Two new coastal time-series of seawater carbonate system variables in the NW Mediterranean Sea: rates and mechanisms controlling pH changes

In this work, we present, for the ﬁ rst time, the seawater carbonate system measurements of two coastal time-series in the NW Mediterranean Sea, L ’ Estartit Oceanographic Station (EOS; 42.05°N 3.2542°E) and the Blanes Bay Microbial Observatory (BBMO; 41.665°N 2.805°E). At these two time-series, measurements of total alkalinity (TA), pH, and associated variables, such as dissolved inorganic nutrients, temperature, and salinity, have been performed monthly since 2010 in surface seawater. Seasonality and seasonal amplitude are analogous in both time-series, with seasonality in pH T in situ (pH at in situ seawater conditions on the total hydrogen ion scale) primarily determined by seasonality in sea surface temperature. The evaluated pH T in situ trends at BBMO (-0.0021 ± 0.0003 yr -1 ) and EOS (-0.0028 ± 0.0005 yr -1 ) agree with those reported for coastal and open ocean surface waters in the Mediterranean Sea and open ocean surface waters of the global ocean, therefore indicating that these time-series are representative of global ocean acidi ﬁ cation signals despite being coastal. The decreases in pH T in situ can be attributed to increases in total dissolved inorganic carbon (DIC; 1.5 ± 0.4 µmol kg -1 yr -1 at BBMO and 1.6 ± 0.6 µmolESkg -1 yr -1 at EOS) and sea surface temperature (0.08 ± 0.02 °C yr -1 at BBMO and 0.08 ± 0.04 °C yr -1 at EOS). The increases in carbon dioxide fugacity ( f CO 2 ; 2.4 ± 0.3 µmol kg -1 yr -1 at BBMO and 2.9 ± 0.6 µmol kg -1 yr -1 at EOS) follow the atmospheric CO 2 forcing, thus indicating the observed DIC increase is related to anthropogenic CO 2 uptake. The increasing trends in TA (1.2 ± 0.3 µmol kg -1 yr -1 at BBMO and 1.0 ± 0.5 µmol kg -1 yr -1 at EOS) buffered the acidi ﬁ cation rates, counteracting 60% and 72% of the pH T in situ decrease caused by increasing DIC at EOS and BBMO, respectively. Once accounted for the neutralizing effect of TA increase, the rapid sea surface warming plays a larger role in the observed pH decreases (43% at EOS and 62% at BBMO) than the DIC increase (36% at EOS and 33% at BBMO).


Introduction
Human activities have exponentially increased the atmospheric concentration of carbon dioxide (CO 2 ) since the Industrial Revolution.As a consequence of the oceanic uptake of about 20-30% of those CO 2 emissions (Gruber et al., 2019;Friedlingstein et al., 2022), ocean surface pH has decreased by 0.1 to 0.2 units (Doney et al., 2009;Rhein et al., 2013;Bates et al., 2014;Rıós et al., 2015), a phenomenon known as ocean acidification.In addition to the pH decrease, the dissolution of anthropogenic CO 2 in seawater leads to an increase in the partial pressure of CO 2 (pCO 2 ) and total dissolved inorganic carbon (DIC), and a reduction in the carbonate ion levels (Doney et al., 2009), thus reducing the saturation states of calcium carbonate minerals, such as aragonite (W Ar ) and calcite.Such changes in ocean chemistry may have direct and indirect consequences for marine life (particularly calcifying organisms), ecosystems, and reliant human communities (Riebesell et al., 2000;Kroeker et al., 2013;Mostofa et al., 2016;Doney et al., 2020).
Although ocean acidification is a global problem, it does not occur uniformly worldwide (Gattuso et al., 2015), presenting a wide range of rates, especially in coastal zones (Hofmann et al., 2011;Duarte et al., 2013a), where natural spatiotemporal variability is an important source of uncertainty when detecting anthropogenic acidification (Duarte et al., 2013a;Carstensen and Duarte, 2019).While the phenomenon of ocean acidification has garnered considerable scientific attention (Doney et al., 2009;Gattuso et al., 2015), the processes and consequences associated with coastal acidification have been less explored.
The ocean carbon cycle community is increasingly recognizing coastal and nearshore areas as hotspots for carbon and biogeochemical variability (Cai et al., 1998;Cai et al., 2020;Cai et al., 2021;Dai et al., 2022), especially vulnerable to global change, likely sites for marine CO 2 removal interventions, and regions of great importance, as coastal ecosystems provide invaluable resources and services, including various for climate change mitigation (Duarte et al., 2013b).The increased CO 2 uptake of coastal waters is not only contributing to their acidification but also exacerbating existing global challenges such as eutrophication, pollution, and habitat degradation (Andersson and Gledhill, 2013;Cai et al., 2023).Therefore, a robust understanding of carbonate chemistry is key to properly assessing habitat vulnerability to ocean acidification.
The dynamics of the seawater carbonate system are intricately shaped by the interplay between physicochemical (such as ocean circulation and mixing, and heat, carbon, and freshwater exchanges with the atmosphere) and biological (such as photosynthesis and calcification) factors.The temporal and spatial dimensions of these interactions add further complexity, emphasizing the need for continuous time-series to disentangle these complex interactions, providing a detailed understanding of how the seawater carbonate system responds to evolving environmental conditions (e.g., Doney et al., 2009;Benway et al., 2019).Long-term observations allow for the identification of trends, patterns, and potential tipping points, enabling to distinguish between natural variability and anthropogenic influences.Continuous datasets are, therefore, essential for advancing our knowledge and facilitating informed management strategies in the context of ongoing global environmental changes.
The Mediterranean Sea has been recognized as one of the most prominent climate-change hotspots (Giorgi, 2006) and is considered a "miniature ocean" (Bethoux et al., 1999), serving as a model to anticipate the responses of the global ocean to diverse pressures (Lejeusne et al., 2010).The relatively rapid overturning circulation of the Mediterranean Sea and the high total alkalinity (TA) of its waters lead to a naturally high capacity of the Mediterranean Sea to absorb and buffer anthropogenic CO 2 (Schneider et al., 2010;Lee et al., 2011;A ́lvarez et al., 2014;Palmieŕi et al., 2015).Despite its importance, the seawater carbonate system of the Mediterranean Sea is still poorly quantified, especially in coastal zones (understood here as those within the continental shelf), and long-term time-series are still sparse (Hassoun et al., 2022).
In this context, we present here the seawater carbonate system measurements of two coastal time-series in the NW Mediterranean Sea: L'Estartit Oceanographic Station (EOS; 42.05°N 3.2542°E) and Blanes Bay Microbial Observatory (BBMO; 41.665°N 2.805°E) (Figure 1).At these two coastal time-series, surface sampling for TA, pH, and associated variables, such as dissolved inorganic nutrients (nitrate, NO 3 -, silicate, SiO 2 , and phosphate, PO 4 3-), temperature, and salinity, have been performed monthly since 2010.Using these newly released datasets, we evaluate the seasonality and assess the long-term pH changes in surface waters and explore the physical and chemical drivers causing them.
Variations in these drivers are the result of changes in ocean circulation and mixing, biological processes, as well as exchanges of heat, freshwater, and carbon with the atmosphere.
2 Materials and methods

EOS and BBMO: site description, sampling, and measurements
The studied coastal time-series are located off the coast of Girona (Catalonia, Spain), ~60 km apart, being EOS in deeper waters and further from the coast than BBMO (92 m deep and ~3.5 km off the main coast and ~2 km off the Medes Islands, and 20 m deep and ~800 m offshore, respectively; Figure 1).Both stations are influenced by the southwest-flowing Northern Current (Figure 1), which originates before the Ligurian Sea and continues south of the Ibiza Channel (Millot, 1999).Both time-series locations present strong stratification in summer and a deep mixed layer depth during the rest of the year (Guadayol et al., 2009;Aparicio et al., 2017;Zamanillo et al., 2021).The most offshore site, EOS, lays close to a natural reserve (Marine Protected Area of the Medes Islands), while BBMO is close to the Blanes harbor area so, a priori, we would expect land-ocean interactions and anthropogenic activities to lead to higher variability at BBMO than at EOS.
Weekly monitoring of sea surface temperature at EOS began in 1973, led by Josep Pascual, being the longest uninterrupted timeseries of oceanographic data in the Mediterranean Sea (Salat et al., 2019).Monthly sampling for pH and TA started in 2010, jointly with Josep Pascual's measurements of sea surface temperature and salinity.Plankton ecology around the BBMO site has been extensively studied since the mid-XX century (e.g., Margalef, 1945), but it was in 1998 when a more continuous sampling effort began, with a focus on microbial biodiversity and biogeochemical function.Since then, monthly water sampling has been carried out for a very broad range of biotic and abiotic variables (Gasol et al., 2012) for which pH and TA also began to be measured at the end of 2009.For this work, we focus our study period at EOS from 22/01/ 2010 to 23/08/2019, since measurements had to be interrupted at that date, and from 22/12/2009 to 02/08/2022 at BBMO.BBMO temperature and salinity were measured with a SAIV-A/ S-SD204 CTD (Environmental Sensors & Systems, Norway), with an accuracy of ± 0.02 in salinity and ± 0.01 °C in temperature.EOS temperature was measured with reversible thermometers with an accuracy of ± 0.02 °C, while EOS salinity was measured with two CTDs, a SAIV-A/S-SD204 CTD (Environmental Sensors & Systems, Norway) with an accuracy of ± 0.02 in salinity, and a CTD75M (Sea & Sun Technology, Germany) with an accuracy of ± 0.01 in salinity.For consistency between BBMO and EOS datasets, the salinity record obtained with the SAIV-A/S-SD204 CTD is considered in this study.When salinity values from the SAIV-A/S-SD204 CTD were missing, salinity values from CTD75M were used.
The salinity records obtained with the two CTDs were quality controlled and calibrated to obtain a consistent dataset.
Discrete seawater samples for pH, TA, and dissolved inorganic nutrients were taken monthly from a depth of 0.5 m at both stations (with a Niskin bottle at EOS and in 10 L polyethylene carboys at BBMO).For pH measurements, at EOS, two cylindrical optical glass cells with a 10 cm path-length were filled directly from the Niskin bottle right after reaching land, ready to be analyzed in the next 3-4 hrs.At BBMO, a 150 mL glass bottle was filled leaving no headspace, and three 10 cm path-length cylindrical optical glass cells were filled from it once at the ICM laboratory (< 2 hrs after sampling).For TA measurements, one sample per site was taken in 500 mL borosilicate glass bottles, rinsed three times, and carefully filled from the bottom with a tube.Samples for TA were poisoned with 300 μL mercuric chloride (HgCl 2 ) saturated solution to halt biological activity (Dickson et al., 2007), right after reaching land for EOS and at the ICM laboratory for BBMO.Bottles were then stored in the dark at room temperature until they were analyzed in the laboratory, within days up to one month.Each TA sample was measured by duplicate.Samples for dissolved inorganic nutrients were stored frozen at -20°C in sterile falcon tubes until analysis.These tubes were filled directly from the Niskin bottle upon reaching land in EOS and from a bottle at the ICM laboratory in the case of BBMO samples.TA was analyzed by potentiometric titration, determined by double endpoint titration (Peŕez and Fraga, 1987;Mintrop et al., 2000), and calibrated with reference materials (Prof.A. Dickson, Scripps Institution of Oceanography (USA); Batches #93, #104, #136, and #179).TA measurement precision was ± 4 μmol kg -1 .
Seawater carbonate system parameters, W Ar , DIC, CO 2 fugacity at in situ seawater conditions (fCO 2 ), and pH at in situ seawater conditions on the total hydrogen ion scale (pH T in situ ) were calculated from pH T25 and TA measurements using the MATLAB ® version of CO2SYSv3 (Sharp et al., 2021), with the carbonic acid dissociation constants of Mehrbach et al. (1973) reformulated on the total hydrogen scale by Lueker et al. (2000), the bisulfate dissociation constant of Dickson (1990), and the total boron to salinity ratio of Lee et al. (2010).We used the CO2SYSv3 uncertainty propagation code (Sharp et al., 2021;adapted from Orr et al., 2018) to estimate the uncertainties in calculated parameters, taking into account the uncertainties in thermodynamic constants suggested by Orr et al. (2018), resulting in an uncertainty of ± 4 μmol kg -1 in DIC, ± 0.02 in W Ar , and ± 10 μatm in fCO 2 .When dissolved inorganic nutrient measurements were unavailable (12% of the EOS dataset and 6% of the BBMO dataset), those values were assumed to be equal to the monthly average values for each time-series.

Trend assessment
To quantify interannual changes, all datasets were detrended for seasonality using the recently developed Trends of Ocean Acidification Time Series (TOATS, https://github.com/NOAA-PMEL/TOATS)software, which is a supplement to the recently published best practices for assessing trends of ocean acidification time-series with monthly or higher periodicity sampling (Sutton et al., 2022).Briefly, the procedure first removes the overall linear trend by applying a simple linear regression and removing the slope from the original dataset.Then, climatological monthly means are calculated using this detrended time-series, and the climatological annual means are determined.Monthly adjustments are then determined by subtracting the climatological annual mean from the climatological monthly means.Finally, the de-seasoned dataset is obtained by subtracting the adjustment value for each month from the time-series of monthly means, which was not detrended [for more details, see Sutton et al. (2022)].The TOATS software also provides the trend detection time, i.e., the minimum observational period needed to statistically distinguish between natural variability (noise) and anthropogenic forcing.
Long-term trends were computed with the de-seasoned dataset, using ordinary least squares regression, and 95% confidence intervals were calculated for the slopes of the regressions.The reported trends were calculated excluding data points identified as outliers in pH after the dataset was de-seasoned (data from 16/05/2017, 22/06/2017, and 24/07/2019 for EOS, and 13/12/2016 and 13/10/2021 for BBMO, data outside 2SD -standard deviation-boundaries).
We tested if the trends were different using the TOATS deseasonalizing software (Sutton et al., 2022) and the widely-applied approach of Bates et al. (2014), where de-seasoned data is computed by subtracting the respective climatological monthly means computed from the time-series and calculating the trends with those anomalies.The two approaches produced statistically indistinguishable trends (Table 1, Supplementary Table 1).
For comparison with atmospheric CO 2 values, we used atmospheric CO 2 data from Plateau Rosa, Italy (courtesy of the World Data Center for Greenhouse Gases; https://gaw.kishou.go.jp/).

Carbonate system driver determination
Observed temporal changes in pH T in situ were decomposed into those associated with each of the potential drivers, assuming linearity and using a first-order Taylor-series deconvolution approach (Equation 1) (Garcıá-Ibañez et al., 2016;Kwiatkowski and Orr, 2018): where represents the slope contribution of changing "Driver" to the observed temporal change in pH T in situ (DpH).The sensitivity of pH to each driver ( ∂ pH ∂ Driver ; Table 2) was estimated by calculating pH T in situ using the true observations of each driver and holding the other three drivers constant (mean value of the time-series) and regressing it to each driver.Sensitivity was then multiplied by the corresponding observed temporal changes in in situ temperature (DTemp), salinity (DS), TA (DTA), and DIC (DDIC) (Table 1).We did not use salinity-normalized TA and DIC as drivers because there was no clear relationship between TA and salinity or between DIC and salinity (not shown).This was also the approach used in another coastal time-series in the NW Mediterranean Sea (Point B; Kapsenberg et al., 2017), where salinity was found to be a poor TA predictor.
The driver decomposition results using the de-seasonalizing TOATS software (Table 2) were statistically indistinguishable from those resulting from applying the de-seasonalizing technique of Bates et al. (2014) (Supplementary Table 2).

Seasonality in biogeochemical variables
The two time-series present a relatively similar seasonality in all the studied variables (Figures 2, 3).Surface waters are warm in summer and temperate in winter (Figure 2A), with maximum temperatures above 23 °C between June and September and minimum temperatures around 13 °C from December to March.
Surface salinity and TA present no clear seasonal cycle (Figures 2B, C), with values quite constant all year round, with annual average salinity values of 37.9 ± 0.3 and annual average TA values of 2557 ± 14 μmol kg -1 .Salinity values have higher variability in March and December (Figure 3B).
Surface DIC presents maximum values of around 2280 μmol kg -1 in March in both time-series (Figure 2D).From that maximum, DIC decreases in spring-summer to minimum values of 2213 ± 22 μmol kg -1 (monthly mean ± standard deviation) reached in September at BBMO and of 2207 ± 21 μmol kg -1 reached in November at EOS.
Seasonal changes in surface pH T in situ , W Ar , and fCO 2 are mainly determined by temperature seasonality because the seasonality in the other variables controlling them (i.e., salinity, DIC, and TA) is relatively small compared to that in temperature (Figure 2).The temperature control on pH T in situ , W Ar , and fCO 2 TABLE 1 De-seasoned time-series {using the Trends of Ocean Acidification Time Series [TOATS, https://github.com/NOAA-PMEL/TOATS;Sutton et al. (2022)] software} regression analyses at EOS and BBMO for pH at in situ seawater conditions on the total hydrogen ion scale (pH T in situ ), temperature (T), salinity (S), total alkalinity (TA), total dissolved inorganic carbon (DIC), and fugacity of carbon dioxide at in situ seawater conditions (fCO 2 in situ ).DIC (μmol kg -1 ) -0.0016 ± < 0.0001 -0.0025 ± 0.0010 88

Site (time
Sensitivity of pH with respect to each driver ( ∂pH ∂Driver ) was multiplied by the de-seasoned regression analyses of each driver ( dDriver dt ; Table 1), where the drivers are changes in sea surface temperature (T), salinity (S), total alkalinity (TA), and total dissolved inorganic carbon (DIC).The addition of the pH T in situ changes from each driver is also given (D pH sum ), as well as the percentage of the contribution of each driver to the observed pH T in situ trends (Table 1).SE is standard error and RMSE is root mean square error.seasonality leads to relatively low pH T in situ and high W Ar and fCO 2 in summer, and relatively high pH T in situ and low W Ar and fCO 2 in winter (Figures 2E-G).The two time-series sites behave as CO 2 sinks during autumn, winter, and spring, and as CO 2 sources during summer, when fCO 2 surpasses atmospheric CO 2 levels (Figure 2G).
In terms of dissolved inorganic nutrients, we show nitrate as the most representative dissolved inorganic nutrient (Figures 2H, 3H).Seasonality in nitrate content in surface waters is higher at EOS than BBMO but, at both sites, there is a maximum around February, with values decreasing into summer, where minimum concentrations are reached, and increasing during autumn.

Trends in the biogeochemical variables
We observed significant increases in sea surface temperature at both sites, with similar trends at BBMO (0.08 ± 0.02 °C yr -1 ) and at EOS (0.08 ± 0.04 °C yr -1 ) (Table 1).Sea surface salinity also increases significantly over time, with a three-times faster increase observed at EOS (0.040 ± 0.012 yr -1 ) than at BBMO (0.013 ± 0.006 yr -1 ) (Table 1).The TOATS-derived temperature trend detection time is 15.5 ± 2.5 years for BBMO and 9.3 ± 1.5 years for EOS, being EOS long enough (9.6 years) to detect statistically-significant trends in sea surface temperature, while BBMO may not be long enough (12.7 years) to detect the reported trend.The TOATS-derived salinity trend detection time is 19.3 ± 2.8 years for BBMO and 9.9 ± 2.1 years for EOS, indicating, therefore, that both time-series may not be long enough to detect those statistically significant trends.Although the trend detection times are sometimes longer than the length of our time-series, the sea surface temperature and salinity trends found in this study are statistically significant for both time-series (Table 1).
In terms of seawater carbonate system parameters, we observe that sea surface pH T in situ decreased significantly and at similar rates at BBMO, -0.0021 ± 0.0003 yr -1 , and at EOS, -0.0028 ± 0.0005 yr -1 (Figure 4; Table 1).The TOATS-derived pH T in situ trend detection time is 9.7 ± 1.5 years for BBMO and 7.6 ± 1.8 years for EOS, being, therefore, both time-series long enough to detect those statistically significant trends.
Significant increasing trends in sea surface TA, DIC, and fCO 2 are observed with similar rates in both sites (Table 1).The TOATSderived trend detection times for BBMO are 14 ± 3 years for DIC and TA and 9.4 ± 1.5 for fCO 2 , while for EOS they are 12 ± 3 years for DIC, 13 ± 3 years for TA, and 7.5 ± 1.7 for fCO 2 .Therefore, both time-series are long enough to detect statistically-significant trends in fCO 2 but may not be long enough for detecting statisticallysignificant trends in DIC and TA.Despite the trend detection times, the reported trends in DIC and TA for both sites are statistically significant (Table 1).Sea surface W Ar decreased at both sites, with the observed trend at EOS (-0.0051 ± 0.0041 yr -1 ) being more than three-times greater than that at BBMO (-0.0014 ± 0.0026 yr -1 ), although the trends are non-significant (Table 1).This is corroborated by the relatively long detection times for W Ar , being 44.7 ± 6.6 years for BBMO and 16.6 ± 2.7 years for EOS.
In terms of dissolved inorganic nutrients, non-significant trends were detected for the studied time period, mainly because of the small long-term changes (not shown) compared to the natural variability (long-term changes were three orders of magnitude smaller than the seasonal amplitude), needing between 10-50 years of data to detect statistically-significant trends.Garcı ´a-Iba ´ñez et al. 10.3389/fmars.2024.1348133Frontiers in Marine Science frontiersin.org

Drivers of ocean acidification trends
To investigate the drivers of the observed long-term changes in sea surface pH T in situ at BBMO and EOS, we decomposed them into their principal underlying drivers: changes in temperature, salinity, DIC, and TA (Figure 5; Table 2).Variations in these drivers are the result of changes in ocean circulation and mixing, biological processes, as well as exchanges of heat, freshwater, and carbon with the atmosphere.
The estimated trends from the decomposition (DpH Sum T ) agree with the observed pH T in situ trends (Figure 5; Table 2), thus indicating that the decomposition analyses accurately represent the observed trends.The predominant driver of the observed pH T in situ decreases was the increase in DIC, followed by sea surface warming.The observed fCO 2 trends (2.4 ± 0.3 μmol kg -1 yr -1 at BBMO and 2.9 ± 0.6 μmol kg -1 yr -1 at EOS) agree with those exhibited by atmospheric CO 2 (data from Plateau Rosa, Italy, for 2010-2018), which increased at 2.36 ± 0.03 ppmv yr -1 (r 2 = 0.98; pvalue < 0.01), therefore suggesting that the main driver of the changes in the inorganic carbon content at BBMO and EOS is the uptake of atmospheric CO 2 .Increases in TA played a major role in counteracting the pH decline.Assuming that the increase in TA was due to increases in carbonate alkalinity (bicarbonate and carbonate ions), then increases in bicarbonate and carbonate ions would contribute to both increases in TA and DIC, and we can sum their contributions to changes in pH T in situ .TA changes then counteracted 60% and 72% of the pH T in situ decrease linked to increasing DIC at EOS and BBMO, respectively.Once accounted for the neutralizing effect of the increase in TA, the observed rapid increase in sea surface temperature plays a larger role in the observed pH T in situ decreases (43% at EOS and 62% at BBMO) than the DIC increase related to anthropogenic CO 2 (36% at EOS and 33% at BBMO, when the TA effect is removed).

Discussion
The two presented coastal time-series in the NW Mediterranean Sea, with continuous monthly data, allowed us to unravel the intricate feedbacks inherent to the carbon cycle, despite the considerable fluctuations often exhibited in coastal regions.This study, therefore, highlights the importance of sustaining continuous time-series observations to help us distinguish natural from human-induced changes, such as rising ocean temperatures or deoxygenation (e.g., Benway et al., 2019).
The two studied time-series exhibit similar seasonal dynamics and ocean acidification trends and drivers.Summer stratification (May to October; Guadayol et al., 2009;Aparicio et al., 2017;Zamanillo et al., 2021) leads to decreases in dissolved inorganic nutrients (Figure 2H) and DIC (Figure 2D), the latter exacerbated by CO 2 outgassing associated with summer warming.Notably, BBMO experiences a more pronounced decrease in DIC and dissolved inorganic nutrients compared to EOS, likely attributable to wind patterns.The occurrence of "Garbins" (south-westerlies) triggers episodic upwelling events that transport deep waters from outside the platform (enriched in DIC and dissolved inorganic nutrients) to the time-series locations (Aparicio et al., 2017).Guadayol and Peters (2006) found that Garbins are more frequent and intense at Roses (nearby EOS) than at Malgrat (nearby BBMO), propitiating a greater impact of upwelling events replenishing the DIC and dissolved inorganic nutrient pools at EOS compared to BBMO.This would explain the more pronounced decrease in DIC and dissolved inorganic nutrients during summer at BBMO compared to EOS.The decrease in DIC observed during summer months would lead to an increase in pH T in situ and a decrease in fCO 2 , but the increase in temperature during the stratification season leads to a net decrease in pH T in situ (Figure 2E) and a net increase in fCO 2 (Figure 2G).Once the Observations of sea surface (A) temperature (T; °C), (B) salinity, (C) total alkalinity (TA; µmol kg -1 ), (D) total dissolved inorganic carbon (DIC; µmol kg -1 ), (E) pH at in situ seawater conditions on the total hydrogen ion scale (pH T in situ ), (F) aragonite saturation state (W Ar ), (G) carbon dioxide fugacity at in situ seawater conditions (fCO 2 ; µatm), and (H) nitrate (NO 3 -; µmol kg -1 ) from the two coastal time-series, BBMO (blue squares) and EOS (magenta asterisks).
summer stratification is over, surface waters are replenished in DIC (Figure 2D) and dissolved inorganic nutrients (Figure 2H) from deeper layers.Both time-series present quite homogeneous values of salinity and TA all year round (Figures 2B, C, 3B, C), with higher variability in March and December, where extreme precipitation events can sporadically lower salinity to values below 36.5 (Figure 3B).These extremes in salinity are not accompanied by extremes in TA (Figures 3B, C), most likely because the freshwater endmember in this region may have a relatively high TA content due to the limestone draining in part of the courses of the rivers and groundwaters in the area (Hartmann et al., 2015).This precludes the detection of relationships between TA and salinity, at least with the frequency of sampling of our dataset.
The decreasing trends in sea surface pH T in situ found at BBMO and EOS (Table 1) agree with other published time-series in the Mediterranean Sea.Specifically, the observed pH T in situ trends at EOS are very similar to those reported for the coastal time-series Point B (-0.0028 ± 0.0003 yr -1 for 2007-2015; Kapsenberg et al., 2017) 2).Trends reported in (x 10 -3 ) yr -1 and error bars represent the standard error of the estimate.

B A FIGURE 4
Sea surface pH T in situ observations (grey circles), de-seasoned monthly means of pH T in situ (black squares; using the using the Trends of Ocean Acidification Time Series (TOATS, https://github.com/NOAA-PMEL/TOATS;Sutton et al. ( 2022)) software), and pH T in situ trends from the deseasoned dataset (values in Table 1) from the two coastal time-series, (A) BBMO and (B) EOS.
2010) and the Strait of Gibraltar (-0.0030 ± 0.0003 yr -1 for 2005-2021; Amaya-Vıás et al., 2023).These pH trends reported at a constant temperature would most likely be larger if the trends were assessed at in situ conditions, due to the warming trends observed in the Mediterranean Sea (e.g., Salat et al., 2019) that contribute to ocean acidification (as shown in Kapsenberg et al. (2017)).The sea surface pH T in situ trends reported here also agree with other sea surface pH trends reported in open ocean waters elsewhere (Bates et al., 2014;Lauvset et al., 2015;Gonzaĺez-Davila and Santana-Casiano, 2023), thus indicating that EOS and BBMO time-series may reflect regional as well as global ocean acidification signals despite being coastal sites.
The rapid warming of the Mediterranean Sea contributes to the acidification of its waters, increasing the ocean acidification signal derived from the CO 2 uptake.This is corroborated by our results (Figure 5; Table 2), which go in line with those reported for the coastal time-series Point B (Kapsenberg et al., 2017), being changes in DIC and temperature the main drivers of the observed sea surface pH changes at these three coastal time-series in the NW Mediterranean Sea (EOS, BBMO, and Point B).However, the trends in each of the drivers do not completely coincide.While the sea surface temperature trends observed at EOS and BBMO (Figure 5; Table 1) agree with those observed at Point B (0.072 ± 0.022 °C yr -1 ), the trends in DIC and TA at Point B are twice as high as those observed at our coastal time-series sites.Additionally, there are contrasting trends in sea surface salinity, where Point B shows a non-statistically significant decrease while BBMO and EOS exhibit a statistically-significant increase.Regarding TA, Kapsenberg et al. (2017) remarked that their changes were faster than those attributable to direct effects of seawater CO 2 uptake, and they suggested that this could be related to increasing limestone weathering.Within our dataset, some TA rises could potentially link to increases in salinity (as shown in Table 1).This suggests that the observed TA increases at EOS and BBMO might relate to increased evaporation and/or reduced river and precipitation.
The ability to remove seasonal patterns from the datasets depends on how comprehensively the dataset covers the entire seasonal cycle.Essentially, the duration of the time-series needed to identify a human-induced trend depends on the level of natural variability present within the signal.In the case of pH in coastal systems, this requires nearly a decade or longer of data before a trend emerges from the noise (Sutton et al., 2019;Turk et al., 2019).To investigate how representative our observed trends in sea surface pH are at EOS and BBMO, we compared trends with those derived from the OceanSODA-ETHZ product (Gregor and Gruber, 2021), which is an observation-based, global gridded product with monthly surface data for all parameters of the seawater carbonate system at a resolution of 1°× 1°derived from the pCO 2 observations from the Surface Ocean CO 2 ATlas (SOCAT; Bakker et al., 2016) and the TA observations from the Global Ocean Data Analysis Product (GLODAP; Olsen et al., 2016).This product produced rates of pH change in agreement with those reported in time-series around the globe (Ma et al., 2023).The version used here was v2023, which expands from 1982 to 2022.The grids selected for our study were those centered at 42.50°N 3.50°E (for EOS) and 41.50°N 2.50°E (for BBMO).The OceanSODA-ETHZ product reproduces the observed seasonal cycle at our sites, but does not reproduce the seasonal amplitude of our observations (Supplementary Figure 1).The pH T in situ trends from the de-seasoned OceanSODA-ETHZ product are similar for both sites (-0.0020 ± 0.0001 yr -1 for BBMO and -0.0021 ± 0.0001 yr -1 for EOS; Supplementary Figures 2A, B black symbols).These pH T in situ trends do not change if we shorten the time period to match the length of our time-series (Supplementary Figures 2A, B red symbols).The pH T in situ trend from the de-seasoned OceanSODA-ETHZ product agrees with that obtained with the BBMO dataset, while for EOS it is slightly lower.Ma et al. (2023) previously noted that pH trends from OceanSODA-ETHZ cannot reproduce those obtained in short (< 30 years) time-series.However, it was already reported that the stronger ocean acidification trends observed toward the coast are not captured by the synthesis and modeling products (McGovern et al., 2022).
To test the capacity of our EOS dataset to resolve long-term trends, we compared our sea surface temperature trends for 2010-2019 at EOS with those resulting from the EOS sea surface temperature dataset for 1974-2018 (Pascual and Salat, 2019).For 1974-2018, sea surface temperature at EOS increased at a rate of 0.032°C yr -1 (Supplementary Figure 2C black symbols), which is the same trend that Salat et al. (2019) reported after de-seasoning the dataset using the Bates et al. (2014) approach.When we use the information from the entire dataset (1974-2018) to de-season the dataset and then compute the sea surface temperature trend for 2010-2018, there is no difference from the trend for 1974-2018 (details not shown).However, if we reduce the period to de-season the dataset to 2010-2018, the trend changes to 0.13 ± 0.03°C yr -1 (Supplementary Figure 2C red symbols), which agrees with the trend in sea surface temperature found in this study.This therefore indicates that our studied time period for EOS may be too short to produce robust long-term trends.

Conclusions
We characterized the seasonality and long-term changes of the seawater carbonate system in the surface waters of two coastal timeseries in the NW Mediterranean Sea, EOS and BBMO.Despite EOS being more offshore and near a natural reserve and BBMO being close to a harbor area, the two sites present similar seasonal dynamics and ocean acidification trends and drivers, suggesting that these changes are mainly determined by changes in temperature and the interplay between seasonal stratification/ mixing that are common to both sites, and not by more local processes that could have differing effects at the two locations.
The observed decreases in sea surface pH T in situ are caused by increases in DIC (related to anthropogenic CO 2 uptake) and sea surface temperature.Once accounted for the neutralizing effect of the observed increase in TA (counteracting 60% and 72% of the influence of increasing DIC at EOS and BBMO, respectively), the rapid sea surface warming plays a larger role in the observed pH T in situ decreases (43% at EOS and 62% at BBMO) than the DIC increase (36% at EOS and 33% at BBMO, when the TA effect is removed).
Future climate-related changes, such as rising temperatures or deoxygenation (Gruber, 2011), are expected to complicate the detection of future changes in seawater carbonate chemistry attributable to the atmospheric CO 2 increase.Maintaining timeseries of ocean carbon data is therefore crucial for assessing seasonal dynamics, annual budgets, and interannual and climatic variability (Tanhua et al., 2013;Vance et al., 2022), as they provide essential information to disentangle the complex interplay of climate-related physical, chemical, and biological feedbacks within the carbon cycle.Emergent climate trends can generally be identified earlier with more frequent measurements, particularly for systems that are highly variable on shorter timescales, like coastal regions.Despite the considerable spatiotemporal fluctuations in coastal regions, our dataset derived from the two presented coastal time-series, covering monthly data collected over 9.6-12.7 years, already reveals robust trends in ocean acidification comparable to those reported in open ocean time-series, offering valuable insights that allow unravelling the complex interplay of climate-induced physical and chemical feedbacks inherent to the carbon cycle.Although our dataset may not provide a complete overview due to the influence of short-term variability on long-term signals, our study nevertheless emphasizes the importance of long-term time-series for evaluating anthropogenic changes.

FIGURE 1
FIGURE 1Locations of the studied coastal time-series, EOS and BBMO, on a map of climatological surface currents in the month of July(Martıńez et al. (2022)), also including an schematic of the Northern Current.
FIGURE 5 Decomposition of the observed long-term trends in sea surface pH T in situ (DpH obs , grey; Table 1) into the contributions of their main drivers (values in Table 2) following Garcıá-Ibañez et al. (2016): changes in sea surface temperature (DpH Temp , green), salinity (DpH S , navy), TA (DpH TA , light blue), and DIC (DpH DIC , red) from the two coastal time-series, (A) BBMO and (B) EOS.The addition of the pH T in situ changes from each driver are also given (DpH sum , white; Table2).Trends reported in (x 10 -3 ) yr -1 and error bars represent the standard error of the estimate.

TABLE 2
Decomposition of de-seasoned pH T in situ trends at EOS and BBMO.