Since 2003, the discrete samples returned back to SNAPO-CO₂ service facilities (LOCEAN, Paris) have been analyzed simultaneously for AT and CT by potentiometric titration using a closed cell (Edmond, 1970; Goyet et al., 1991). The same technique was used at sea for surface water underway measurements during OISO and MINERVE cruises (indicated by an asterisk in Table 1). In the late 1980s the JGOFS–IOC Advisory Panel on Ocean CO₂ recommended the need for standard analysis protocols and for developing certified reference materials (CRMs) for inorganic carbon measurements (Poisson et al., 1990; UNESCO, 1990, 1991). The CRMs were provided to international laboratories by Andrew Dickson (Scripps Institution of Oceanography, San Diego, USA), starting in 1990 for CT and 1996 for AT, respectively. These CRMs were thus always available to us and used to calibrate the measurements (CRM batch numbers used for each cruise are listed in the Supplement, Table S2). The CRM accuracy, as indicated in the certificate for each batch, is around ±0.5 µmol kg⁻¹ for both AT and CT (https://www.nodc.noaa.gov/ocads/oceans/Dickson_CRM/batches.html, last access: 22 January 2025). The concentrations of CRMs we used vary between 2193 and 2426 µmol kg⁻¹ for AT and between 1968 and 2115 µmol kg⁻¹ for CT, corresponding to the range of concentrations observed in open-ocean water. In the Mediterranean Sea the concentrations are higher (AT > 2600 µmol kg⁻¹ and CT > 2300 µmol kg⁻¹), and in the coastal zones or near the Amazon River plume the concentrations were often lower than the CRMs (AT < 1500 µmol kg⁻¹ and CT < 1000 µmol kg⁻¹). Results of analyses performed on 1242 CRM bottles (different batches) in 2013–2024 are presented in Fig. 2. The standard deviations (SDs) of the differences in measurements were on average ±2.69 µmol kg⁻¹ for AT and ±2.88 µmol kg⁻¹ for CT. For unknown reasons, the differences were occasionally up to 10–15 µmol kg⁻¹ (1.2 % of the data; Fig. S2). These few CRM measurements were discarded for data processing. We did not detect any specific signal for CRM analyses (e.g., larger uncertainty depending on the batch number or temporal drifts during analyses; Fig. 2), but for some cruises the accuracy based on CRMs could be better than 3 µmol kg⁻¹ (e.g., < 3 µmol kg⁻¹ for AMAZOMIX cruise using six batches of no. 197 and for MOOSE-GE 2022 using 19 batches of no. 204, or < 1.5 µmol kg⁻¹ for SOMLIT-Point-B in 2022 using six batches of no. 204).

For some projects, duplicates have been regularly sampled (SOMLIT-Point-B, SOMLIT-Brest) or replicate bottles sampled at selected depths at fixed stations during the cruises (e.g., STEP, CARBODISS). In the first synthesis of the SNAPO-CO₂ dataset, we showed the results from several time series (SOMLIT-Point-B, SOMLIT-Brest and BOUSSOLE/DYFAMED). Here we present the results for the new data obtained at SOMLIT-Point-B in the coastal Mediterranean Sea and SOMLIT-Brest in the Bay of Brest (Fig. 3). Results of AT and CT repeatability are synthesized in Table 2. For the OISO cruises conducted in 2019, 2020, and 2021, the repeatability was evaluated from duplicate analyses (within 20 min) of continuous sea surface underway sampling at the same location (when the ship was stopped). Similarly to what was found for the CRM measurements (Fig. S2), differences in duplicates are occasionally higher than 10–15 µmol kg⁻¹ (Fig. 3), but most of the duplicates for all projects are within 0 to 3 µmol kg⁻¹. Compared to previous results (Kapsenberg et al., 2017; Metzl et al., 2024a), there are larger differences between duplicates at SOMLIT-B in 2019–2023 (up to 30 µmol kg⁻¹; Fig. 3), leading to relatively large SDs around 5 and 6 µmol kg⁻¹ for both AT and CT (Table 2). The same was observed for duplicates at SOMLIT-Brest (Table 2). We do not yet have a clear explanation for this large SD, although larger variability was observed in recent years, and the measurements were performed later after the sampling (e.g., more than 6 months for some samples during and after the COVID period). We will see that, given the temporal variability in the properties, this does not lead to suspicious interpretation for the seasonality or the trend analyses of these time series.

Identifying all data with an appropriate flag is very convenient for selecting the data (good, questionable, or bad). Here we used four flags for each property (flag 2 for good, 3 for questionable, 4 for bad, and 9 for no data) following the WOCE program and those used in other data products such as SOCAT (Bakker et al., 2016) and GLODAP (Olsen et al., 2016; Lauvset et al., 2024). During the data processing, we first assigned a flag for all AT and CT data based on the standard error in the calculation of AT and CT concentrations (nonlinear regression; Dickson et al. 2007). By default, if the standard deviation on the regression is > 1 µmol kg⁻¹, we assigned flag 3 (questionable), although the data could be acceptable and then used for interpretations. Flag 3 was also assigned when salinity was doubtful or when differences in duplicates were large (e.g., ±20 µmol kg⁻¹). Flag 4 (bad or certainly bad) was assigned when clear anomalies were detected for unknown reasons (e.g., a sample probably not fixed with HgCl₂ or analysis performed late during COVID). A secondary quality control was performed by the PIs of each project based on data inspection, duplicates, the AT–salinity relationship, or the mean observations in deep layers where large variability in AT and CT is unlikely to occur from year to year.

An example of quality flag is presented for all data from the MINERVE cruises conducted in 2002–2018 in the Southern Ocean, where clear outliers have been identified (Fig. S3). For the MINERVE cruises in 2002–2018 and a total of 12 335 AT and CT analyses, 24 were identified as bad (flag 4), 978 for AT and 971 for CT were listed as questionable (flag 3), and all others are considered good data (flag 2, i.e., about 92 %). For the MOOSE-GE cruises in 2021, 2022, and 2023 (new data in SNAPO-CO₂-v2) and a total of 1373 AT and CT analyses, 2 were identified as bad (flag 4), 38 for AT and 33 for CT were listed as questionable (flag 3), and all others were considered good data (flag 2, i.e., 97 %). This is better than the statistics we evaluated for the SNAPO-CO₂-v1 dataset (90 % flag 2 for MOOSE-GE in 2010–2019). A similar control was performed for each project.

Intercomparisons of measurements performed for different cruises or with different techniques help to evaluate the quality of the data and detect potential biases when merging the data in the same region obtained by different laboratories at different periods. This is especially important to interpret long-term trends in AT and CT as well as for pCO₂ and pH calculated with AT–CT pairs. The synthesis of various cruises in the same region and periods also offers verification and secondary control of the data.

Compared to the open ocean, AT concentrations are much higher in the Mediterranean Sea (Copin-Montégut, 1993; Schneider et al., 2007; Álvarez et al., 2023) with values up to 2600 µmol kg⁻¹. The AT and CT data obtained in 2014–2023 show a clear contrast between the northern and southern regions of the western Mediterranean Sea with higher concentrations in the Ligurian Sea and the Gulf of Lion (Fig. 7). This contrast is associated with the circulation and the frontal system in this region (e.g., Barral et al., 2021). New data in the coastal zones in the Gulf of Lion (ACCESS, DICASE, CARBODELTA, COCORICO₂, MESURHOBENT) also have very high AT and CT concentrations (AT > 2600 µmol kg⁻¹; CT > 2350 µmol kg⁻¹). Very low AT and CT concentrations (AT < 2500 µmol kg⁻¹; CT < 2200 µmol kg⁻¹) were also occasionally observed in the coastal zones (COCORICO₂ stations; Petton et al., 2024).

In summer 2022 the Mediterranean Sea experienced an exceptional warming (Fig. S6) superposed to the long-term warming in the ocean (Cheng et al., 2024). Such an event would impact the internal ocean processes such as thermodynamic, stratification, and biological processes (Coppola et al., 2023) and the interannual variability and trends in CT, pH, fCO₂, and air–sea CO₂ fluxes (Yao et al., 2016; Wimart-Rousseau et al., 2023; Chau et al., 2024b). As in 2003, the warming in summer 2022 was associated with the drought event that occurred in Europe and over the Mediterranean Sea (Faranda et al., 2023). In July 2022, the maximum temperature of 28.42 °C was observed at station SOMLIT-Point-B. In the Ligurian Sea the temperature trend has been faster in recent years: +0.173 ± 0.072 °C per decade over 1990–2010 and +0.678 ± 0.143 per decade over 2010–2023 (Fig. S6). With the new data added to the SNAPO-CO₂-v2 synthesis (DYFAMED, MOOSE-ANTARES, and MOOSE-GE), we evaluated a temperature trend of +0.84 ± 0.20 °C per decade over 1998–2022, indicating that the discrete sampling captured the property changes at regional scale. Based on the data in the Ligurian Sea, the trends in CT appeared faster in summer (+1.53 ± 0.46 µmol kg⁻¹ yr⁻¹) than in winter (+0.94 ± 0.64 µmol kg⁻¹ yr⁻¹; Table 6). On the other hand, the trends in AT were the same (+0.72 ± 0.36 µmol kg⁻¹ yr⁻¹ in winter and +0.69 ± 0.42 µmol kg⁻¹ yr⁻¹ in summer). The trend in CT on the surface in winter was close to the one derived at 100 m (below the Chl a maximum), CT100m=+1.10 ± 0.17 µmol kg⁻¹ yr⁻¹ (Fig. 8), whereas for AT the trend was the same at the surface and at depth (+0.76 ± 0.12 µmol kg⁻¹ yr⁻¹). This suggests that the winter CT data recorded the anthropogenic CO₂ uptake of around +1 µmol kg⁻¹ yr⁻¹; Fig. S7). Note that, given the observed CT trends, the spatial view presented in Fig. 7b for 2014–2023 would be the same based on CT concentrations normalized to a reference year. As noted by Touratier and Goyet (2009), the CT concentrations in the Mediterranean Sea should increase in parallel with the level of atmospheric anthropogenic CO₂. For an atmospheric CO₂ rate of +2.16 ppm yr⁻¹ over 1998–2023 (Lan et al., 2024) and at fixed sea surface temperature (17.75 °C), salinity (38.25), and AT (2567 µmol kg⁻¹), the theoretical CT increase would be +1.24 µmol kg⁻¹ yr⁻¹. Interestingly, an anthropogenic flux of -0.3 ± 0.02 molC m⁻² yr⁻¹ in the Mediterranean Sea (Bourgeois et al., 2016) would correspond to an increase in CT of 1.07 ± 0.07 µmol kg⁻¹ yr⁻¹ in the top 100 m. This is again close to what is observed in winter or at 100 m (Table 6, Fig. 8). On the other hand the faster CT trend observed in surface waters during summer might be associated with a decrease in biological production and/or changes in circulation/mixing over time that deserve specific investigations, such as an analysis of the oxygen budget in this region (Ulses et al., 2021). It is worth noting that the CT and AT trends in coastal zones of the Mediterranean Sea are the opposite of those observed offshore: for example at station SOLEMIO (Bay of Marseille; Wimart-Rousseau et al., 2020), the CT and AT concentrations decreased over 2016–2022 and thus opposed the anthropogenic CO₂ signal, indicating that processes such as riverine inputs, advection, and biology control the carbonate system decadal variability at local scale. This calls for developing dedicated complex biogeochemical models to resolve these processes (Barré et al., 2023, 2024), especially when extreme events occur, such as the very hot summer in 2024 with sea surface temperature (SST) up to 30 °C in the Mediterranean Sea (platform buoy and/or mooring AZUR, EOL, and La Revellata, data available at https://dataselection.coriolis.eu.org/, last access: 11 October 2024). The data obtained in the Mediterranean Sea are important not only to validate biogeochemical models but also to reconstruct the carbonate system from AT and pCO₂ data (Chau et al., 2024a) as the global AT–sea surface salinity (SSS) relationships (e.g., Carter et al., 2018) are not suitable for this region.

The North Atlantic Ocean is an important CO₂ sink (Takahashi et al., 2009) due to biological activity during summer and heat loss and deep convection during winter. As a result this region contains high concentrations of anthropogenic CO₂ (Cant) in the water column (Khatiwala et al., 2013). Decadal variations in the Cant inventories were recently identified at basin scale, probably linked to the change in the overturning circulation (Gruber et al., 2019; Müller et al., 2023; Pérez et al., 2024). This region experienced climate modes such as the North Atlantic Oscillation (NAO) and the Atlantic Multidecadal Variability (AMV) that imprint variability in air–sea CO₂ fluxes at interannual to multidecadal scales (e.g., Thomas et al., 2008; Jing et al., 2019; Landschützer et al., 2019) but not always clearly revealed at regional scale (Metzl et al., 2010; Schuster et al., 2013; Pérez et al., 2024). In addition it has been recently shown that extreme events such as the marine heat wave in summer 2023 led to a reduce CO₂ uptake in this region (Chau et al., 2024b). Although the annual CO₂ fluxes deduced from global ocean biogeochemical models (GOBMs) seem coherent with the data products at basin scale (-0.30 ± 0.07 and -0.24 ± 0.03 PgC yr⁻¹ for the North Atlantic subpolar seasonally stratified, NA-SPSS biome), the pCO₂ cycle seasonality is not well simulated (Pérez et al., 2024). Therefore to correct the GOBMs outputs, comparisons with the observed CT and AT cycles are also needed.

In this context regular sampling in the North Atlantic (OVIDE cruises, Mercier et al., 2015, 2024; SURATLANT transects, Reverdin et al., 2018) and time-series stations in the Irminger and Iceland seas (Olafsson et al., 2010; Lange et al., 2024; Yoder et al., 2024) are important to explore the variability in the biogeochemical properties from seasonal (Fig. S8) to decadal scales (Fig. 9). The SURATLANT data added to the SNAPO-CO₂-v2 dataset over 2017–2023 offer new observations in the North Atlantic Subpolar Gyre (NASPG in the NA-SPSS biome) and new transects from Norway to Iceland and reaching the coast of Greenland (Fig. 9). In 2010 the winter NAO was negative, moved to a positive state in 2012–2020, and was again very low in 2021. The new SURATLANT data after 2017 confirm the cooling and the freshening in the NASPG since 2009 (Holliday et al., 2020; Leseurre et al., 2020; Siddiqui et al., 2024), whereas the most recent data in 2022 and 2023 suggest a reverse trend (increase in salinity and temperature; Fig. S8). After 2016, large CT anomalies in the NASPG were observed. For example, in April 2019 and 2022, the CT concentrations were low compared to 2016 (Fig. 9) and opposed to the expected anthropogenic CO₂ uptake. In September 2023 the CT concentrations were much lower than in 2022 (Fig. 9), probably linked to biological productivity when the NAO index was negative (Fröb et al., 2019), as observed in summer 2023 (NAO < -2 in July 2023). Despite this variability the CT trends are relatively well evaluated (Table 6). As in the Mediterranean Sea the CT trends in the NASPG appeared to be different depending on the season (Fig. 9). The CT increase was faster in September than in April (+1.09 ± 0.37 µmol kg⁻¹ yr⁻¹ and +0.78 ± 0.23 µmol kg⁻¹ yr⁻¹). This is either close to or lower than the theoretical CT increase due to the rise of atmospheric CO₂ (+0.91 µmol kg⁻¹ yr⁻¹) and in the range of recent results evaluated for the subpolar mode waters in the Irminger Sea (Cant trend = 0.95 ± 0.17 µmol kg⁻¹ yr⁻¹ for the period 2009–2019; Curbelo-Hernández et al., 2024).

In the tropical Atlantic, previous studies highlighted the large variability of biogeochemistry and the difficulty in detecting long-term trends in CT (e.g., Lefèvre et al., 2021). This is related to the variability of circulation, equatorial upwelling, biological processes (some linked to Saharan dust), and inputs from large rivers (Congo, Amazon, and Orinoco). The new data added to version SNAPO-CO₂-v2 (Fig. S9) show the contrasting zonal CT distribution in this region with lower concentrations in low-salinity regions of the North Equatorial Counter Current and Guinea Current (Fig. 5; Oudot et al., 1995; Takahashi et al., 2014; Broullón et al., 2020; Bonou et al., 2022). For exploring the temporal changes, we selected the data in the western region, available for at least 10 years, and separated them into the northern and southern sectors. In both regions the CT trend is close to +3 µmol kg⁻¹ yr⁻¹ (Table 6, Fig. S9), much higher than the excepted anthropogenic signal. In this region where coastal water masses mix with oceanic waters, the interannual variability in CT is large and the changes driven by competitive processes (circulation, biological processes). More observations and dedicated models are needed to separate the anthropogenic and natural variability in this region (Pérez et al., 2024).

In the Southern Ocean there are a few regular multiannual observations of the carbonate system. Time series of more than 10 years were obtained in the Drake Passage (Munro et al., 2015) and in the southern Indian Ocean (Leseurre et al., 2022; Metzl et al., 2024b). Observations were also obtained for more than 20 years southeast of New Zealand from the Munida Time Series (MTS) in the subtropical and sub-Antarctic frontal zones (Currie et al., 2011; Vance et al., 2024). To complement these datasets we have added the data collected in the southeastern Indian Ocean between Tasmania and Antarctica in the framework of the MINERVE cruises (Fig. 10; Brandon et al., 2022). These cruises were conducted from October to March, offering each year a view of the seasonal changes between late winter and summer from the sub-Antarctic zone to the coastal zone near Antarctica (Adélie Land). In all sectors (here from 45 to 67° S), the CT concentrations were higher in October when the mixed-layer depth (MLD) was deep and were lower during the productive summer season (e.g., Laika et al., 2009; Shadwick et al., 2015). An example is presented at 60° S, 151° E from the data obtained along a reoccupied track in 2011–2012 (Fig. S10). At this location south of the polar front in the POOZ/HNLC area (permanent open ocean zone/high-nutrient low-chlorophyll area ), the CT concentrations were +25 µmol kg⁻¹ higher in October compared to February. The same seasonal amplitude was observed in the western Indian sector of the POOZ (Metzl et al., 2006, 2024b), suggesting that the CT seasonality is relatively homogeneous in this region, corresponding to the Indian SO–SPSS biome (Fay and McKinley, 2014). The difference in the climatological CT between October and January is on average +28.3 ± 9.8 µmol kg⁻¹ in the Indian Ocean POOZ (Takahashi et al., 2014). Given this seasonality and potential change in the seasonal amplitude over time (Gallego et al., 2018; Landschützer et al., 2018; Shadwick et al., 2023), the property trends have to be evaluated for October and January–February separately, here over 2002–2012 in the POOZ (Fig. 10, Table 6). In both seasons, the average CT concentrations reached a minimum in 2008 and increased faster in 2008–2012 (up to +4.8 µmol kg⁻¹ yr⁻¹). Interestingly, such an acceleration of the trend after 2009 was observed for pCO₂ at the MTS station (Vance et al., 2024). We note that the CT trend over 2002–2012 was slightly faster in October (Fig. 10), probably linked to deeper MLD, as suggested from the cooling and the salinity increase observed during this season (Fig. S10).

In the western Indian sector, the new data in the SNAPO-CO₂-v2 dataset from the OISO cruises at high latitudes also recorded a rapid CT trend over 5–8-year periods (e.g., +3.4 µmol kg⁻¹ yr⁻¹ in 2015–2020 at 56° S; Fig. 11, Table 6). Although the interannual variability in CT, between 10 and 20 µmol kg⁻¹, is often recognized (Fig. 11), the evaluation of the trends over more than 20 years indicated a faster trend in the subtropical Indian Ocean (+1.1 µmol kg⁻¹ yr⁻¹) compared to higher latitudes (Indian POOZ, +0.6 µmol kg⁻¹ yr⁻¹); they are close to the expected anthropogenic signal in these regions (+1.1 µmol kg⁻¹ yr⁻¹ in the subtropics and +0.8 µmol kg⁻¹ yr⁻¹ at higher latitudes).

Coastal waters experience enhanced ocean acidification due to increasing CO₂ uptake, due to accumulation of anthropogenic CO₂ (Bourgeois et al., 2016; Laruelle et al., 2018; Roobaert et al., 2024a; Li et al., 2024), and from local anthropogenic inputs through rivers or from air pollution (e.g., Sarma et al., 2015; Sridevi and Sarma, 2021; Wimart-Rousseau et al., 2020). The changes in the CO₂ uptake in coastal zones are also linked to biological processes (Mathis et al., 2024) or to circulation and local upwelling (Roobaert et al., 2024b), all controlling large variability in AT and CT in space and time, leading to uncertainties in detecting long-term changes in pCO₂ and air–sea CO₂ fluxes in heterogeneous coastal waters (Dai et al., 2022; Resplandy et al., 2024). At seasonal scale, large differences between observations and models were also identified, leading to differences in the coastal ocean CO₂ sink of up to 60 % (Resplandy et al., 2024). It is thus important to document the seasonal cycles of AT and CT to compare and correct models and thus to better predict future changes in biogeochemical properties in coastal waters and their impact on marine ecosystems. A better understanding of the processes and their retroaction in the coastal regions is also required regarding marine carbon dioxide removal (MCDR) experiments and for their evaluation (e.g., Ho et al., 2023).

In the SNAPO-CO₂-v2 dataset, new data have been added to the coastal zones at stations SOMLIT-Brest, SOMLIT-Roscoff, and SOMLIT-Point-B. They extend the period to 2022 or 2023 for temporal analysis. New data from the French coastal zones have been also included from the COCORICO₂ project documented in detail by Petton et al. (2024). The observations in coastal zones could be identified in the MARCATS regions (Margins and CATchment Segmentation; Laruelle et al., 2013) (Fig. 12), where little information is available for quantifying the ocean CO₂ sink at the decadal scale and for evaluation of the anthropogenic CO₂ uptake (Regnier et al., 2013; Dai et al., 2022; Li et al., 2024). To explore the change in the observed properties in the coastal zones and have an idea of the long-term CT trends, we selected the time series with at least 10 years of data (Table 7, Fig. 13). Except at high latitudes (Greenland and Antarctic coastal zones), we observed a warming in coastal zones (Fig. S11). Changes in salinity are also identified (increase or decrease), and results of the trends are presented for salinity-normalized CT at 34, 35, or 38 depending on the region. Although the interannual variability is large in coastal waters, sometimes linked to extreme events (e.g., river discharges), we observed an increase in N-CT at most of the eight selected locations. The exceptions are the coastal zones in the Gulf of Lion near the river Rhône and near Tasmania in October.

In the Gulf of Lion, the new data in the coastal zone confirmed the first view at the SOLEMIO station over 2016–2018 (Bay of Marseille; Wimart-Rousseau et al., 2020). In this region the lowest CT was observed in summer 2022 (average CT of 2238.6 ± 21.0 µmol kg⁻¹), much lower than in 2015 (2290.8 ± 44.7 µmol kg⁻¹). Over the continental shelf south of Tasmania (MARCATS no. 34), the trend in N-CT was positive in summer but not significant in October. In October this was associated with an increase in salinity and in AT probably linked to advective processes via the reversal and variability in the Zeehan or the East Australian currents. From our data a warming of +0.06 °C yr⁻¹ was identified for both seasons over 2002–2012, as previously observed south of Tasmania over 1991–2003, impacting the pCO₂ trend and air–sea CO₂ fluxes in this region (Borges et al., 2008). The difference in the N–CT trends in austral summer and spring calls for new detailed studies with extended data in this region. At high latitudes in Adélie Land (Antarctic coast MARCATS no. 45), the variability in N-CT was large (range from 2150 to 2200 µmol kg⁻¹; Fig. 13), and the trend over 10 years in summer was not significant (Table 7). As opposed to the open zone at 60° S (Fig. 10), the CT concentrations in the coastal zone near Antarctica did not increase, probably linked to competitive processes between anthropogenic uptake, changes in primary production, mixing, or ice melting (Shadwick et al., 2013, 2014). More data are needed to better evaluate the changes in the carbonate system in Antarctic coastal zones where bottom waters are formed and transport anthropogenic CO₂ at lower latitudes (Zhang et al., 2023).

For the coastal time series SOMLIT where annual trends could be estimated (sampling at monthly resolution), the N-CT increase (+2.1 to 3.4 µmol kg⁻¹ yr⁻¹) is close to or higher than the anthropogenic signal, leading to a decrease in pH ranging between -0.05 and -0.06 TS (total scale) per decade. The new data added to the SNAPO-CO₂-v2 dataset (2016–2023) confirm the progressive increase in CT and the acidification in the western Mediterranean Sea and in the northeastern Atlantic coastal zones (Kapsenberg et al., 2017; Gac et al., 2021).