Air–sea carbon flux from high-temporal-resolution data of in situ CO2 measurements in the southern North Sea

An important element to keep track of global change is the atmosphere–water exchange of carbon dioxide (CO2) in the ocean as it provides insight in how much CO2 is incorporated in the ocean (i.e. the ocean as a sink for CO2) or emitted to the atmosphere (i.e. the ocean as a source). To date, only few high-resolution observation sets are available to quantify the spatiotemporal variability of air–sea CO2 fluxes. In this study, we used observations of pCO2 collected daily at the ICOS station Thornton Buoy in the southern North Sea from February until December 2018 to calculate air–sea CO2 fluxes. Our results 10 show a seasonal variability of the air–sea carbon flux, with the sea being a carbon sink from February until June switching to a carbon source in July and August, before switching back to a sink until December. We calculated that the sink was largest in April (-0.95 ± 0.90 mmol C m d), while in August, the source was at its maximum (0.08 ± 0.13 mmol C m d). On an annual basis, we found a sink for atmospheric CO2 of 130.19 ± 149.93 mmol C m y. Apart from regionand basin-scale estimates of the air–sea CO2 flux, also local measurements are important to grasp local dynamics of the flux and its interactions 15 with biogeochemical processes.


Introduction
Increased anthropogenic emissions of greenhouse gases (GHGs) lead to global warming (IPCC, 2019), and observing their balance is an important way to keep track of global change (Steinhoff et al., 2019). A key element in this balance is the airsea exchange of CO2 in the ocean, as the oceans are responsible for the uptake of 25% of anthropogenic CO2 emissions 20 (Friedlingstein et al., 2019). The air-sea CO2 flux provides insight in how much CO2 is added to the marine environment from the atmosphere (i.e. the sea being a sink for atmospheric CO2) or emitted by the marine environment to the atmosphere (i.e. the sea being a source). The North Atlantic Ocean is one of the major sinks with an uptake of 680 mmol C m -2 y -1 (Watson et al., 2009(Watson et al., ) in 2005, and between 800 and 4000 mmol C m -2 y -1 (Woolf et al., 2019) in 2010. Continental shelfs are regarded as sinks of carbon with an average air-sea CO2 rate of 1900 mmol C m -2 y -1 for the European continent . 25 However, the Southern Bight of the North Sea (SBNS), i.e. a European shelf sea, is shown as a source area in Thomas et al. (2004). The latter is in contrast with other studies that suggest that the SBNS and the whole North Sea can be regarded as a sink for CO2 (Borges and Frankignoulle, 2002;Laruelle et al., 2018;Schiettecatte et al., 2007). The southern part of the North Sea includes the Belgian Continental Shelf (BCS), which is a well-studied area in terms of air-sea surface dynamics and carbon https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License. biogeochemical cycling (e.g. Borges et al., 2019;Gypens et al., 2011Gypens et al., , 2004. In terms of air-sea carbon fluxes, the BCS shifted 30 from being a carbon source in the 1950s to a carbon sink in the 1980s (Gypens et al., 2009), with more recent source-sink turnovers . Changing seawater physical and biogeochemical characteristics in the BCS result in seasonal patterns of air-sea CO2 flux (Gypens et al., 2004. The dynamic nature of the BCS in terms of annual CO2 fluxes, which were often based on short term measurements and simulated values in past studies, highlight the necessity of high-resolution robust CO2 observations. Therefore, in the present study, we monitored the local dynamics of CO2 flux using high-temporal-35 resolution data of both partial pressure of CO2 in the sea (pCO2, sea) and in the air (pCO2, air). Our aims were to quantify the airsea carbon flux, to identify what drives the seasonality of the flux in a specific year and to identify the annual source-sink dynamics in a specific location of the BCS.

Materials and methods
The North Sea has a surface of 670.000 km² (EEA, 2015) of which the Belgian Continental Shelf (BCS) occupies about 0.5% 40 (or 3.454 km²; Belpaeme et al., 2011). The BCS is relatively shallow with water depths gradually increasing to 45m from the Southeast towards the Northwest (Van Lancker et al., 2015). Apart from extreme observations, surface water temperatures vary seasonally between 5°C and 20°C. The salinity is strongly influenced by the river plumes of the Scheldt, Rhine, Seine and Meuse (Lacroix et al., 2004) and varies between 29 to 35 PSU. https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License.
The Flanders Marine Institute is operating the Fixed Ocean Station "BE-FOS-VLIZ Thornton Buoy" in this area as part of the 50 European research infrastructure "Integrated Carbon Observation System" (ICOS -ERIC, https://www.icos-cp.eu/). The station is equipped with commercial sensors to measure in-situ the sea surface xCO2, atmospheric xCO2, sea surface Temperature (SST), sea surface salinity (SSS) and the total gas pressure of CO2. The BE-FOS-VLIZ Thornton Buoy is located approximately 30 km away from Zeebrugge in the area of the Thornton bank wind turbine farm (51.579N, 2.993E; Fig. 1). In this study, we used observations from the year 2018. A schematic of the mooring and position of the sensors is depicted in 55 Figure A1. The equipment details and data collection information are listed in Table 1. and robust 2-way interactive communication with the buoy system and individual sensors, and provides the means to adapt 65 sampling strategies of the sensors and identify issues very effectively.
The sensors used for this study (SBE37-SMP-ODO and CO2-PRO ATM) are calibrated by the manufacturers once per year.
Additionally the pCO2 measurements of the buoy were validated monthly against calculated pCO2 values from measurements of Total Dissolved Inorganic Carbon (CT), Total Alkalinity (TA) and pH of manually collected samples. Water sampling followed the SOP1 described in Dickson et al. (2007). TA, CT and pH were determined using the methodologies described in 70 Dickson et al. (2007). For TA the method follows SOP3b of Dickson et al. (2007; commercially available system VINDTA 3s). The pH analysis and setup follows SOP6a (Dickson et al., 2007) using the Thermo Scientific Orion pH meter (STAR A211) and ROSS Sure Flow glass body pH electrode and we report pH at Total Scale at 25 o C as measurements are performed in a thermostatic environment (Grant Water Bath). Total Dissolved Inorganic Carbon (CT) is determined using the commercially available Automated Infra Red Carbon Analyzer (AIRICA). For all methods, we use CRMs from Scripps 75 Institute of Oceanography (UCSD). The uncertainties for each method are mentioned in Table 2. For the calculation of pCO2, sea, we have used the R package 'seacarb' (Gattuso et al., 2020). https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License.
The calculated pCO2, sea values were used to calibrate the sensor data using a linear regression method (Fig. A2). The SST and SSS data of the buoy were validated against data obtained from RV Simon Stevin's CTD system (SBE3 & SBE4, respectively for SST & SSS -Seabird Scientific) and underway Thermo-Salinograph sensor (SBE21 -Seabird Scientifics) when visiting 80 the station and collecting samples. The pCO2, air measurements were evaluated against xCO2 data from nearby ICOS atmospheric stations on land. For this comparison, we used the non-parametric Kruskal-Wallis rank sum test and the pairwise Wilcoxon Rank Sum test in the R package 'stats' (R Core Team, 2019).
The air-sea CO2 flux (F) is calculated (Eq. 1) according to the wind-driven turbulence diffusivity model of Nightingale et al. (2000) expressed in partial pressure: where kNightingale is the gas transfer velocity (length • time -1 , Eq. 2), K0 is the solubility of CO2 in seawater (mass • volume -1 • pressure -1 ) and pCO2, sea and pCO2, air are the partial pressure of CO2. We calculated pCO2 by multiplying xCO2 measurements 90 with the total gas pressure of CO2 respectively in seawater or atmosphere. The solubility of CO2 in seawater depends on the sea surface temperature (SST) and the sea surface salinity (SSS; Wanninkhof, 2014 Wind speed (10m above sea level) data were acquired from Meetnet Vlaamse Banken (MVB) for the Westhinder platform, 95 Wandelaar platform and Scheur Wielingen platform, which are located approximately 20 km to 40 km more to the South and Southwest. Wind speed was measured every ten minutes. In the SST and SSS records, there are no data in September 2018 due to a malfunction in the buoy's SBE37-SMP-ODO sensor. To account for the lacking SSS data, we completed our times series with salinity data from by the RV Simon Stevin's CTD system in the same period (Flanders Marine Institute, 2019).
The SST data gaps were completed by data from a second water temperature sensor installed on an Aanderaa Seaguard 100 multiparametric platform (Fig. A1b). Timestamps were used in order to combine data sets from various sensors and systems.
All data were assessed for potential outliers. As in Salgado et al. (2016), outliers were defined as values lying outside the borders of the lower quartile minus three times the interquartile range (Q25 -3*IQR) and the upper quartile plus three times the interquartile range (Q75 + 3*IQR). The daily mean air-sea CO2 flux was calculated from 1891 time points. We took the day-night cycle of the CO2 flux into account by using daily means. Besides, we calculated monthly means and standard 105 https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License. deviation. Our data covers the period from February 2018 until December 2018. To quantify the annual CO2 flux based on eleven months of data, we calculated a weighted mean for the winter, i.e. February and December, and the remaining nine months, respectively, using weights 0.25 and 0.75. We, then, extrapolated the weighted mean to a year. A summary of the input data is provided in Table 3. We investigated if the CO2 flux calculated with xCO2, air from the Thornton Buoy was different from the CO2 fluxes based on xCO2, air measurements at nearby atmospheric stations (Sect 3.1). We compared them 110 using the non-parametric Kruskal-Wallis rank sum test and the pairwise Wilcoxon Rank Sum test in the R package 'stats' (R Core Team, 2019). We adopted the method developed by Takahashi et al. (2002) to separate and assess the seasonal effects of biological processes and temperature on pCO2 and CO2 flux dynamics over an annual cycle. We applied eq. 3 and 4, where Tmean is the mean annual temperature (13.4 °C) and Tobs is the in situ temperature. The relative importance of the components effects is expressed by the thermal-biological ratio (T/B) or the difference (T-B), where T is pCO2, therm and B is pCO2, bio. 120 2, A T/B ratio between zero and one implies the dominance of biological processes over thermal effects (T-B < 0), whereas a T/B ratio larger than one implies that temperature effects are dominant (T-B > 0).

Environmental conditions
In 2018, sea surface salinity (SSS) varied between 32.1 PSU and 34.7 PSU (Fig. 2a) with a mean value of 33.4 ± 0.58 PSU.
The water temperature followed a seasonal pattern with high temperatures in summer time (max. 22.2 °C) and low water temperatures during the winter (min. 3.3 °C; Fig. 2b). No seasonal pattern was observed for the wind speed, i.e. the wind speed is highly variable throughout the year. The lowest wind speed measured was 0.3 m s -1 and the highest was 17.9 m s -1 (Fig. 2c). 130 https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License.
The pCO2, air fluctuated between 389.4 µatm and 464.7 µatm (Fig. 2d). The pCO2, sea data were validated against calculated values of pCO2, sea from DIC and pH values of manually collected samples. After that the pCO2, sea data were corrected with a linear regression method and validated against the manually collected (spot) samples. The pCO2, sea had a large range (126.9 µatm -525.6 µatm), and reached its lowest value in May and highest value in August (Fig. 2d). These observed pCO2, sea concentrations corroborate with data found by Gypens et al. (2011) and Borges et al. (2006) for the 140 English Channel (ECH) and the Southern Bight of the North Sea (SBNS). Borges et al. (2006) found that the spring bloom in early spring was followed by an increase in pCO2, sea in late spring-summer. Schiettecatte et al. (2007) observed that the SBNS was oversaturated in CO2 during winter and strongly undersaturated in April-May. Schiettecatte et al. (2007) reported a minimum pCO2, sea value of 192.35 ± 35 µatm in the SBNS in April and a maximum of 455 ± 36 µatm in August. They observed higher pCO2, sea values for the BCS, up to 900 µatm, but high values were measured close to the Scheldt plume 145 (Schiettecatte et al., 2007). In the present research, we observed a seasonal trend of pCO2, sea, which increased in the summerearly autumn and decreased in the winter-spring. We did not observe a strong seasonality in the pCO2, air record.
To evaluate our atmospheric xCO2 data, we compared data from the BE-FOS-VLIZ Thornton Buoy with 2018 data from nearby (i.e. < 900 km) atmospheric stations (Fig. 3): i.e. two ICOS atmospheric stations Cabauw (207m above sea level; https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License. Frumau et al., 2020) and Tacolneston (185m; O'Doherty et al., 2020) and one atmospheric station in Mace Head (24m; 150 Delmotte et al., 2020). Usually, the CO2 mole fraction data and products of these land-based atmospheric stations are used to calculate air-sea CO2 fluxes, (e.g. Borges and Gypens, 2010). A basic comparison between the different data sets highlights the following. The minimum and maximum CO2 mole fraction registered at the Thornton Buoy in 2018 was 389.4 ppm CO2 and 464.7 ppm CO2. The atmospheric CO2 mole fraction from sampling station Cabauw fluctuated between 394.0 ppm -473.5 ppm CO2, Tacolneston between 386.5 ppm -455.1 ppm CO2 and Mace Head between 394.2 ppm -451.7 ppm CO2. A similar 155 trend was observed in the xCO2, air data of the Thornton Buoy as in the xCO2, air data of the other stations (Fig. 3). Our xCO2, air data is in range with the xCO2, air data of the land-based atmospheric stations (Fig. 3), which supports our use of local field observations. The use of local field observations of xCO2, air at sea provides useful information that complements the use of land-based stations because: 1) the sampling happens close to the water surface where the air-sea carbon exchange occurs, and 2) the xCO2, air observations are more specific to the sampling location than land-based stations. 160

Air-sea CO2 flux
The air-sea CO2 flux was estimated based on the salinity (Fig. 2a), temperature (Fig. 2b), wind speed (Fig. 2c), and pCO2 for seawater and atmosphere (Fig. 2d) time series at the Thornton Buoy in the BCS. We found that the wind speed had a large https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License.
impact on the magnitude of the CO2 flux, i.e. higher wind speed increased the air-sea exchange of CO2 in either way. The daily means of the CO2 flux varied between -2.99 mmol m -2 d -1 and 0.37 mmol m -2 d -1 (Fig. 4). We calculated monthly means 170 (-0.95 ± 0.90 mmol m -2 d -1 to 0.08 ± 0.13 mmol m -2 d -1 ) and distinguished a clear seasonal pattern (Fig. 4). We compared these air-sea CO2 fluxes with air-sea CO2 fluxes calculated with pCO2, air of the atmospheric stations. Only the carbon flux using the atmospheric CO2 data of Cabauw differed from the carbon flux using pCO2, air of the Thornton Buoy (p = 0.031; Fig.3), showing the importance of local atmospheric pCO2 measurements. Overall, the air-sea CO2 flux calculated with different pCO2, air sources, i.e. Thornton Buoy and atmospheric stations, followed a very similar seasonal trend (Fig. 3). 175 Coinciding with other studies in the SBNS, we noted a seasonal effect in the air-sea carbon flux Borges and Gypens, 2010;Gypens et al., 2011;Kitidis et al., 2019;Schiettecatte et al., 2007). The BCS at our location acted as a carbon sink from February until June (-0.95 ± 0.90 to -0.34 ± 0.23 mmol C m -2 d -1 ; Fig. 4). The sink was the largest in April with a monthly mean of 0.95 ± 0.90 mmol C m -2 d -1 . The flux direction switches to a weak carbon source from mid-July and until August with a monthly mean of 0.08 ± 0.13 mmol C m -2 d -1 . However, our findings contradict with the other studies from August onwards. We found that the BCS at our measuring station switched back again to a small sink from September until 185 December (-0.15 ± 0.15 to -0.04 ± 0.10 mmol C m -2 d -1 ; Fig. 4). We believe that the frequency and quality of our local observations allowed us to identify the weak source in July and August, whereas it may have been unnoticed with different https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License. observational capacity, e.g. sporadic sampling cruises. The Thornton Buoy ICOS setup allows for the collection of robust and high-frequency time series observations, , whereas sampling cruises can provide excellent spatial coverage however time resolution can be sporadic (Borges and Frankignoulle, 2002;Schiettecatte et al., 2007). In that respect, it is possible that if 190 samples and observations were obtained during a cruise in autumn over a relatively short period (days or weeks) when CO2 was emitted, then the extrapolation of those observations could have led to the BCS being described as a source in autumn instead of a sink. We also need to acknowledge that environmental factors, e.g. temperature and biological activity can have significant effect on carbon fluxes Thomas et al., 2005Thomas et al., , 2007Wimart-Rousseau et al., 2020). Extreme events, such as the heat wave in the summer of 2018, may have also contributed to some of the differences (e.g. increase in 195 CO2 concentrations) that we present in this study (Borges et al., 2019). Additionally, the solubility of CO2 is lower in warmer water (Wiebe and Gaddy, 1940), reducing the uptake of atmospheric CO2 (Yamamoto et al., 2018). Gypens et al. (2011) also simulated that the North Sea would change to a source for atmospheric CO2 with warmer conditions (biological processes excluded). Other factors, such as wind and input of river plumes, are known to affect the air-sea carbon flux (Arndt et al., 2011;Gypens et al., 2011;Laruelle et al., 2018;Nightingale et al., 2000;Thomas et al., 2005). High wind speed during the 200 winter can amplify the CO2-uptake in this season and so influence the yearly carbon exchange between the atmosphere and the sea (Kitidis et al., 2019). It is known that either temperature driven or biological processes are the dominant driving factor of the pCO2, sea (Schiettecatte et al., 2007;Thomas et al., 2005). In order to determine the main driver of the pCO2, sea dynamics, and as such to quantify the influence of temperature driven and biological processes on the observed CO2 flux, we applied the computational method of Takahashi et al. (2002). 205 We found that on an annual scale the biological activities dominated the pCO2, sea (T/B ratio = 0.69 and T-B = -113.32) and so CO2 flux in the BCS. We also observed that the dominant factor changed by season. For the winter, i.e. February to March and October to December, we found that the thermal effect is dominant (T/B ratio = 1.24 and T-B = 42.28). However, in spring and summer biological processes are dominant over the thermal effect (T/B ratio = 0.84 and T-B = -34.74). Our results 210 correspond with the results of Schiettecatte et al. (2007), who found a T/B ratio of 0.74 (T-B = -70). This is, however, in contrast with the results reported by Thomas et al., (2005) who suggested that temperature rather than biological activity controlled the pCO2, sea dynamics seasonally. The data analysed in Thomas et al. (2005) were collected in four short term cruises and one cruise (i.e. in May) did not consider a CO2 undersaturation (Schiettecatte et al., 2007). This CO2 undersaturation occurs in the declining phase of the phytoplankton bloom and is typically observed mid-April when the bloom is at its peak 215 in the SBNS (Borges, 2003;Borges and Frankignoulle, 2002;Gypens et al., 2004). Based on our high-temporal-resolution measurements, we found that biological activities in BCS controlled the pCO2, sea and consequently the CO2 flux (T/B ratio = 0.69). The high-temporal resolution is important to determine the seasonal variations in the pCO2, sea, CO2 flux and their underlying mechanisms. Linking high-temporal-resolution phytoplankton dynamics with our pCO2 and CO2 flux data (Hilligsøe et al., 2011) may provide new insights in the CO2 flux variation and its underlying drivers. 220 https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License.
According to our data in our location, the air-sea CO2 flux in 2018 was found to be mainly driven by biological processes, and we found that the BCS at our measuring station acted as a sink for atmospheric carbon on an annual scale (-130.19 ± 149.93 mmol C m -2 y -1 ). Our result is in line with other studies, who identified the SBNS as a CO2 sink on an annual scale (Borges and Frankignoulle, 2002;Gypens et al., 2004;Kitidis et al., 2019;Schiettecatte et al., 2007;Fig. 5). 225 Figure 5: The annual air-sea CO2 flux from different studies (letter) with data from 1994 to 2018. Studies a, e, f, g, h and i provide an annual air-sea CO2 flux for the SBNS, whereas studies b, c, d and j for the BCS. Please note that the high values (> 1000 mmol C m -2 y -1 ) of study b and d were located close (< 5 km) to the coast near Zeebrugge. Where possible, the standard deviation (or the standard error in case of study g) is shown by error bars. The horizontal line around a letter indicates that a mean was taken over 230 the indicated period during that study. Gypens et al. (2004Gypens et al. ( , 2011 simulated annual CO2 fluxes in range of our findings, e.g. -170 mmol C m -2 y -1 in 1996 -1999 and -103 mmol C m -2 y -1 in 2002. However, the observed annual carbon sinks in other studies were twice (e.g. -300 mmol C m -2 y -1 ; Borges and Frankignoulle, 2002), to four (e.g. -700 mmol C m -2 y -1 ; Schiettecatte et al., 2007), to 20 times as large (e.g. -2000 mmol C m -2 y -1 ; Kitidis et al., 2019) than our quantifications. Indeed, previous studies show a high inter-annual variability 235 in CO2 flux within the SBNS. Other studies (Borges et al., 2008;Thomas et al., 2004Thomas et al., , 2005Fig. 5) have observed that, in contrast to our study, the southern North Sea was a source of atmospheric CO2 on an annual scale, e.g. 220 mmol C m -2 y -1 (Thomas et al., 2005). It should be noted that many of the CO2 flux data of the southern North Sea are several years old, dating back from 2001(Thomas et al., 2005), 2003(Schiettecatte et al., 2007 and 2015 (Kitidis et al., 2019). These previous studies as well as our study show the high inter-annual variability in the BCS. The high inter-annual variability 240 stresses the need to keep track of the air-sea CO2 flux in a high dynamic area, such as the BCS. Having access to recent and high-temporal-resolution in situ data is important for robust coastal and ocean research but is also useful for policy makers, as https://doi.org/10.5194/bg-2020-442 Preprint. Discussion started: 4 December 2020 c Author(s) 2020. CC BY 4.0 License.
it could refine policy decisions. The carbon fluxes play a major role in the development of the ocean, i.e. ocean acidification (IPCC, 2019) and the global carbon cycle by absorbing anthropogenic carbon emissions (Friedlingstein et al., 2019). Our findings are both in line, i.e. an annual sink for atmospheric carbon, and in contrast, i.e. an annual source for atmospheric 245 carbon, with findings of others studies, demonstrating the high inter-annual variability.
The air-sea CO2 flux does not only vary in time. It also varies in space. Having data on the spatial variability on a local scale, e.g. Thornton Buoy (this study) and Zeebrugge (Borges et al., 2008;Borges and Frankignoulle, 2002;Fig. 5), could be used to assess the spatial variability within a larger area, such as continental shelf seas. Continental shelf seas showed an increase 250 in absorbing atmospheric CO2 and variability within the shelf, but also across different shelf systems (Landschützer et al., 2016;Laruelle et al., 2018). Though, it remains uncertain if the increase in atmospheric CO2 absorption will continue (Legge et al., 2020). As global warming endures, seawater temperature will rise, consequently decreasing the solubility of CO2 (Wiebe and Gaddy, 1940), reducing the uptake of atmospheric CO2 (Yamamoto et al., 2018). In addition, global warming can affect the CO2 uptake indirectly by decreasing or stopping ocean circulation. Less ocean circulation will decrease the nutrient supply, 255 weakening the biological processes and the CO2 export (Yamamoto et al., 2018). The variability (Laruelle et al., 2018) and insufficient quantification (Legge et al., 2020) of air-sea CO2 flux stresses the need for more and extensive in situ observations on local, such as in this study, and global scale (Bozec et al., 2006;Wimart-Rousseau et al., 2020;Woolf et al., 2019). Hightemporal resolution of CO2 flux monitoring is key to gain more knowledge about the inter-annual variability, its drivers and the evolution of CO2 flux. We also suggest extending the observations to investigate the spatial variability in the BCS. 260

Conclusion
We calculated monthly mean air-sea carbon flux at a station in the BCS using high-temporal-resolution data, i.e. daily measured values of pCO2, sea and pCO2, air. By doing so, we revealed a large range of the variability in the air-sea carbon flux (-2.99 mmol m -2 d -1 and 0.37 mmol m -2 d -1 ). The air-sea carbon flux displayed a seasonal pattern, with a sink in the winterspring months, a source in the summer and a small sink in autumn. We measured a carbon sink for atmospheric CO2 in 2018 265 with an estimated uptake of 130.19 ± 149.93 mmol C m -2 y -1 . We advocate for long-term sustained observations, that will allow to improve the quantification of coastal air-sea CO2 flux and constrain the associated variations and drivers.