Sea surface pCO2 variability and air-sea CO2 exchange in the coastal Sudanese Red Sea

The dynamics of sea surface pCO2 (pCOw2 ) and air–sea CO2 exchange of the Sudanese coastal Red Sea has for the first time been studied over a full annual cycle (October 2014–October 2015) based on semi-continuous measurements from moored autonomous sensors. pCOw2 showed a seasonal amplitude of approximately 70 μatm, overlaid by a high frequency (3-4 days) signal of around 10 μatm. The highest values, of about 440 μatm occurred during summer and fall, while the lowest values of about 370 μatm occurred during winter. The monthly pCOw2 change was primarily driven by temperature, i.e., heating and cooling of the water surface. Additionally, Dissolved Inorganic Carbon (DIC) and Total Alkalinity (AT) contributed significantly to the observed change in pCOw2 as a consequence of alongcoast advection and upwelling of CO2-rich deep water, and likely biological production, and uptake of atmospheric CO2. The area is a net annual source for atmospheric CO2 of 0.180 ± 0.009 mol CO2 m−2 y−1. Based on a compilation of historic and our new data, altogether covering the years 1977 to 2015, long term trends of pCOw2 were determined for the seasons winter–spring (1.75 ± 0.72 μatm y−1) and summer -fall (180 ± 0.41 μatm y−1), both weaker than the atmospheric trend (1.96 ± 0.02 μatm y−1). We are suggesting that the study region has transformed from being a source of CO2 to the atmosphere throughout the year to becoming a sink of CO2 during parts of the year. The long term pCOw2 trend was to a large degree driven by increasing DIC, but increasing AT and temperature also played a role. © 2021 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Anthropogenic emissions of carbon dioxide (CO 2 ) into the atmosphere result from burning of fossil fuel and land use change. During the last decade (2009-2018) these processes have been responsible for an input to the atmosphere of 9.5 ± 0.5 Pg C y −1 (Friedlingstein et al., 2019), of which the ocean and terrestrial biosphere have absorbed about 2.5 ± 0.6 and 3.2 ± 0.6 Pg C y −1 , respectively (Friedlingstein et al., 2019). The global carbon budget is based on an enormous amount of data from all over the world, but there is still an ongoing effort to more precisely determine the amount of carbon exchanged between the different reservoirs. Generally, an improved understanding of the ocean CO 2 variability is needed since models seem to underestimate this (Friedlingstein et al., 2019). On a regional scale, the marginal seas are generally not considered in the global carbon budget presented by Friedlingstein et al. (2019), and this is also the exceptionally rich biodiversity which is supported by coral reefs that extends thousands of kilometres along the coast (Klaus, 2015, and references therein). Although the productivity is low, the sea supports fisheries (mainly artisanal) that are an important source of proteins in the region (FAO, 1989). The Red Sea is also known for its recreational diving sites and a dozen popular tourist destinations.
The Red Sea is heavily under sampled and for this reason, only limited knowledge regarding the inorganic carbon cycle and air-sea CO 2 exchange is available in the literature. This is particularly true for the western coastal areas (Rasul et al., 2015). Surveys have typically been conducted along the north-south central axis of the Red Sea, such as: the GEOSECS cruise in 1977 (Weiss et al., 1983), the MEROU cruises in 1982 (Beauverger, 1984a;Beauverger et al., 1984b;Souvermezoglou et al., 1989;Metzl et al., 1989), and the MINERVE cruises in 1991, 1992(Metzl et al., 1995Metzl, 2016). Consistently, these have shown that the Red Sea is a source of CO 2 to the atmosphere, which was later confirmed by Ali (2008), who used the abovementioned data in addition to western coastal Red Sea data from 2007. Souvermezoglou et al. (1989) used two methods (direct and indirect) to determine a carbon budget for the summer and the winter in the Strait of Bab-El-Mandab at the entrance of the Red Sea. They concluded that the Red Sea is a source for CO 2 during both seasons. Metzl et al. (1989) used GEOSECS (December 1977;Weiss et al., 1983) and MEROU (June and October 1982;Beauverger, 1984a;Beauverger et al., 1984b) data to initialize an inverse model for transport and carbon exchange. In their model, the air-sea CO 2 flux evaded into the air during both summer and winter. In Metzl et al. (1995), data from the MINERVE cruises, which primarily covered the Indian Ocean, were presented, and the Red Sea appeared to be a source for CO 2 to the atmosphere, as previous studies had shown. Metzl et al. (1995) showed, however, that the size of the CO 2 source decreased from south to north, and the northernmost part of this enclosed sea was a sink during the summer of 1991.
In this study, we investigate the seasonal dynamics of sea surface partial pressure of CO 2 (pCO w 2 ) and air-sea CO 2 exchange in the Sudanese coastal Red Sea off Port Sudan (Fig. 1) using new carbon and hydrography data. We present, for the first time in this area, the full annual cycle of pCO w 2 and determine the drivers of the seasonal pCO w 2 variations. Further, we determine the longterm trend of the pCO w 2 in the central Red Sea and discuss the associated drivers.

Data
The data used in this study are collected from one mooring equipped with two instruments deployed in Sudanese coastal waters over the period 2014-2015. In addition, we used data from the open sea collected through international expedition (GEOSECS, Weiss et al., 1983;MEROU, Beauverger, 1984a;Beauverger et al., 1984b;Souvermezoglou et al., 1989;Metzl et al., 1989;MINERVE, Metzl et al., 1995;Metzl, 2016;VOS_Pacific_ Celebes, Hydes et al., 2011) and data from a coastal carbon time series ( and 2009( , (Ali, 2017, see Fig. 1 and Table 1. The moored instruments were deployed in October 2014 and retrieved one year later, in October 2015. The mooring, positioned at 37.395 • E, 19.720 • N, was located 18 km north east of Port Sudan at the Sudanese coast ( Fig. 1), at the same position as one of the stations in the coastal carbon time series (Ali, 2017). The moored instruments were a Sea Bird MicroCat  with temperature and conductivity sensors (accuracy ±0.01 • C and ± 0.02 mS cm −1 , respectively) and a SAMI-CO 2 (Submersible Autonomous Moored Instrument) sensor from Sunburst Sensors (pCO 2 accuracy ± 3 µatm based on factory calibration to NIST traceable infrared detector). The bottom depth at the mooring site is about 800 m, and the MicroCat and SAMI-CO 2 were positioned at 37 and 34 m depth, respectively. Temperature and conductivity were determined hourly, while the pCO w 2 was measured every three hours throughout the year of deployment. To evaluate the SAMI sensor pCO w 2 measurements, discrete samples of DIC (Dissolved Inorganic Carbon) and AT (Total Alkalinity) were collected from 40 m depth immediately after the mooring deployment and at the time of mooring retrieval a year later. Unfortunately, the samples collected during retrieval were destroyed during transport. The DIC and AT samples were conserved using saturated HgCl 2 solution, kept cold and dark, and analysed at the Geophysical Institute, University of Bergen, Norway, 7 weeks after collection. DIC was determined using gas extraction of acidified water samples followed by coulometric titration (Johnson et al., 1993;Dickson et al., 2007) with a VINDTA 3C interfaced with a UIC Coulometer (CM5012, accuracy ± 2 µmol kg −1 ), while AT was analysed using an open cell titration system; the Marianda VINDTA 3S (accuracy ± 2 µmol kg −1 ). The accuracy was determined by using certified reference material (CRM) supplied by prof. Andrew Dickson, Scripps Institution of Oceanography, USA. Further, the software CO2SYS (van Heuven et al., 2011) with the carbon system constants of Mehrbach et al. (1973) refitted by Dickson and Millero (1987), KSO − 4 constant from Dickson (1990), and total boron from Uppström (1974) was used to calculate pCO w 2 . Comparison between the SAMI pCO w 2 measurements and the calculated pCO w 2 resulted in an adjustment of the SAMI pCO w 2 measurements of +15.7 µatm. In addition, the mooring pCO w 2 data were compared with calculated pCO w 2 from a coastal carbon time series described in Ali (2017), where monthly averaged DIC and AT over 4-40 m depth and constants as above were used as input to CO2SYS (van Heuven et al., 2011). Average offset between the mooring pCO w 2 and the time series pCO w 2 was 5.3 µatm with a standard error of 4.8 µatm (n=7).
The coastal carbon time series are described in (Ali, 2017), and only a brief description is included here. The time series consist of DIC, AT, temperature and salinity, which were collected between Port Sudan and Sanganeb atoll reef located 30 km northeast of Port Sudan ( Fig. 1 and Table 1). Samples for the coastal time series were collected at 4 m depth from 10 stations in July 2007 (SPS1 section; (Ali, 2017), while over the years 2009-2013, samples were collected from 4 m depth from 5 stations (St1, St2, St3, St5, St6) and from 4-60 m depth from one station (St4) (SPS2 section; (Ali, 2017). The SP2 section was sampled 2-4 times each year, typically February, April, July, and October.  Schlitzer, 2020). The coastal study area is enclosed in the black-bordered box at the Sudanese coast. (b) The location of the mooring between Port Sudan and Sanganeb atoll reef is indicated (yellow circle), together with the coastal carbon time series (SPS1 and SPS2), green circles and red triangles) described in Ali (2017) and Ali et al. (2018). Note that the mooring position is the same as that of station 4 (St4) in the coastal carbon time series SPS2. Ali et al. (2018) examined in detail the seasonal variations of the water masses along the section between Port Sudan and the Sanganeb atoll reef (Fig. 1). They found that the mixed layer depth (MLD) is approximately 40 m deep during most of the year and deeper during winter mixing, and thus, the water shallower than 40 m is part of a homogeneous surface layer. To examine if this was also the case for pCO w 2 , the coastal carbon time series in Ali (2017) was used to calculate monthly averaged pCO w 2 (software and constants as described in Section 2.1) at two depth levels: 4 m and 30-40 m. The difference between calculated pCO w 2 at 4 m and 30-40 m was on average 8.6 µatm, which was half the average variability within the calculated pCO w 2 at 4 m depth (19 µatm). Thus, it is reasonable to assume that the mooring data from 34 and 37 m depth in the current study represents the surface. On a few occasions, the moored instruments were pulled down to 42-48 m, apparent from the Sea-Bird MicroCat pressure data (not shown). This happened e.g., towards the end of July and September 2015; however, no noticeable changes were observed in pCO w 2 , salinity or temperature during these events, and consequently, these data are included in the analyses.

Data processing
The net air-sea CO 2 flux (F CO2 ) was determined using the pCO 2 difference between the seawater (w) and atmosphere (air): where K is the gas transfer velocity and S is the CO 2 solubility in seawater. K is commonly parameterized as a function of windspeed ten metres above the sea surface (U 10 ), and several studies have examined the relationship between U 10 and K (e.g., Liss and Merlivat, 1986;Wanninkhof, 1992;Nightingale et al., 2000;Sweeney et al., 2007;Wanninkhof, 2014). At low wind speeds, the selection of K-U 10 relationship does not have a big impact on the calculated fluxes (e.g. Sweeney et al., 2007), but at wind speeds above approximately 5 m s −1 , the deviation between the various K-U 10 relationships becomes notable. In the current study, where wind speeds typically were less than 10 m s −1 (Fig. 2e), the K-U 10 relationship from Nightingale et al. (2000) was used. This results in K values that are similar to those determined using the relationship by Sweeney et al. (2007) for wind speeds below 10 m s −1 . U 10 , was calculated from daily averaged 6-hourly u-and v-wind velocity components extracted from NCEP/NCAR reanalysis 2 data (https://climatedataguide.ucar.edu/climate-data/ncepreanalysis-r2) at the position 20.00 • N and 37.50 • E (Kalnay et al., 1996). The CO 2 solubility, S, was determined according to Weiss (1974). The factors controlling monthly pCO w 2 changes were quantified according to the framework of Takahashi et al. (1993), where temperature (SST), salinity (SSS), DIC, and AT contribute to the pCO w 2 change: This equation is rearranged to the following form: where τ = 0.0423 • C −1 , η = 0.9, γ = 8.6, and Γ = -7.0. The two first (τ and η) were determined from Takahashi et al. (1993), while the two latter (γ and Γ ), which are the Revelle factors for DIC and AT, respectively, were determined by using the averaged values from 0-40 m deep of the coastal carbon time series (Ali, 2017) as input in the CO2SYS software (van Heuven et al., 2011). Further, to examine the factors controlling pCO w 2 changes, AT and DIC data over a full year are required, according to Eq. (3). Daily AT data were reconstructed from mooring salinity, using the linear regression: AT = 35.3 S + 1070 (R 2 = 0.44), which was developed from the coastal carbon time series described in Ali (2017)  plementary material). The R 2 of 0.44 in our AT-SSS relationship indicate that S can explain only about 40% of the AT variation in the coastal area. Thus, by using our AT-SSS relationship for TA computation possible changes in AT from other processes are neglected. This highlights the need for better understanding of the observed AT variations, however, an in-depth analysis of this is out of scope for this paper. Further, daily DIC was calculated from the reconstructed AT, mooring pCO w 2 salinity, and temperature using CO2SYS (van Heuven et al., 2011). Nutrient concentrations were assumed to be zero in these calculations, which induced an error in calculated pCO w 2 of less than 2 µatm.
The drivers for the observed pCO w 2 changes were determined on a monthly basis, where the change in monthly pCO w 2 represents the difference from one month to the subsequent month.

Results
The summer season is defined as the months May-August while the winter is December-February, as in Ali et al. (2018). Spring and fall months are March-April and September-November, respectively. For the coastal Red Sea, the seasonal cycle is the primary mode of variability (Fig. 2a-c), but there is also higher frequency (3-4 days) variability (Fig. 2c). The seasonal temperature change was about 6 • C (Fig. 2a) while that of salinity was approximately 1.1 (Fig. 2b). The highest temperatures occurred during fall and the lowest at the end of January, while the summer was a transition period with relatively large variations over short times. In mid-May, the temperature abruptly increased to its average summer value of 28.5 • C. The salinity (Fig. 2b) was highest (approximately 40) during fall and winter and lowest (around 38.7) during spring and summer. Several abrupt and large changes in salinity occurred, e.g., in early March and in late August (Fig. 2b). The seasonal peaks and lows for the salinity lagged those of temperature by about 3 months, which, according to Ali et al. (2018) most likely is caused by temperature variation being governed by air-sea heat exchange in addition to horizontal fluxes associated with advection of water along the coast. This transports fresher water from the south into the area of study (Ali et al., 2018), which partly explains the observed salinity variation, , and (f) CO 2 fluxes (mmol CO 2 m −2 day −1 ). Positive ∆pCO 2 and F CO2 indicate gas transport from sea to air. The horizontal ticks represent day one in a month.
while evaporation and precipitation are additional factors, with evaporation dominating.
The pCO w 2 followed the temperature during parts of the year (Fig. 2c), with a seasonal amplitude of approximately 70 µatm, and a higher frequency (3-4 days) signal of about 10 µatm.
The correlation between daily pCO w 2 and SST is relatively strong (R 2 =0.6; Supplementary material, Fig. S2). The highest pCO w 2 occurred during late summer and fall (approximately 440 µatm) when the water was warmest, while the lowest pCO w 2 was observed in January (ca. 370 µatm) when the water was coldest.
During spring, the pCO w 2 value was relatively stable, which was also the case for temperature and to some degree salinity. Abrupt pCO w 2 changes occurred during early January and mid-May, where the first was a sharp and short-lasting peak, most likely connected to wind speed, while the latter was a sharp transition to summer conditions when warmer and less saline surface water was advected from the south. This was also seen as a simultaneously abrupt decrease in salinity (0.5) and increase in temperature (about 2 • C; Fig. 2b).
The atmospheric pCO 2 (pCO air 2 , Fig. 2c) was highest during winter and spring and lowest during summer, with a seasonal amplitude of about 15 µatm, i.e., 25% of that of pCO w 2 . The seasonal cycle of ∆pCO 2 (Fig. 2d, Eq. (1)), was thus governed by the pCO w 2 . The region was undersaturated with respect to CO 2 in parts of the winter while during the rest of the year, the ocean was supersaturated.
The daily CO 2 fluxes (F CO2 ) are shown in Fig. 2f, and the values were periodically low, a result of low wind speeds and periods of near CO 2 equilibrium between ocean and atmosphere. Over the year, the flux was highly variable and mostly directed from ocean to atmosphere, with CO 2 being degassed to the atmosphere at a rate of up to 4.7 mmol CO 2 m −2 day −1 . During winter 2015, a significant CO 2 undersaturation (∆pCO 2 up to −32 µatm) developed and in late January, the area was a sink for atmospheric CO 2 with fluxes as high as -3 mmol CO 2 m −2 day −1 . The strongest CO 2 uptake occurred during events of high wind speeds (Fig. 2e), and over the year of 2014 to 2015, the area was a net source for atmospheric CO 2 of 0.180 ± 0.009 mol CO 2 m −2 y −1 , where the uncertainty is the root sum of squares of the daily uncertainties over a year, and uncertainties of U 10 and pCO w 2 were assumed to be ±1 m s −1 and ±2 µatm, respectively.

Discussion
To quantify the drivers of the observed changes in pCO w 2 , a decomposition analysis as described in Eqs. (2) and (3) was performed, where the driving variables considered were sea surface temperature (SST), sea surface salinity (SSS), DIC, and AT. This is discussed in Section 4.1, while Section 4.2 focuses on the longterm trend in air-sea CO 2 exchange in the Red Sea over a period of nearly four decades from 1977 to 2015.

Factors controlling monthly change in oceanic pCO 2
DIC calculated from mooring pCO w 2 and AT (from SSS) is presented in Fig. 3a while AT is plotted in Fig. 3b. Between September and February, DIC increased likely due to upwelling of colder waters with remineralized organic matter, however, during January also uptake of atmospheric CO 2 (Fig. 2f) contributed to the DIC increase. In February, DIC dropped strongly due to advection of fresher water from the south, described in Ali et al. (2018), and seen as decreasing salinity (Fig. 2b), which was also reflected in the AT in February (shown in Fig. 3b). The counteracting effect of SST and DIC was also seen during July and August, but this time with an opposite sign: warming of the water and decreasing DIC likely due to net primary production.
The decomposition analysis was performed by using monthly averages of the mooring data, with the aim to quantify the drivers of the seasonal changes. However, we are aware of that monthly averages will smoothen the abrupt changes in pCO w 2 (Fig. 2c) as well as the high frequency pCO w 2 variability that occurred on top of the seasonal cycle. The seasonal cycle of AT (estimated from SSS) (blue line) and the accompanying monthly average (blue triangles). Monthly SSS averaged from mooring data (green squares) and are also shown. Lower The seasonal cycle of DIC calculated from mooring pCO w 2 and AT (estimated from SSS) (blue line) and the accompanying monthly average (blue triangles). Monthly SST averaged from mooring data (green squares) are also shown. Fig. 4a shows the monthly changes in measured pCO w 2 over a year, and the plot is similar to that of Fig. 4b, which represents the sum of the calculated drivers. Thus, it is reasonable to believe that the decomposition approach was representative for the situation in the coastal Red Sea.
Changes in temperature (Fig. 4c) was a strong driver for the pCO w 2 changes, but DIC also contributed to the observed changes in pCO w 2 (Fig. 4e). From September 2014 to January 2015, the surface water was cooling (Figs. 2a and 3a), which lowered the pCO w 2 (Fig. 4c). From February to August, on the contrary, the SST increased (Figs. 2a and 3a), and induced a pCO w 2 increase (Fig. 4c). Effects of SSS are overall negligible (Fig. 4d), except in February 2015 (representing the change from February to March), when freshening (∆SSS = 1) of the surface water led to a small negative change in pCO w 2 . This salinity change was described by Ali et al. (2018), who observed that during early February 2015, (their Fig. 4d) the geostrophic current in the area of study was reversed from southwards along-coast to northwards. This introduced water from the south with lower salinity combined with relatively higher temperature into the coastal area.
The strong drop in DIC (Fig. 3a) in February caused a highly negative change in pCO w 2 (approximately -30 µatm; Fig. 4e), which was counteracted by a drop in AT caused by salinity change, corresponding to a positive change in pCO w 2 of more than 20 µatm (Fig. 4f), and, thus, the net pCO w 2 change was very low during this month. The increase in DIC between September and February (Fig. 3a) led to a corresponding positive change in pCO w 2 (Fig. 4e), which to some degree counteracted the negative pCO w 2 change induced by decreasing SST in January.

The long-term pCO 2 trend
The Red Sea is located in the sub-tropical area, which is globally characterized as a net annual source area of CO 2 (Takahashi et al., 2002). We have focused on the central Red Sea delimited by the box 19 -20 • N, 36.2 -39.6 • E (Fig. 5, upper panel) and used data from GEOSECS 1977 (Weiss et al., 1983), MEROU I 1982(Beauverger, 1984aBeauverger et al., 1984b), MINERVE 1991, 1992(Metzl et al., 1995Metzl, 2016), VOS_Pacific_Celebes 2009 (Hydes et al., 2011), coastal carbon time series (Ali, 2017), and the current study to estimate interannual trends of pCO w 2 (Fig. 5).  Table 1. DIC and AT data from the GEOSECS expedition were extracted from GLODAPv2 (Global Ocean Data Analysis Project) synthesis activity in which the data have been adjusted (Olsen et al., 2016;Key et al., 2015). Motivated by the observation that the mooring area emitted CO 2 to the atmosphere during some seasons and absorbed atmospheric CO 2 during winter, the monthly means of pCO w 2 from all cruises are split into winter-spring (GEOSECS, VOS_Pacific_ Celebes, coastal carbon time series, and data from the current study) and summer-fall (MEROU I, MINERVE 1991, 1992, coastal carbon time series, and data from the current study). The mean values and trends of pCO w 2 are plotted in Fig. 5, and it should be bear in mind that both coastal and open ocean  1977, MEROU I 1982, MINERVE 1991, 1992, VOS_Pacific_Celebes 2009, coastal carbon time series (SPS1 and 2), and the current data. Summer-fall (May-November) pCO w 2 data are indicated by open circles and dashed line, while winter-spring (December-April) pCO w 2 data are the filled circles and dotted line. The red colour refers to open sea data, while blue colour refers to coastal data. The monthly averages are based on one station (GEOSECS, MEROU I, VOS_Pacific_Celebes) and multiple stations (MINERVE cruises, coastal carbon time series SPS1 and SPS2, and current mooring data). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) data have been used to extend the number of years, which means that the results must be taken with care. Fig. 5 also include monthly averages of pCO air 2 . These are based on atmospheric measurements from Sede Boker station in the Negev desert from 1995 and onwards. We have extrapolated the atmospheric data series back to 1977 by assuming a growth rate similar to that of Mauna Loa time series. This was checked by first comparing atmospheric xCO air 2 values from Mauna Loa with those from the Sede Boker station. The seasonal amplitude was slightly larger at the latter station, but the interannual trend was similar at the two stations. Thus, it is reasonable to assume that the change in xCO air 2 between 1977 and 1995 at the Israeli station was similar to the change in xCO air 2 observed at Mauna Loa during the same period, i.e., 27 ppm. From this, we estimated the average xCO air 2 value at Sede Boker to 338 ppm in 1977, which corresponds to a pCO air 2 value of approximately 323 µatm (Fig. 5). Then, we could, with confidence, draw a line representing the monthly Red Sea pCO air 2 over the years 1977-2015 (Fig. 5, grey line). The growth rate of pCO air 2 over these nearly four decades is 1.96 ± 0.02 µatm y −1 (R 2 =0.9).  5 shows that, over the years, the pCO w 2 increase regardless of season. The trend of winter-spring pCO w 2 (1.75 ± 0.72 µatm y −1 ) is slightly weaker than that of summer-fall pCO w 2 (1.80 ± 0.41 µatm y −1 ) and both are weaker than the atmospheric pCO air 2 trend. According to Souvermezoglou et al. (1989) and Metzl et al. (1989), the Red Sea was a year around source of CO 2 to the atmosphere. The data used in the current analysis are from both coastal and open ocean, however, we might speculate that, according to our findings of the area being a sink for CO 2 during winter, the year around CO 2 source must have changed at some point. It is difficult to determine when such a transformation might have occurred, but a similar transformation has previously been determined in the Mediterranean Sea. This sea was classified as a source of CO 2 during 1980s, while in the 2000s equilibria between pCO w 2 and pCO air 2 was reached, and at present the ocean is considered a minor sink of CO 2 (Taillandier et al., 2012). Their result was an average of the western and the eastern Mediterranean basin, where the western basin changed from being a weak sink of CO 2 during the 1980s to become a stronger sink in the 2000s, while the eastern basin changed from being a strong source to becoming a weaker source over the same period of time (Taillandier et al., 2012). The Red Sea is similar to the Mediterranean Sea in terms of being a semiclosed sea with relatively high temperature and salinity. Both seas are influenced by the same monsoon system, with northnorthwesterly winds controlled by the Mediterranean weather system reaching all the way to Bab Al Mandab during summer. Thus, it is reasonable to assume that the same factors drive the source to sink transformation in the Mediterranean Sea and Red Sea. During summer-fall, when the pCO w 2 trend (1.80 µatm y −1 ) is less than the pCO air 2 trend (1.96 µatm y −1 ), the area emits a decreasing amount of CO 2 to the atmosphere from one year to the next. Thus, it is reasonable to assume that in the future, equilibrium might be reached between atmosphere and sea surface also during summer and fall. The coastal Red Sea might then be converted to a sink area for CO 2 throughout the year. These results have been compared with reconstructed monthly surface pCO 2 data from the Red Sea at 19.7 • N 37.4 • E. A neural network model (Denvil-Sommer et al., 2019) has been used for reconstruction, and the data were downloaded from Copernicus Marine Service (CMEMS, 2019). Over the years 1985 to 2019, the annual pCO w 2 trend from reconstructed Red Sea data was +1.8 µatm y −1 (Fig. S3 in Supplementary material). When focusing on the winter only (December-February), the pCO w 2 trend was lower (1.5 µatm y −1 ; Fig. S4 in Supplementary material). Thus, over the recent years this area of the Red Sea appeared to be a stronger CO 2 sink during winter compared to summer, which is in line with our results. Over the longer period 1977 to 2015, the amount of surface pCO w 2 data is low, and the assumption of the Red Sea was previously a year around CO 2 source must be drawn with care. The historical winter data are from the open ocean while the coastal data are not available until 2007. Therefore, it might be that the negative CO 2 fluxes are only apparent in the coastal area and not in the open ocean, which can only be unravelled by collecting more data.
Global warming leads to increasing SST, and this is described in Alawad et al. (2020) who found that for the period between 1982 and 2017 the annual mean SST in the northern half of the Red Sea has increased by about 0.04 • C y −1 . This is similar to the warming of the northern Red Sea of 0.4-0.45 • C decade −1 found by Chaidez et al. (2017), but four times larger than the global warming trend of 0.11 • C decade −1 (Rhein et al., 2013). If increasing temperature was the only factor influencing the pCO w 2 , the observed warming of the northern Red Sea would lead to a change in pCO w 2 of about +20 µatm over the years 1977 to 2015.
This would explain about a third of the observed pCO w 2 trend of +1.71 µatm yr −1 . If, on the contrary, the warming rate of Chaidez et al. (2017) of +0.17 • C decade −1 was used (average for the whole Red Sea and taking into consideration that the warming differs from region to region), this would increase the pCO w 2 about 10 µatm over 38 years, i.e.about 15% of the observed pCO w 2 trend. These estimates are, however, theoretical approaches, and it was shown in Fig. 4 that also DIC and to some degree AT contributed to the pCO w 2 changes. Further, higher SST will lower the solubility of CO 2 , and thus slightly increase the CO 2 flux to the atmosphere (Weiss, 1974). This effect is counteracted by an increase in atmospheric CO 2 over time, which will decrease the CO 2 flux to the atmosphere in this area. Increasing SST will also result in more stratification of the upper ocean, which leads to less available nutrients from below and thus, reduced primary production (Behrenfeld et al., 2006;Raitsos et al., 2011;Taillandier et al., 2012). Low vertical mixing in CO 2 source areas like the Red Sea means less CO 2 exported from the deep layer into the surface, and consequently decreasing CO 2 levels in the surface layer. This, in turn, results in CO 2 equilibrium with the atmosphere and transformation into a sink area for atmospheric CO 2 . Quantification of this has not been further elaborated due to lack of data. The long term trend in pCO w 2 divided into the seasons summerfall and winter-spring has been decomposed into drivers (SST, DIC, AT, SSS) similarly to the method described in Eqs.
(2) and (3). Here, we have used the monthly means of pCO w 2 developed for the trend analyses, and further, determined monthly means of SST, SSS, DIC, and AT for all datasets. The datasets, where only pCO w 2 , SSS and SST were measured (MINERVE 1991(MINERVE , 1992(MINERVE , 1999, and the current study) were extended with calculated TA from SSS using a AT-SSS relationship (see Section 2.2) based on data described in Ali (2017), and, further, DIC was calculated from CO2SYS (van Heuven et al., 2011) with AT, pCO w 2 , SST, and SSS as input as explained in Section 2. The results are presented in Fig. 6, and the SSS seems to be of negligible importance for the pCO w 2 change over the years during the winter-spring seasons, while its importance increases slightly during the summer-fall seasons. DIC turns out to be the most important driver regardless of seasons. During the winter-spring seasons, DIC completely dominates over SST and TA, while during the summer-fall seasons, changes in SST and AT also play a role as drivers for the long term pCO w 2 trend. The small and negative trend in SST during winter-fall (Table 2) is in contradiction to the observed warming of the area (Alawad et al., 2020;Chaidez et al., 2017), however, this trend value has a relatively high error (Table 2) and is not significant, likely due to few data before 2009. All year around, but strongest during summer-fall seasons, a positive change in AT induces a negative pCO w 2 change. This is likely connected to the increasing AT observed along the central Red Sea between 1998 and 2018 by Steiner et al. (2018, their Fig. 2), which the authors connected to declining calcification rates in the central and southern Red Sea. An increase in AT over years Ali, 2017) is also seen in the coastal carbon time series (not shown). Table 2 presents the trend values for SST, DIC, AT, and SSS over the years in addition to their uncertainties (1 SD). The uncertainties in Table 2 also reflect the uncertainties in Fig. 6 and is clear from their size that there are no significant trends in SST, SSS, and AT during winter-spring seasons, and thus, SST, SSS, and AT do not have significant effect as drivers during these seasons. The high uncertainties presented in Table 2 also contribute to the differences between the observed (Obs) trend in pCO w 2 (change from one year to next) and the sum (Sum) of drivers.

Summary
We used data from a SAMI-CO 2 sensor moored at about 37 m depth in the coastal Red Sea off Port Sudan to study the annual cycle of oceanic pCO 2 (pCO w 2 ) in the period October 2014-October 2015. The pCO w 2 varied with a seasonal amplitude of approximately 70 µatm with highest values during summer and fall. The pCO w 2 seasonality was overlaid with a high frequency (3-4 days) fluctuation of about 10 µatm. Decomposition analyses revealed that the seasonal changes in pCO w 2 were mainly driven by changes in temperature, although DIC and AT also played a significant role as drivers. The area was a net annual source for atmospheric CO 2 of size 0.180±0.009 mol CO 2 m −2 y −1 , with the spring to fall being the time for emission to the atmosphere, while the winter being the time for CO 2 absorption by the sea surface.
We augmented the mooring data with data from literature all together covering the period 1977 to 2015 and assessed the long-term trends in pCO w 2 , which were determined for the seasons winter-spring and summer-fall of size 1.75 ± 0.72 µatm y −1 and 1.80 ± 0.41 µatm y −1 , respectively. These are weaker trends compared to that of the atmosphere (1.96 ± 0.02 µatm y −1 ), and we speculate that the area has transformed from being a source area for CO 2 to the atmosphere throughout the year to becoming a sink area for CO 2 during parts of the year. A similar transformation happened during the 2000s in the Mediterranean Sea. The long term pCO w 2 trend was primarily driven by increasing DIC during all seasons, and during summer-fall seasons, increasing SST and TA also contributed significantly to the long-term pCO w 2 trend. The increase in AT over the years was likely connected to the declining calcification rates in the central and southern Red Sea. However, the long term DIC and AT trends were highly uncertain due to the scarcity of data, highlighting the need for maintaining a long time series as presented in this study.
The new data presented in this manuscript represents the first semi-continuous pCO w 2 dataset over a full annual cycle. Future elaboration of these data is expected to contribute to improved understanding of the annual cycle of pH and saturation concentration of calcite in this region.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.