Seasonal and Interannual Variability of the CO2 System in the Eastern Mediterranean Sea: A Case Study in the North Western Levantine Basin

The seasonal variability of the carbonate system in the eastern Mediterranean Sea (EMed) was investigated based on discrete total alkalinity (AT), total dissolved inorganic carbon (CT), and pH measurements collected during three cruises around Crete between June 2018 and March 2019. This study presents a detailed description of this new carbonate chemistry dataset in the eastern Mediterranean Sea. We show that the North Western Levantine Basin (NWLB) is unique in terms of range of AT variation vs. CT variation in the upper water column over an annual cycle. The reasons for this singularity of the NWLB can be explained by the interplay between strong evaporation and the concomitant consumption of CT by autotrophic processes. The high range of AT variations, combined to temperature changes, has a strong impact on the variability of the seawater pCO2 (pCO2SW). Based on Argo float data, an entire annual cycle for pCO2SW in the NWLB has been reconstructed in order to estimate the temporal sequence of the potential “source” and “sink” of atmospheric CO2. By combining this dataset with previous observations in the NWLB, this study shows a significant ocean acidification and a decrease in the oceanic surface pHT25 of −0.0024 ± 0.0004 pHT25 units.a–1. The changes in the carbonate system are driven by the increase of atmospheric CO2 but also by unexplained temporal changes in the surface AT content. If we consider that the EMed will, in the future, encounter longer, more intense and warmer summer seasons, this study proposes some perspectives on the carbonate system functioning of the “future” EMed.

The seasonal variability of the carbonate system in the eastern Mediterranean Sea (EMed) was investigated based on discrete total alkalinity (A T ), total dissolved inorganic carbon (C T ), and pH measurements collected during three cruises around Crete between June 2018 and March 2019. This study presents a detailed description of this new carbonate chemistry dataset in the eastern Mediterranean Sea. We show that the North Western Levantine Basin (NWLB) is unique in terms of range of A T variation vs. C T variation in the upper water column over an annual cycle. The reasons for this singularity of the NWLB can be explained by the interplay between strong evaporation and the concomitant consumption of C T by autotrophic processes. The high range of A T variations, combined to temperature changes, has a strong impact on the variability of the seawater pCO 2 (pCO 2 SW ). Based on Argo float data, an entire annual cycle for pCO 2 SW in the NWLB has been reconstructed in order to estimate the temporal sequence of the potential "source" and "sink" of atmospheric CO 2 . By combining this dataset with previous observations in the NWLB, this study shows a significant ocean acidification and a decrease in the oceanic surface pH T 25 of −0.0024 ± 0.0004

INTRODUCTION
Since the beginning of the industrial era, the rise in atmospheric CO 2 due to anthropogenic activities is considered to be the main factor responsible for current climate change (IPCC, 2018). The ocean plays a significant role in modulating atmospheric CO 2 as it has sequestrated ca. 31% of the global anthropogenic CO 2 emissions in the past few decade (Gruber et al., 2019). Between 2009 and 2018, the ocean CO 2 sink was estimated to be equal to 2.5 ± 0.6 PgC.a −1 (Friedlingstein et al., 2019). Ocean CO 2 uptake induces an increase in hydronium ion concentration (i.e., a decrease in oceanic pH) commonly referred as ocean acidification (Doney et al., 2009). This ocean acidification represents a significant threat to marine organisms (Kroeker et al., 2013) and is likely to affect marine ecosystems (Feely et al., 2004). The marginal Mediterranean Sea (MedSea) is a singular oceanic basin in terms of carbonate chemistry and deserves specific study. Due to the relatively short residence time of its water masses, this semi-enclosed, basin is considered to be more reactive to external forcing than other oceanic areas (Durrieu de Madron et al., 2011). The warm and highly alkaline waters absorb CO 2 from the atmosphere and transport it to the interior by active overturning circulation (Schneider et al., 2010;Álvarez et al., 2014). Indeed, while representing only 0.3% of the global oceanic volume, the anthropogenic carbon content of the MedSea was estimated to represent 1.1% of the world's ocean content in 1994 (Schneider et al., 2010;Lee et al., 2011). Moreover, several studies have reported a marked decline in the pH of the MedSea over the last few decades (e.g., Touratier and Goyet, 2011;Hassoun et al., 2015b;Palmiéri et al., 2015;Flecha et al., 2019).
Detailed descriptions of the circulation and water masses of the MedSea can be found in Millot and Taupier-Letage (2005), Bergamasco and Malanotte-Rizzoli (2010), and Durrieu de Madron et al. (2011). The water masses of the Eastern Mediterranean Sea (EMed) are warmer, more haline, more oxygenated and more alkaline than those in the Western Mediterranean Sea (WMed) (Álvarez et al., 2014). The EMed water column can be schematically divided into three layers: (1) The surface layer, filled with Modified Atlantic Waters (MAW) with specific regional and seasonal characteristics [e.g., Levantine Surface Waters (LSW)]; (2) Intermediate waters characterised, in the presence of MAW, by a local salinity maximum and generally described by the generic name Levantine Intermediate Waters (LIW); (3) The Eastern Mediterranean Deep Waters (EMDW), mostly retained in the EMed, consisting of a mixture of Adriatic Deep Waters (AdDW) and Aegean Deep Waters (AeDW). EMDW have undergone drastic changes over the last few decades (known as the Eastern Mediterranean Transient; Roether et al., 1996).
The MedSea is already exhibiting a consistent ocean acidification trend as a direct consequence to oceanic CO 2 uptake. It is therefore important to observe carbonate chemistry over sustained time-series to understand the long-term changes in ocean chemistry. The seasonal dynamics of the carbonate system, crucial in understanding the variability in the airsea CO 2 exchanges, also requires these important time-series observations. When compared to other oceanic areas, including the WMed, the oligotrophic EMed  is characterised by low primary production rates (Moutin and Raimbault, 2002). This low productivity reduces the vertical gradients of dissolved inorganic carbon, making the detection and understanding of decadal and seasonal changes in the carbonate system particularly challenging in this area. Over the last few decades, a considerable amount of work has been devoted to the EMed (e.g., Schneider et al., 2010;Álvarez et al., 2014;Hassoun et al., 2015b;Hainbucher et al., 2019), however, these cruises do not cover a full seasonal cycle leading to biased observations. Most of the time-series measurements recorded in the MedSea have been taken in the coastal (e.g., De Carlo et al., 2013;Ingrosso et al., 2016;Kapsenberg et al., 2017) and oceanic WMed (Lefèvre, 2010;Coppola et al., 2018). In the EMed, time-series measurements are scarce and mostly based in the Cretan Sea (Petihakis et al., 2018) or coastal sites such as the Lebanese coast (Hassoun et al., 2019) or the Israeli coast (Sisma-Ventura et al., 2017), precluding a rigorous description of the temporal variability of the carbonate system in the open-ocean EMed.
In the MedSea open-ocean, studies based on data derived from satellite observations have been conducted to decipher, over a seasonal and interannual scale, the variations in pCO 2 (D'Ortenzio et al., 2008;Taillandier et al., 2012). Nonetheless, understanding the variability in the seasonal carbonate system in the EMed is required to evaluate the effects of the increasing threats in this area, such as warming (Nykjaer, 2009) and ocean acidification.
In the frame of the PERLE project (the Pelagic Ecosystem Response to deep water formation in the Levant Experiment), an intense in situ survey of the Levantine area was carried out during 2018(D'Ortenzio et al., 2020. This study reports on a new oceanic inorganic carbon dataset acquired over three different periods of the year in the South Cretan area (described as the North Western Levantine Basin or NWLB hereafter) (Figure 1).
This study gives a detailed description of this new dataset and the oceanographical context (section "Descriptive Carbonate Chemistry in the Context of the PERLE Cruises"). In section "Atypical Drivers of the Seasonal Dynamics of the Carbonate Chemistry Within the Mixed Layer of the North Western Levantine Basin, " using these new annual observations in the NWLB, the physical and biological drivers explaining the seasonal variability of the carbonate parameters in the upper water column will be investigated and the impact of the variations on air-sea CO 2 fluxes will be discussed. In section "Long Term Temporal Changes in Carbonate Chemistry in the North Western Levantine Basin, " the main drivers of carbonate chemistry changes will be considered on longer timescales, based on the estimated trends in the surface carbonate chemistry of the NWLB derived from existing data over the last 20 years. Some hypotheses on the future of the carbonate system functioning of the EMed will be discussed.

Cruise and Sampling Strategy
This study focuses on three PERLE cruises: PERLE0, PERLE1, and PERLE2 (Figure 1). These cruises were carried out in the EMed between 2018 and 2019. At all stations, a CTD-Rosette was deployed (1) to acquire data with sensors (Conductivity Temperature and Depth-CTD and associated parameters) along vertical profiles and (2) to collect discrete seawater samples from Niskin bottles for chemical analysis. Over the 11, 31, and 125 casts performed during the PERLE0, PERLE1, and PERLE2 cruises, seawater was sampled from 1, 12, and 17 casts, respectively, for carbonate parameter analysis (see Supplementary Table 1 and Supplementary Figure 1). Details for the cruises and parameters measured during each PERLE cruise are summarised in Table 1.

CTD and Seawater Sampling
A SeaBird TM 911+ underwater unit was used to interface a pressure sensor, an external temperature probe (SBE3plus) and an external conductivity cell (SBE4C). Sensors were calibrated by the manufacturer. Additional sensors were interfaced and data from a fluorescence (Chelsea Aqua 3) and an oxygen (SBE43) sensor are used in this study. Fluorescence and oxygen are expressed in A.U. (Arbitrary Unit) and µmol.kg −1 , respectively, in this study. For vertical profiles, 24 Hz data on the downcast were averaged on 1 dbar bins by the SeaBird TM dedicated software. Water samples were collected from CTD-Rosette casts with a carousel equipped with 22 Niskin bottles (12 L). Water was sampled from 10 to 21 depths, from a few meters above the seafloor up to the surface (0-5 dbars). From 0 to 200 dbars, a higher sampling resolution was applied (every ca. 20 dbars) than below 200 dbars (every ca. 200 dbars).
In addition, the "Real-time" CTD data from the WMO 6902913 Argo float (Argo, 2000) deployed during the PERLE1 cruise were used in this study to complete the hydrological data. Data collected from October 2018 to July 2020 were used (Figure 1). Because the Argo float considered in this study is still operational, no "Delayed Mode" data were available at this stage. The Argo real-time quality control procedures have been applied by the Coriolis data centre (Wong et al., 2020). A visual comparison of the Argo CTD data with collocated PERLE cruise CTD data was carried out on two profiles to exclude major deviations in the Argo data. Salinity measurements (derived from conductivity-SBE41CP sensor, Seabird TM ) were recorded with an accuracy of 0.005 psu.

Total Alkalinity and Total Dissolved Inorganic Carbon
Samples for total dissolved inorganic carbon (C T ) and total alkalinity (A T ) were collected into acid-washed 500 cm 3 borosilicate glass bottles, poisoned with 200 mm 3 of a 36 g.dm −3 HgCl 2 , as recommended by Dickson et al. (2007) and stored in the dark at 4 • C. Analyses were performed after 5 months of storage. Measurements of C T and A T were performed simultaneously by potentiometric acid titration using a closed cell following the methods described by Edmond (1970) and Dickson and Goyet (1994). Analyses were performed at the National Facility for Analysis of Carbonate System Parameters (SNAPO-CO2, LOCEAN, Sorbonne University-CNRS, France) with a prototype developed at LOCEAN. The average accuracy of A T and C T analysis (estimated from repeated measurements of Certified Reference Material provided by Prof. Dickson's laboratory from the Scripps Institution of Oceanography, San 1 | Summary of the cruise information and the parameters measured during each PERLE cruises including availability, number of samples (n) and their associated accuracy.
Diego) was 1.8 and 2.1 µmol.kg −1 , respectively, for PERLE0 and 4.6 and 4.7 µmol.kg −1 , respectively, for PERLE2. Although A T and C T measurements were carried out during the PERLE1 cruise, the accuracy of the dataset did not conform to the quality control procedure (see section "Primary Quality Control of the Measured Data") therefore the measured PERLE1 A T /C T dataset was not used in this study. However, A T values were reconstructed for PERLE1 based on a published A T -S relationship (see section "Derived Parameters").

pH
The pH was measured directly on board. Samples for pH measurements were collected in cylindrical optical glass vials and analyses were performed manually using purified m-Cresol Purple (mCP) following the spectrophotometric protocol (at 25 • C) described by Clayton and Byrne (1993) (see details in Supplementary Material). This method is based on the dissociation of the pH-sensitive mCP dye (provided by Prof. Byrne, University of Southern Florida) in the water sample. pH is reported on the total scale at 25 • C (pH T 25 ) using the equation by Liu et al. (2011). The reproducibility of measurements was estimated to be ± 0.0009 by measuring replicates from the same Niskin bottle. The accuracy was determined to range within ± 0.007 for PERLE1 and ± 0.003 for PERLE2 by analysing replicates of TRIS solution (provided by Prof. Dickson, Scripps Institution of Oceanography, San Diego). No direct pH measurements were carried out during the PERLE0 cruise. The effect of the addition of the indicator on the seawater pH was evaluated and corrected (see details in the Supplementary Material).
The recommendations of Langdon (2010) were followed for sampling, reagent preparation and sample analysis. The thiosulfate solution was calibrated by titrating it against a potassium iodate certified standard solution of 0.0100 N (CSK standard solution-WAKO). The reproducibility of measurements, calculated by measuring replicates from the same Niskin bottle, was estimated to be ± 0.86 µmol.kg −1 (n = 42, PERLE2).
Oxygen measurements from the SBE43 sensor from the CTD rosette were systematically adjusted for all cruises with the "Winkler" values on the whole water column. Based on the raw data processing algorithm (Owens and Millard, 1985), 3 calibration coefficients were adjusted (the oxygen signal slope, the voltage at zero oxygen signal and the pressure correction factor) by minimising the sum of the square of the difference between the Winkler oxygen values and oxygen derived from the sensor signal. The accuracy of the SBE43 adjusted values is around ± 2 µmol.kg −1 .

Nutrients
Samples for dissolved inorganic nutrients were collected from Niskin bottles in 20 mL polyethylene bottles. Samples were analysed directly on board during PERLE2 and frozen before analysis on land for PERLE0 and PERLE1. Analyses were performed after less than a month of storage. All nutrient samples were analysed by a standard colorimetric method on a segmented flow analyser (Autoanalyser II Seal Bran& Luebbe R ) following Aminot and Kerouel (2007). The relative precision of these analyses ranged from 5 to 10% (Aminot and Kerouel, 2007).

Primary Quality Control of the Measured Data
Systematic primary quality control of the measured data was performed on each PERLE dataset. No significant problems have been detected for Winkler oxygen and pH measurements. During PERLE1, for a few casts, a CTD pump dysfunction significantly altered the quality of the CTD oxygen: oxygen measurements from these casts were disregarded. A systematic quality control procedure for A T and C T was conducted based on internal consistency tests between A T , C T and pH T (see details in the Supplementary Material). Following these steps, only 15 PERLE2 casts were validated, leading to the loss of ca. 60% of the PERLE2 A T /C T dataset. All the A T /C T PERLE1 dataset was lost. A comparison of the quality controlled PERLE dataset with previously collected data does not reveal systematic biases for A T , C T , or pH T 25 (Figure 2A-C).

Statistical Tests on the Linear Model
Relationships between years and carbonate parameters (A T , C T , and pH T 25 ) and between A T and salinity were computed using a linear regression model. Linear regression statistics, including the standard error of the slope (i.e., the error of the estimated trend), the coefficient of determination (r 2 ) and the significance of the trend (p-value) were calculated using the R software. Linear relationships have been tested using the Pearson coefficient for parametric test (Sokal and Rohlf, 1969) with a significance level of 95%.
Parameters derived from the A T -S linear relationship were tested against previously published A T -S relationships in the area using a Student's t-test for the slope and intercept. The null hypothesis, H 0 , was that our observations were not significantly different from these linear models.

Derived Parameters
Absolute salinity (S A ), conservative temperature ( ) and potential density (σ θ ) were derived from practical salinity, temperature and pressure and the geographic position based on the TEOS-10 (The International Thermodynamic Equation of Seawater-2010). In this study, following the recommendations of the Intergovernmental Oceanographic Commission (Valladares et al., 2011), S A and were used to study the hydrological context ( − S A diagrams). Calculations were made with the "oce" R package (Kelley et al., 2017). Note that practical salinity (labelled Salinity) and in situ temperature (labelled Temperature) were used in this study to facilitate comparisons with previous studies in particular, for A T -S relationships.
Apparent Oxygen Utilisation (AOU-µmol.kg −1 ) was calculated from the difference between oxygen solubility concentration (at P = 0 dbar) estimated with the "Benson and Krause coefficients" (Garcia and Gordon, 1992) and in situ A density threshold of 0.03 kg.m −3 with a reference depth of 10 dbars was used to compute the Mixed Layer Depth (MLD) (D'Ortenzio et al., 2005).
Salinity data were used to reconstruct an A T time-series using the sub-surface A T -S relationship proposed by Hassoun et al. (2015a) (see discussion in section "Total Alkalinity and Salinity Relationships Within the Mixed Layer"). In this study, the PERLE1 and the Argo float A T datasets were reconstructed following this A T -S relationship. Considering the standard deviation of the A T -S relationship proposed by Hassoun et al. (2015a), the accuracy of the calculated A T values is ± 19 µmol.kg −1 .
Salinity-normalised changes in A T (NA T 39.3 ) and C T (NC T 39.3 ) were calculated dividing by in situ salinity and multiplying by 39.3 (i.e., the mean PERLE salinity above 200 dbars).
Seawater carbonate system parameters were derived from A T and C T values. Calculations were made with the software program CO2SYS-MATLAB ( van Heuven et al., 2011) using silicate and phosphate concentrations. When nutrient data was not available, silicate and phosphate mean concentrations for each depth were used. As recommended for the MedSea by Álvarez et al. (2014), the carbonic acid dissociation constants K 1 and K 2 from Mehrbach et al. (1973) as refitted by Dickson and Millero (1987) and the dissociation constant for HSO 4 − from Dickson (1990) were used. Uppström (1974) was used to calculate the ratio of total boron to salinity and Dickson and Riley (1979) to calculate the hydrogen fluoride constant K F .
The buffer factors γA T (γC T ), βA T (βC T ) and ωA T (ωC T ) provide an estimation of the seawater's ability to buffer changes in the aqueous CO 2 [CO 2 ], protons [H + ] and the carbonate saturation state ( ) when A T (C T ) changes at constant C T (A T ) (Egleston et al., 2010). The calculations were performed following the formula proposed by Álvarez et al. (2014).

Quantification of Biological Processes
Net Ecosystem Production (NEP) is defined as the sum of biotic and abiotic carbon fluxes in the ecosystem (Borges et al., 2008). Net Ecosystem Calcification (NEC) is a measure of the balance between CaCO 3 formation (calcification) and dissolution (Smith and Kinsey, 1978). Based on the NA T

39.3
and NC T 39.3 plot, the reaction path can take on variable slopes depending on the ratio of different processes, such as photosynthesis/respiration, carbonate dissolution/formation and CO 2 release/invasion (Zeebe, 2012 Following equation (2), NEP can be expressed according to NEC as: Then, by replacing the NEP term in equation (1) by equation (3), NEC can be calculated as: NEC and NEP are expressed in µmolC.kg −1 .d −1 . Salinitynormalised A T and C T values "exclude" the "precipitationevaporation" influence in the layer where biological activity is at a maximum. It is assumed that the layers considered (MLD-200 dbars) to estimate the NEP and NEC processes are not influenced by air-sea CO 2 fluxes, which were therefore not considered. Colored points correspond to A T on (D), to C T on (E) and to pH T 25 on (F). Isopycnal horizons based on potential density referenced to a pressure of 0 dbar (σ θ ) are represented by grey contour lines. On (D-F), different dots have been used for each PERLE cruise. Because no A T and C T data were available for PERLE1 cruise, only pH T 25 data have been represented (C,F).

CARIMED Database
CARIMED (CARbon, tracer and ancillary data In the MEDsea) aims to be an internally consistent database containing inorganic carbon data relevant for this basin (Álvarez et al., in preparation). Ancillary (hydrographic, inorganic nutrients and dissolved oxygen), CO 2 (pH, A T , and C T ) and transient tracer (CFC-11 and 12, Tritium, SF 6 , Neon, CCl 4 , and He 3 ) data from several cruises in the MedSea from 1976 until 2018 were assembled. Primary and secondary quality control procedures following the GLODAP (Global Ocean Data Analysis Project) philosophy  are locally adapted to this marginal sea. This work only uses data collected in the Levantine basin (Supplementary Table 2).

DESCRIPTIVE CARBONATE CHEMISTRY IN THE CONTEXT OF THE PERLE CRUISES
Carbonate Chemistry Along the Water Column Below the Surface Layer Intermediate waters (mostly LIW) were located around the 29.0 kg.m −3 isopycnal layer (Lascaratos and Nittis, 1998; see Supplementary Figure 1) and were characterised by an A T maximum evolving from 2,600 to 2,640 µmol.kg −1 (Figures 2A,D). As observed by Álvarez et al. (2014), the LIW was located above the layer of maximum organic matter mineralisation in the EMed and was associated with low C T concentrations (ca. 2,290 µmol.kg −1 ) and high pH T 25 values (ca. 8.000) in contrast to the deepest water masses. It can be observed that slightly colder, more haline and denser Cretan Intermediate Waters (Velaoras et al., 2019) were detected during PERLE2 in the western part of the Cretan Sea with the highest A T value for PERLE2 cruise (ca. 2,660 µmol.kg −1 , Figure 2A).
In the deep-water layer (i.e., EMDW), both AeDW and AdDW presented similar C T values ( Figure 2E) while slightly higher pH T 25 ( Figure 2F) and A T (Figure 2D) values were measured in the AeDW (see Supplementary Figure 2). On the Cretan shelf, deep waters were comprised of dense EMDW with high A T (≈ 2,650 µmol.kg −1 ) and C T values (≈ 2,350 µmol.kg −1 ). Deep waters of the Cretan Sea were filled with CDW with low pH T 25 (≈ 7.950) values resulting from relatively low A T and high C T content (Figures 2D-F). This description of the carbonate chemistry in the deep and intermediate water masses in the PERLE area is in good agreement with previous studies (Schneider et al., 2010;Álvarez et al., 2014). However, the PERLE strategy based on an intense observation period over a year is not appropriate to describe changes in deep-water masses. For the rest of this study, in order to tackle the seasonal dynamics of the surface waters, only data in the NWLB (Figure 1) where all three PERLE cruises were conducted, will be discussed further.

Seasonal Variability in the Upper Water Column
The highest spatial and temporal variability in carbonate chemistry parameters was encountered in the upper water layer which has been defined to be approximately the first 200 dbars. Discrete pH T 25 values (measured and calculated), taken from the southern part of the PERLE sampling area (the NWLB) illustrate the seasonal variability of the carbonate chemistry in the upper layer ( Figure 3A). The pH T 25 was the most measured carbonate parameter in this study and, when normalised to 25 • C, can be considered as an indicator of the carbonate chemistry status by including the changes in A T and C T . An overview of the upper layer seasonal dynamics is also presented for temperature, salinity, fluorescence, and AOU profiles in Figures 3B-E, respectively.
The lowest pH T 25 values were encountered in March 2019 during the PERLE2 cruise and correspond to the relatively higher C T values and lower A T values. During this cruise, a significant range in the MLD was encountered with the deepest values observed. This cruise coincided with the abrupt stratification observed in the EMed after the deepening of the MLD from November to February-March (D'Ortenzio et al., 2005). Increased fluorescence values were observed in shallow waters at the end of the cruise (in the eastern part of the area) in comparison to the beginning of the cruise (in the western part).
Intermediate pH T 25 values were measured in June 2018 during the PERLE0 cruise corresponding to increased surface alkalinity and a moderate depletion in inorganic carbon. The PERLE0 cruise is an early summer cruise characterised by a shallow MLD. The highest fluorescence values were recorded during this cruise well below the MLD (ca. 90 dbars) and light oxygen supersaturation (AOU ≈ −20 µmol.kg −1 ) just beneath the MLD.
Finally, high pH T 25 values (>8.000) were measured up to 100 dbars during the PERLE1 cruise, probably in association with a high A T content due to evaporation. During this late summer cruise, the deepest Deep Chlorophyll Maximum (DCM) with the lowest fluorescence values but also the deepest negative AOU concentrations were encountered. Moreover, during this cruise, the mesoscale Ierapetra Eddy (IE) was crossed (see Supplementary Figure 3 and Ioannou et al., 2019). The core of this warm and salty eddy (Figures 3B,C) was characterised by a deepening of the MLD associated with a deep DCM and negative AOU values. Nonetheless, no clear IE signal was observed on the pH T 25 values (Figure 3A). In the EMed, spring and autumn seasons need to be considered as short transition periods between the summer and winter, which come later than on the continent (Özsoy et al., 1989). Moreover, in the EMed, summer is characterised by maximum heat in the surface layer that can remain up until November, whereas winter is identified with minimal heat that can occur until April. Considering each cruise as representative of a period within the annual cycle, the PERLE0 cruise

Total Alkalinity and Salinity Relationships Within the Mixed Layer
When no A T values were available (see section "Primary Quality Control of the Measured Data"), A T can be estimated based on an A T -S relationship. In the MedSea, several linear relationships between A T and salinity in the surface waters have been proposed for different sub-basins (e.g., Schneider et al., 2007;Cossarini et al., 2015;Hassoun et al., 2015a;Gonzaìlez-Daìvila et al., 2016).
During the PERLE cruises, in the NWLB, AT was significantly (n = 14, p-value = 0.014, r 2 = 0.36) influenced by salinity variations within the mixed layer (Figure 4). Figure 4 also displays the A T -S distribution in the Cretan Sea (grey dots on Figure 4). The mixing of high alkalinity Black Sea waters (values of ca. 2,967 µmol.kg −1 ; Hiscock and Millero, 2006) in the Cretan Sea shifts the A T -S characteristics of surface waters in agreement with Schneider et al. (2007) who demonstrated that freshwater and Black Sea inputs affect the A T -S relationship. More pronounced deviations from the expected linear A T -S relationship are observed for stations with deeper mixed layers (Figure 4). This might be the result of the mixing of water masses with different A T -S relationships during winter mixing.
As A T values were available only for PERLE0 and PERLE2 cruises, the A T -S relationship derived for the PERLE cruises in the mixed layer (and in the NWLB) have been based on a very limited number of data. The PERLE A T -S linear relationship was tested against the Hassoun A T -S linear model (Hassoun et al., 2015a). No significant differences were found on either the slope (t-test = 1.86, n = 14, p < 0.05) or the intercept (t-test = 0.27, n = 14, p < 0.05). Therefore, the annual timeseries were reconstructed based on the A T -S linear relationship measured by Hassoun et al. (2015a) in the surface waters (0-25 m) of the eastern Mediterranean sub-basin, and A T has been estimated based on this relationship.

ATYPICAL DRIVERS OF THE SEASONAL DYNAMICS OF THE CARBONATE CHEMISTRY WITHIN THE MIXED LAYER OF THE NORTH WESTERN LEVANTINE BASIN Seasonal Variations in Total Alkalinity and Total Inorganic Carbon
During the PERLE cruises, the NWLB exhibited a greater range in A T than C T values within the mixed layer (see section "Total alkalinity control on the seasonal air-sea CO 2 exchanges"). A T ranged between 2,610 and 2,693 µmol.kg −1 whereas C T ranged between 2,292 and 2,332 µmol.kg −1 . Over an annual scale, the ratio of the range in A T variations to the range in C T variations ( A T / C T ) can be used to infer the sensitivity to A T and C T changes in the upper ocean. Over the period studied, in the NWLB, the ratio A T / C T is equal to 2.1. In the global ocean, long-term time-series A T / C T ratios are lower than 1.0 ( Table 2).
The reasons for these apparent and rather unique ranges of A T and C T over the year in the NWLB can be attributed to several factors: (1) The main drivers of the C T gradient in the water column are, primary production transforming the C T into organic carbon in the photic layer, and respiration transforming the organic carbon into C T . As the EMed is an area of low productivity (Moutin and Raimbault, 2002), the vertical C T gradient is lower than in other oceanic areas. Consequently, the C T range in surface waters, driven by C T consumption during the stratified period and replenishment via vertical mixing with subsurface waters enriched in C T , is greatly reduced. (2) The high levels of evaporation that affect the MAW in the EMed during the summer season increases salinity by nearly 1 g.kg −1 (Figure 2) between the end of winter (PERLE2) and the end of summer (PERLE1). The A T and C T parameters should be equally affected by evaporation in a closed system. However, when reported on a A T /C T diagram (with normalised axes-see Figure 5), a higher range of A T variation compared to C T is observed. This indicates that when salinity increases in surface waters, a concomitant consumption of C T must occur to compensate for the C T increase due to evaporation to maintain an apparent stability in C T concentrations. The biological consumption of C T   1994-2018 1988-2019 1988-2018 1996-2004 2014-2019 A T / C T 2.1 0.6 0.9 0.2 0.9 0.8 0.5 References This study Lefèvre, 2010Coppola et al., 2020Bates et al., 1996Dore et al., 2009Santana-Casiano and González-Dávila, 2010Olafsson et al., 2009 will be discussed in the next section as a possible mechanism to explain this low C T variability.

Impact of Biological Processes on Variations in Seasonal Carbonate Parameters
To understand the overall impact of biological processes on the seasonal variations in the carbonate system in the NWLB, changes in A T and C T need to be considered independently from the changes induced by dilution and evaporation. For this purpose, salinity-normalised changes of A T and C T in the upper 200 dbars are plotted in Figure 5. To differentiate waters affected by air-sea exchanges from sub-surface waters, the upper 200 dbars of water column has been divided into two layers: within and below the mixed layer (0 dbars-MLD and MLD-200 dbars). The barycentre of all observational points, defined as the coordinate of the mean A T and C T values during each cruise, is reported and considered to be representative of the "season" sampled. The barycentres are spread along the photosynthesisrespiration line between the three cruises, reflecting the effects of biological processes on the carbonate system over the year. From the early summer period (PERLE0-red dots on Figure 5) to the end of the summer period (PERLE1-blue dots on Figure 5), for both layers, the barycentre shift was a signature for increased photosynthetic processes compared to respiration processes. The deepening of the DCM observed between the PERLE0 and PERLE1 cruises and the negative AOU values recorded during these cruises supported this observation. The deepening of the DCM is a signature to the downward displacement of primary producers related to surface nutrient depletion (Sigman and Hain, 2012), and negative AOU values reflect oxygen production. All these elements indicate that autotrophic processes dominate the upper water column between early and late summer. Based on these assumptions, between the end of the summer period (PERLE1) and the end of the winter period (PERLE2-green dots on Figure 5), the barycentre shift indicates that heterotrophic processes were dominant in the upper water column. Whilst observations cannot be time related, it can be assumed that between the late winter period of PERLE2 and the early summer period of PERLE0, the "theoretical" shift of the barycentre indicates a balance in favor of autotrophic processes during this period. When considered together, these seasonal changes in normalised A T and C T confirm that during periods of high evaporation, autotrophic processes are consuming C T and increasing A T . This can explain the apparent C T stability and the important change in A T over an annual cycle. Based on the assumption that, below the mixed layer, the PERLE sampling area is a closed system (unimpacted by airsea CO 2 fluxes), the temporal evolution in NA  , 2005). Therefore, it can be assumed that the water masses below the mixed layer remain isolated from surface CO 2 inputs between the PERLE0 and PERLE1 cruises. However, due to the late winter deepening of the MLD (Figure 3), between the end of the summer period (PERLE1) and the late winter period (PERLE2), NEC and NEP could be biased by air-sea exchanges.
The seasonal NEP values estimated in this study confirm previous estimations based on oxygen concentration changes monitored with short-time incubations during the stratified period. In June 2006, Regaudie-de-Gioux et al. (2009 reported a positive NEP value of 0.22 ± 1.30 mmol O 2. m −3 d −1 in waters above 100 meters in the EMed and in summer 2008, Christaki et al. (2011) reported positive NEP values of 4 ± 14 mmol O 2. m −2 d −1 . As previously observed by Schneider et al. (2007), the contribution of calcification and dissolution processes to variations in the carbonate system could be assumed to have a minor role in the MedSea. The NEC values calculated in the NWLB confirm this. The spreading of PERLE2 data points along the CaCO 3 formation/dissolution line in Figure 5 (green dots) might be associated to the spatial changes in alkalinity content across the geographical distribution of sampling sites during this cruise rather than to calcification and dissolution processes.

Total Alkalinity Control on the Seasonal Air-Sea CO 2 Exchanges
To address the question of the control of A T and C T changes on the "source" (pCO 2 SW > pCO 2 ATM ) or "sink" (pCO 2 SW < pCO 2 ATM ) of CO 2 in the NWLB, PERLE's A T and C T values are reported in Figure 6. The temperature range in the area has been used to draw the red and blue "iso pCO 2 SW -lines" as representative of the pCO 2 SW values encountered during the winter and summer PERLE cruises. Considering a mean atmospheric partial pressure (pCO 2 ATM ) value of 403 µatm (recorded at Lampedusa site from October 2018 to December 2019; Dlugokencky et al., 2021), the upper seawaters encountered at the warm end of summer with high alkalinity (PERLE1) were a "source" of CO 2 . In contrast, the cold and low alkalinity end of winter (PERLE2) surface waters were a "sink" of CO 2 with pCO 2 SW . Although the C T content remained almost stable between the PERLE cruises, the A T variability was noticeable with the lowest A T values measured at the end of the winter period (PERLE2) and the highest A T values estimated during PERLE1, at the end of the summer period. When considering the large pCO 2 SW variations due to the temperature variability represented by the shift between the red and blue isolines, the high alkalinity seawater at the end of summer (PERLE1-blue dots on Figure 6) induces low pCO 2 SW values when seawater starts to cool and therefore highlights the potential for surface waters to absorb atmospheric CO 2 . In the NWLB, the variability of the A T content of the surface waters over an annual cycle impacts the air-sea CO 2 exchanges. The "classical" vision that the pCO 2 SW variability is not driven by temperature change but by the biological control on C T , must be largely revisited in light of the important effect that variations in A T have on the pCO 2 SW regulation capability in the EMed.
In order to estimate the effect of the A T variability on the pCO 2 SW over an annual cycle, alkalinity was derived from salinity data from an Argo float that cycled in the NWLB for over a year. The temperature and total alkalinity (derived from salinity) values recorded by the float in the upper 20 dbars of the water column representative of the surface mixed layer affected by air-sea exchanges are presented in Figure 7. The cruise data within the mixed layer are also reported. In Figure 7, the red "iso pCO 2 SW -line" indicates the pCO 2 equilibrium between the ocean and the atmosphere. This isoline was derived at constant C T , based on the assumption that the pCO 2 SW is, apart from temperature, controlled by A T rather than by C T in the NWLB. The distribution of data above and below this line highlights the "source" or "sink" status of the NWLB for atmospheric CO 2 , respectively.
The float derived data agreed with data measured during the PERLE cruises and indicate a penetration of atmospheric CO 2 into the EMed from December to April, and a release of CO 2 into the atmosphere from May to November. It must be noted that these estimates are sensitive to the C T value used. Indeed, by considering a high C T content (grey isoline labelled "C T max" in Figure 7), the period of CO 2 "sink" for the atmosphere will be shorter (from February to April). Conversely, if the lowest C T mean value is considered (black isoline labelled "C T min" in Figure 7), the area will act as a "sink" from December to May. The observed "iso pCO 2 SWlines" shift (grey and black isolines in Figure 7) from the "iso pCO 2 SW -line" at mean C T (red isoline in Figure 7) due to the C T variability over a year induces a temporal change in the status of "source" or "sink" of the upper water masses.  Hassoun et al. (2015a)]. The colour bar corresponds to the "month of the year." The red "iso pCO 2 SW -line" corresponds to the mean pCO 2 ATM value at Lampedusa site (estimated from the mean mole fraction of CO 2 in ppm) calculated with the mean C T values for all PERLE cruises ( = 403 µatm). The two others grey isolines correspond to the same constant pCO 2 SW with the minimum and maximum C T values (from PERLE cruises) (2,292 and 2,332 µmol.kg −1 , respectively). Arrows reflect the theoretical changes in A T and temperature throughout the year. The coloured area represents the error associated to the red "iso pCO 2 SW -line" deduced by combining the uncertainty associated to the A T values (i.e., ± 19 µmol.kg −1 ) with the default standard uncertainties from the constants (Orr et al., 2018).
Moreover, by considering the accuracy of ± 19 µmol.kg −1 associated to the A T estimation (according to Hassoun et al., 2015a), the uncertainty of the estimated pCO 2 SW has been calculated (Orr et al., 2018) and ranged between the two "iso pCO 2 SW -lines" deduced from the maximum and minimum C T values (red area on Figure 7). Although the displacement of the air-sea pCO 2 equilibrium might shift considering the A T uncertainty, the temporal succession of the "sink" or "source" status for atmospheric CO 2 throughout a year in the NWLB is evidenced. It confirms that the A T content of the surface waters is a significant driver of the air-sea CO 2 fluxes in the NWLB.
These are, to the best of our knowledge, the first estimates of the succession of the "sink" and "source" status in the NWLB based on in situ data. Previous estimates based on satellite observations of sea surface properties, and on a model characterising the evolution of the mixed layer pCO 2 SW (D'Ortenzio et al., 2008;Taillandier et al., 2012) are confirmed by this study. Moreover, coastal observations in the South eastern Levantine basin close to the Israeli shelf have also reported a CO 2 source for the atmosphere in summer (from May to December) and a sink of atmospheric CO 2 in winter (from January to April) (Sisma-Ventura et al., 2017).

LONG TERM TEMPORAL CHANGES IN CARBONATE CHEMISTRY IN THE NORTH WESTERN LEVANTINE BASIN Decadal Carbonate Chemistry Trends in Surface Waters in the NWLB
Based on historical observations from the CARIMED dataset and observations from the PERLE cruises, temporal changes in carbonate chemistry between 2001 and 2019 in the surface NWLB have been assessed to study the mechanisms that could explain the carbonate system changes over the last twenty years (Figure 8). The surface layer has been defined to a depth of 50 dbars to include sufficient data. Due to the seasonal changes in surface salinity in the EMed (Grodsky et al., 2019), salinity-normalised A T (NA T 39.3 ) and C T (NC T 39.3 ) were used to facilitate the comparison between the different datasets across space and time. Indeed, due to the strong salinity dependency of alkalinity, by normalising by salinity, a significant part of the seasonal signal for alkalinity is removed.
While being higher (even when salinity-normalised) than the trends observed in the North Western MedSea (i.e., 1.40 ± 0.15 µmol.kg −1 .a −1 ; Merlivat et al., 2018), the temporal C T increase in the NWLB surface waters (Figure 8A) is consistent with other trends measured in the eastern Levantine basin (i.e., 5 ± 2 µmol.kg −1 .a −1 ; Hassoun et al., 2019). However, when compared to other time-series over the global ocean, the trends measured in the surface NWLB waters are 3.7-1.5 times higher (if the NC T 39.3 trend is considered) than the global ocean range which lies between 0.78 µmol.kg −1 .a −1 (Munida South Pacific timeseries) and 1.89 µmol.kg −1 .a −1 (CARIOCA time-series; Bates et al., 2014). This suggests that distinct mechanisms explaining the increasing C T trend exist in the NWLB.
While A T is considered insensitive to atmospheric CO 2 penetration (Zeebe, 2012), positive trends in C T and negative trends in pH T 25 (Figures 8A,E) can be explained, at least partially, by the increase in atmospheric CO 2 . Indeed, between 2006 and 2018, a mean annual increase of 2.2 ± 0.08 ppm.a −1 in xCO 2 ATM (mole fraction of CO 2 ) was recorded at the Lampedusa site (equivalent to the trend recorded on a global scale; Dlugokencky et al., 2021). To estimate the sensitivity of the estimated trends to the increase in atmospheric CO 2 , the increase in xCO 2 ATM was assumed to be equivalent to a surface ocean increase in pCO 2 SW . Based on the estimated trends in pCO 2 SW , NA T 39.3 , and NC T 39.3 , annual changes in carbonate chemistry pCO 2 SW have been calculated by solving thermodynamic equations ( Table 3). The observed annual decrease in pH T 25 ( Figure 8E) and increase in C T (Figure 8A) lies between the values estimated with and without an A T increase. This suggests that an A T increase must exist to compensate for the decrease in pH and the increase in C T or, in other words, that the high observed C T trend is the consequence of the observed A T increase. Although a positive A T trend has been observed elsewhere in a coastal site of the MedSea (Kapsenberg et al., 2017), it remains unexplained. These changes could be related to changes in riverine inputs or changes in Black Sea water inputs (Schneider et al., 2007).
It is worth noting that the CARIMED database, by merging data measured over the past 20 years, has a large overrepresentation of the spring season ( Supplementary Figure 1 and Supplementary Table 2). Moreover, the spatial distribution of the sampled stations was different for each cruise. The Estimated trends are obtained from slope values of a linear regression between the studied parameters and time. The confident interval has been added for each trend with the coefficient of determination (r 2 ), the number of values used (n) and the significance of the trend (p-value).  Figure 2C) or winter could modulate the observed temporal trends. Nonetheless, when data collected during "not spring" cruises are not considered to estimate the trends, despite shifting the temporal trend values, tendencies remain significant for each parameter. Thus, the conclusion that a decadal A T increase must exist to counterbalance the pH decrease associated to the C T increase remains coherent and valid.

Perspectives on the Future Functioning of the Eastern Mediterranean Carbonate System
In the projected warmer MedSea (Nykjaer, 2009), increased stratification but also reduced nutrient inputs from river discharge caused by more frequent drought periods could increase the oligotrophy of the MedSea (e.g., Moon et al., 2016;Pagès et al., 2019Pagès et al., , 2020. As this study suggests that the magnitude of the annual C T variation in surface waters is reduced in the EMed due to the low C T vertical gradients, all processes that could decrease primary production in the future could reduce the C T contribution to the air-sea exchanges. Even if internal thermohaline oscillation needs to be considered to draw solid conclusions about salinity trends, over the past 30 years, a positive long-term trend in salinity for the LSW and LIW has been recorded (Ozer et al., 2017). Because of the salinity impact on alkalinity concentrations (Figure 4) and of the A T impact on the air-sea CO 2 fluxes (Figure 7), if the PERLE1 conditions are exacerbated in the future with marine heatwaves extending over longer periods of the year, even more alkaline waters can be expected at the end of the summer. An even greater potential pCO 2 ATM sink will result when surface seawaters cool. The gyres (such as the IE), which have a higher A T content due to their saltier waters, might be even more efficient at catching atmospheric CO 2 when seawater cools. The control of air-sea CO 2 exchange by alkalinity that is suggested in this study could be enhanced in a future warmer and less productive EMed. However, as C T and A T are equally affected by evaporation and as, in the future less productive EMed, the C T biological consumption will be less efficient, the mechanisms leading to stable inorganic carbon content described in this study might be altered.
In an attempt to quantify the sensitivity of the carbonate system to future C T and A T changes, estimated buffer factors within the MLD for each PERLE cruise are presented in Table 4. At a comparable period of the year (March-April for PERLE2 cruise), the estimated buffer factors are in good agreement with former estimates (Álvarez et al., 2014) whereas the estimated buffer factors for PERLE0 and PERLE1 cruises during summer are significantly higher. Higher absolute buffer values imply higher buffering capacity and lower changes in [CO 2 ], pH or for a given change in A T or C T . Assuming that the PERLE1 conditions will be exacerbated in the future (Darmaraki et al., 2019), the EMed surface water is moving toward an overall increase in its buffering capacity (relative to changes in A T and C T ).
It is worth noting that, when atmospheric CO 2 dissolves in seawater, the CO 2 concentration in solution changes due to the carbonate ion buffering effect. The future effects of the decadal trends measured in the NWLB on the buffering capacities of the carbonate ion can be discussed using three different perspectives: (1) By considering the observed decrease in pH T 25 , the carbonate ion availability will decrease accordingly, reducing the atmospheric CO 2 uptake by the MedSea. (2) The greater increase in C T in comparison to the increase in A T will reduce the carbonate ion availability, but, nevertheless, will compensate for the impact of a pH decrease on the carbonate ion content, so allowing the CO 2 uptake into the atmosphere. (3) The positive trend in A T , and its impact on the CO 2 atmospheric uptake and on mitigating the decreasing pH trend, may indirectly increase the C T .

CONCLUSION
Based on data collected in the EMed over three different seasons of the year, this study provides for the first time, an annual overview of the seasonal dynamics of the carbonate chemistry in the NWLB. In this area, an atypical seasonal range in A T variations compared to the range in C T variations results from the combination of high rates of evaporation and biological processes. The high A T content at the "end of summer" period has a strong impact on the air-sea exchanges of CO 2 . In the NWLB, the status of "source" or "sink" for atmospheric CO 2 is adjusted by the A T variability more than the C T variability. Over longer time scales, and by compiling historical data, the reported increasing trends in A T and C T impact with divergent effects the observed acidification. These "end of summer" conditions will occur more frequently and lasting longer in the future. This ocean warming up will result in an increased buffer capacity that could mitigate the ocean acidification of the EMed.

DATA AVAILABILITY STATEMENT
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://mistrals.sedoo. fr/MERMeX/ and http://www.coriolis.eu.org.

AUTHOR CONTRIBUTIONS
CW-R, TW, and DL initiated and design the study. MÁ provided the CARIMED database and contributed to carbonate chemistry interpretation. PR helped supervising the study. MP-P and PC provided the nutrients database. MF, LC, TM, LN-C, CW-R, and TW performed on board carbonate parameters and oxygen analytical measurements. VT and FD'O provided CTD and ARGO dataset. FD'O, XD, and PC planned and designed the PERLE Research cruises. CW-R, TW, and DL wrote the first draft of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.

FUNDING
This study takes part of the PERLE (Pelagic Ecosystem Response to the Levantine Experiment) of the MISTRALS-MERMEX project. The project leading to this publication has received funding from European FEDER Fund under project 1166-39417. The SNAPO-CO2 service at LOCEAN was supported by the CNRS-INSU and OSU Ecce-Terra.