevertheless, the sensitivity of these parameters should be observed in more detail, as carbonate calcification and dissolution processes also seem to play an important role in controlling A T and C T (Fig. 3b,c). Furthermore, because we found minimum pH values in winter (7.92) lower than those at Ryder Bay in 1994 (8.11) and 2010 (8.00) 7 as well as between 2011 and 2014 (7.95) 19 , these waters may be experiencing ocean acidification, although counterintui- www.nature.com/scientificreports/ tive processes may be offsetting the effects in the studied region 22 . In fact, the waters of the Gerlache Strait have previously been reported to show signs of acidification in summer below the mixed layer 20,22 , with surface pH values lower than those found at Ryder Bay (8.21-8.48 18 ). The effects of intensified summer CO 2 uptake on calcite and aragonite saturation in surface waters may emerge in the coming years. However, supersaturation of these carbonate species is associated with decreased pCO 2 sw values in summer 15 . This reveals that these feedback effects need to be further investigated, especially considering the residence time of these waters in coastal regions. As strong summer CO 2 sink periods are extended, an inverse effect of sea surface acidification may occur, as observed in the southernmost portion of the Gerlache Strait. Nevertheless, the acidification process should occur in the deep layers of these strong CO 2 sink regions and in adjacent deep waters due to horizontal advection. Indeed, this will likely be the case because the residence time of surface waters in this region was estimated to be less than 7 days, while the residence time in adjacent larger basins ranges between 13 and 40 days 50 . Therefore, assuming a steady increase in both atmospheric CO 2 68 and temperature 69 , the Southern Ocean coastal regions may become intense hotspots of deep-ocean acidification, with some expected implications for organisms throughout the water column and the food web as a whole. For example, on the sea surface, there may be a restructuring of the food web due to a shift in the dominant groups Figure 8. Distinct processes driving surface CO 2 partial pressure (pCO 2 ) and seasonal sea-air CO 2 fluxes in a coastal region of the northern Antarctic Peninsula (NAP). From (a) December to March, sea ice melting provides a shallow mixed layer that leads to phytoplankton growth. This spring-summer scenario coupled with less intense modified Circumpolar Deep Water (mCDW) intrusions into the NAP and a decrease in total dissolved inorganic carbon (C T ) from meltwater causes pCO 2 drawdown. Therefore, in these months, the region behaves as a strong sink of atmospheric CO 2 . Conversely, from (b) April to November, under sea ice cover conditions, more intense mCDW intrusions coupled with a deeper mixed layer lead to intensified vertical mixing, resulting in the upwelling of CO 2 -rich waters. Such processes, in association with the rejection of C T through brine release during sea ice growth, lead to a significant increase in surface pCO 2 . Then, the region becomes a moderate to strong CO 2 source to the atmosphere during the autumn-winter. The theoretical depth of the shallowest spring-summer mixed layer is approximately 50 m, reaching approximately 150 m in the autumn-winter 61 24,53 and references therein]. Such changes will potentially decrease the transfer of carbon, energy and nutrients through organisms such as diatoms to pelagic and benthic ecosystems, with complex feedbacks on ocean biogeochemistry and climate 24 . In this sense, these findings shed light on the importance of clarifying the real impacts of these changes throughout the water column. This is because, despite showing signs of acidification, most studies provide only snapshots, and coupled ocean-land-ice processes can mask the real ocean acidification state of Southern Ocean coastal regions.
Annual budget of sea-air CO 2 exchanges. We have identified the Gerlache Strait as a weak CO 2 source from 2002 to 2017, with an annual budget of sea-air CO 2 exchanges at near-equilibrium conditions. This contrasts with the expectations for other Antarctic coastal regions, which demonstrate annual CO 2 sink behaviour 5,13,25 , such as in summer and spring 11,14,33 . The studied region acts as a moderate CO 2 source in autumn and a strong CO 2 source in winter. The CO 2 outgassing that occurs during 8 months of the year (i.e., from April to November) is almost fully compensated for in only 4 months (i.e., from December to March), when the region acts as a moderate to strong CO 2 sink. Although this behaviour is not considered typical for Antarctic coastal regions, the Gerlache Strait lies at approximately 64°S, where Takahashi et al. 70 verified an approximately neutral annual sea-air CO 2 flux. Nevertheless, here, we hypothesize that this scenario is more common to coastal regions of the Southern Ocean than previously thought because incipient signs of this behaviour have already been identified in other Antarctic coastal regions. For example, Bakker et al. 26 found strong supersaturation of seawater CO 2 relative to atmospheric CO 2 in autumn and winter in the Weddell Sea but suggested that the region was an annual CO 2 sink. These contrasting summer/winter behaviours, with an annual CO 2 sink budget, also extend to other Southern Ocean coastal regions, such as the western Antarctic Peninsula 5,7,8,11 , the Ross Sea 13 , the Indian Antarctic sector 6,71 and even the Antarctic Zone south of 62°S as a whole 72 . However, the relatively low monthly and interannual coverage in most of these studies may have biased the integrated FCO 2 budget throughout the year. This is particularly true if we take into account recent estimates of FCO 2 from long-term climatology for global coastal regions 4 . In this climatology, the NAP, as well as the Weddell Sea and much of the Atlantic and Indian sectors of the Southern Ocean, was a net CO 2 source between 1998 and 2015. Despite this, the CO 2 uptake by CO 2 sink regions was so intense that the annual FCO 2 budget for this period was approximately − 17 Tg C year -14 .
Expected scenarios for the future of sea-air CO 2 exchanges. The recent changes observed in the NAP, mainly related to the intensification of the westerly winds 49 , rising temperatures 73 and the prolongation of ice-free water periods 74,75 , are expected to persist in the coming years 24,28 . In this sense, two future scenarios for net sea-air CO 2 fluxes can be projected. First, with longer ice-free water periods, these coastal regions could release CO 2 that would otherwise remain in the seawater isolated by sea ice, intensifying the annual CO 2 source. This release may be enhanced by intensified mCDW intrusion into the western Antarctic Peninsula shelf that have been projected 24,45 , although little is known about its periodicity and variability. On the other hand, nutrient-rich mCDW intrusions coupled with the delayed sea ice cover period and rising temperatures should lead to prolonged phytoplankton growth 75 . Thus, strong CO 2 sink periods should also extend beyond late summer. As CO 2 uptake has intensified in the summer 14,15 and proved to nearly counteract annual CO 2 evasion, this region could become an annual CO 2 sink in future years, particularly assuming that the Southern Ocean is becoming greener 75 . Actually, this second scenario seems likely to occur, as the magnitude and frequency of FCO 2 in months when the region is a strong CO 2 sink are increasing and in months when the region is a strong CO 2 source have been less frequent (Fig. 5), leading to intensified annual CO 2 uptake since 2010 (Fig. 6a). These scenarios become more complex when we take into account the influence of the modes of climate variability. For example, the positive phase of SAM has been associated with more intense CO 2 outgassing due to the deepening of the mixed layer 76 . Conversely, it was also associated with higher CO 2 uptake due to the intensification of upwelling, which supplies iron and nutrients to the sea surface and hence increases phytoplankton growth 77 . This reveals the sensitivity of sea-air CO 2 exchanges to these feedback mechanisms and the urgent need to broaden investigations for a coupled analysis of ocean-climate systems. Nevertheless, signs of intensifying summer CO 2 sink behaviour 14,15 suggest that the influence of SAM should be reversing the flux to encourage annual net CO 2 uptake in Antarctic coastal regions.

Methods
Dataset and carbonate system properties. We used the data available from Surface Ocean CO 2 Atlas version 6 (SOCATv6) 78 to compile a temporal series spanning 2002 to 2017 ( Figure S2) of the sea surface (up to a depth of 5 m) temperature (SST), salinity (SSS) and seawater CO 2 partial pressure (pCO 2 sw ) of the Gerlache Strait. Here, we evaluated the seasonal variability of the net sea-air CO 2 flux (FCO 2 ) and hydrographic and carbonate system parameters. Therefore, the seasons were defined as (1) summer: January to March; (2) autumn: April to June; (3) winter: July to September; and (4) spring: October to December. We analysed the months in which the data covered the majority of the Gerlache Strait in all seasons ( Figure S3).
The pCO 2 sw data extracted from SOCATv6 were directly measured using air-water equilibrators and an infrared analyser for CO 2 quantification 78 . However, SOCATv6 provides surface pCO 2 sw data with only corresponding SST and SSS values. Hence, we used total alkalinity (A T ) from the High Latitude Oceanography Group (GOAL) 79 ; Table S1; 15,22 82 . Using the estimated A T and pCO 2 sw from SOCATv6, we calculated the total dissolved inorganic carbon (C T ), pH and saturation states of calcite (Ω Ca ) and aragonite (Ω Ar ) with CO 2 SYS version 2.1 83,84 . This program determines these parameters from the thermodynamic equilibrium relation between the carbonate species using carbonate dissociation constants. Because of the good response obtained in high-latitude regions 14,15,20,85,86 , we used the constants K1 and K2 proposed by Goyet and Poisson 87 and the sulphate and borate constants proposed by Dickson 88 Table 1). Then, the ΔpCO 2 drv values were separated into categories representing the contributions of differences in C T , A T , SST and SSS. The relative contributions of the drivers changing pCO 2 sw were assessed by converting their relative changes into pCO 2 sw units (μatm) following Lenton et al. 55 as in Eq. 2: where ΔC T , ΔA T , ΔSST and ΔSSS are the differences between the values of the parameters and their respective averages in previous seasons. This analysis was conducted in each year, and the results were averaged to represent an average year. The partial derivatives were calculated using Eqs. 3 to 6 (see details in Takahashi et al. 3 ). These approximations have been widely used in the Southern Ocean 12,21,51 to evaluate pCO 2 sw drivers, both seasonally and spatially. Here, we used the average Revelle and Alkalinity factors of 14 and − 13, respectively. net sea-air co 2 flux (FCO 2 ). We calculated FCO 2 using Eq. 7 4, 90 : where ∆pCO 2 is the difference between pCO 2 sw and atmospheric pCO 2 (pCO 2 air ); K t is the gas transfer velocity, depending on wind speed 91 ; K s is the CO 2 solubility coefficient, as a function of both SST and SSS 92 ; and Ice is a dimensionless coefficient corresponding to the fraction of the air-water interface (between 0 and 1) covered by sea ice. We used monthly averages of pCO 2 air and wind speed (m s -1 ) data from the U.S. Palmer Station, located in the southern part of the Gerlache Strait. The station continuously measures meteorological parameters throughout the year 68 . We calculated pCO 2 air from the monthly averages of the atmospheric molar fraction of CO 2 (xCO 2 air ) and atmospheric pressure (both from the Palmer Station), which was corrected by the water vapour pressure estimated from SST and SSS by the widely used equations of Weiss and Price 93 . Sea ice cover was obtained from the monthly mean of the 0.25° daily satellite products by Reynolds et al. 94 , which cover the entire length of the Gerlache Strait ( Figure S9e-h). Table 1. Average differences (Δ) and standard deviations for the sea surface temperature (SST; °C), salinity (SSS), total alkalinity (A T ; μmol kg -1 ), and total dissolved inorganic carbon (C T ; μmol kg -1 ) involved in seawater CO 2 partial pressure (pCO 2 sw ; μatm) changes. The table shows the differences between the values of the parameters in each season and their respective averages in previous seasons (ΔpCO 2 drv ). www.nature.com/scientificreports/ Spatial distributions of properties. All spatial distribution maps for the properties in this study were interpolated using Data-Interpolating Variational Analysis (DIVA) gridding 95 . We used a length scale value of 15‰ for both the X and Y axes to ensure optimal preservation of data structure and smoothness. The averaging and all other calculations performed in this study were based only on the observed or reconstructed data and not on the interpolated data. Map interpolations were made to provide reader-friendly visualization of the results.

Limitations and uncertainties.
We estimated the propagated uncertainty from the partial derivatives of all calculated parameters (Table 2) in relation to each variable involved in the calculation as follows: where the derived functions f (x) are the calculated parameters (i.e., FCO 2 , C T , Ω and pH) and σ is the uncertainty associated with each variable involved in calculation of the parameter. Because SSS uncertainties are expected to be low enough to be negligible (i.e., < 0.001, according to the GOAL and PANGAEA datasets), they were not considered here. Hence, the propagated uncertainties in C T , Ω and pH fundamentally represented the errors associated with the estimated A T (± 4.4 μmol kg -1 ), SST (± 0.05 °C) and measured pCO 2 sw . We used pCO 2 sw data from SOCATv6 with uncertainties < 2 µatm (55% of total) and < 5 µatm (45%). We calculated the propagated uncertainties for all carbonate system properties with the CO 2 SYS error tool 96 . For FCO 2 , uncertainty was related to the standard error of the averaged wind speed for each season, the measured pCO 2 sw and xCO 2 air , and sea ice cover. The analytical error for xCO 2 air measurements from the U.S. Palmer Station was estimated to be ± 0.07 µmol/mol for the studied period 68 . Sea ice concentrations were computed to a precision of 1% coverage 94,97 .
Finally, we used a first-order polynomial relationship between A T and SSS to estimate A T and calculate the other parameters of the carbonate system based on summertime data. We assumed this relationship for all seasons because the summer was the only period with available A T data for the study region. However, the summer is characterized by greater A T variability than other seasons, implying that the ranges of A T and SSS may represent the annual range (i.e., A T : 2200-2320 μmol kg −1 ; SSS: 32-34.5). Such limitations are mainly due to the scarcity of data in periods other than summer and highlight the need for additional efforts to better understand the dynamics of the carbonate system parameters in coastal regions of the Southern Ocean.
Received: 15 January 2020; Accepted: 20 August 2020 Table 2. Average and standard deviation of the uncertainties propagated in the calculations of the carbonate system properties and net sea-air CO 2 flux (FCO 2 ) for each season. The units of uncertainty are the same as the units of the evaluated parameters: FCO 2 (mmol m -2 day -1 ), total dissolved inorganic carbon (C T ; μmol kg -1 ), pH (total scale) and saturation states of calcite (Ω Ca ) and aragonite (Ω Ar ) (unitless). Standard deviations ~ 0 are smaller than the limit of significant digits in the averages.