The CO₂ flux to the atmosphere in the East Atlantic during November 2019 was driven by upwelled water with low SST, low Chl-a concentration and high wind speed (Supplementary Figure S3, Supplementary Material). In this context, the physical upwelling of CO₂-rich deep waters to the surface can exceed the capacity of biological drawdown, leading to persistent CO₂ outgassing (Siddiqui et al., 2023; Roobaert et al., 2024).

The pCO_2sw thermal component (pCO_{2sw_TD}) followed the changes in SST and decreased by more than 100 µatm from the beginning of November 2019, when the region was characterised by elevated temperature (> 24 °C), to January 2020 over a period of which SST decreased more than 14 °C (Supplementary Figure S2, Supplementary Material). However, the non-thermal component, pCO_{2sw_N}, counteracted the changes in pCO_{2sw_TD} by increasing approximately 100 µatm over the same period (Supplementary Figure S2, Supplementary Material). Between 25 and 28 October 2019 the SD 1030 moved southward and crossed an area with higher Chl-a, which primarily affected the pCO_{2sw_N} and decreased the CO₂ flux. In the first weeks of November 2019, heading southwards, the CO₂ flux was positive (directed into the atmosphere), mainly driven by the SST increase (Figure 3B) and the relatively low Chl-a concentrations. During this period, some small negative CO₂ flux values were observed concurrently with a strong reduction in pCO_2sw (Figure 3D) suggesting that the SDs crossed an area previously characterised by elevated primary productivity (e.g., Supplementary Figure S3A, 28 October 2019. After 28 November 2019, the SDs crossed an upwelling area with lower SST and the pCO_2sw increased by about 30 µatm as the surface water was enriched with CO₂ from below (Figures 3A, D). Here, the CO₂ flux was mainly driven by the non-thermal component (pCO_{2sw_N}). After 6 December 2019, the SDs moved away from the upwelling area, and a sharp decrease in pCO_2sw was observed concurrently with an increase in SST. Until 16 December the flux was almost close to zero with the exception of a strong peak that occurred on 12 December, mainly triggered by wind driven coastal upwelling (i.e., increase in wind speed and strong decrease in SST; Supplementary Figure S3, Supplementary Material). From January to early March 2020, moving northwards from the Canary upwelling system, the SDs sampled an area characterised by low variability in pCO_2sw and pCO_2atm. In this period the thermal and non-thermal components of pCO_2sw were relatively similar, and at the end of this period, the non-thermal component started to increase inversely to the thermal component (Supplementary Figure S2, Supplementary Material).

Eddies are frequently observed in this area (McGillicuddy et al., 2003; Katzenmeier et al., 2024), which was also the case during the R/V Meteor cruise (M160) in fall 2019. Cyclonic eddies promote the upwelling of deeper water rich in CO₂ and nutrients from remineralized organic matter. Inside these eddies, the surface pCO_2sw can easily exceed the pCO_2atm, and thus facilitating the CO₂ efflux to the atmosphere. The CO₂ flux may also have been influenced by the island orography, which affected wind patterns and led to the formation of small frontal systems. North of Cape Verde and towards the Strait of Gibraltar, the sea surface was significantly undersaturated with respect to CO₂, which resulted in CO₂ influx.

A particularly interesting aspect is the high variability in SST, SSS, and O₂ (Figure 3C) concentration observed between 11 and 16 December, coinciding with measurements conducted by METEOR (Devresse et al., 2023). During this period, the SD crossed twice a small cyclonic eddy, and the CO₂ flux response was markedly different in the same area just two days apart (Supplementary Figure S4). The signal clearly indicates the presence of an upwelling, as evidenced by a sudden decrease in SST accompanied by a corresponding increase in SSS, which is also clearly observed in the METEOR data (Devresse et al., 2023), along with a strong increase in O₂ concentration and oxygen supersaturation coinciding with the SST minima.

This structure, which was identified as a cyclonic eddy (Brava eddy, Devresse et al., 2023) caused a general increase in CO₂, mainly due to outgassing associated with upwelled deep waters enriched in CO₂. In the core of the eddy the water mass was colder with higher concentration of dissolved inorganic nitrogen and Chl-a concentrations than the surrounding waters with a strong doming of the nitracline. This could explain the sudden changes of CO₂ fluxes observed around mid-December (Figure 3E). Furthermore, wind effects, which were stronger on the 11^th December than on the 16^th December, influenced phytoplankton biomass, leading to its increase following the reduction in wind intensity. The non-thermal component of pCO_2sw increased, resulting in a decrease in R. This can be explained by the initial upwelling-driven outgassing, enhanced by wind, followed by a reversal of the CO₂ flux toward the ocean due to the biological response associated with primary production, which removed CO₂ from the surface waters. This highlights how high resolution observations are essential to capture the strong variability of CO₂ fluxes associated with fine-scale oceanographic features such as mesoscale and submesoscale eddies.

After 7 March 2020 the SDs reached the Mediterranean and crossed the Alboran Sea in almost a week (Figure 4A). The area showed strong variations in SST and SSS (Figure 4B), which influenced the biogeochemical variables. In early March 2020 the variability of pCO_2sw increased, and pCO_2sw reached the lowest value recorded during the mission (350 µatm). The observed variability can be explained taking into account the characteristics of the region, where Atlantic and Mediterranean waters are coexisting and interact at different spatial and temporal scales (Capó et al., 2021; Poulain et al., 2021). Respiration processes, winter mixing, and spring blooms played a key role when the SDs moved from the Alboran Sea to the Ligurian Sea. Eddies are widespread in the area, as shown in Figure 4A which highlights the anticyclonic eddies in both the western and eastern Alboran Sea from satellite SST images. These eddies are caused by the surface inflow of Atlantic water that crosses the Strait of Gibraltar, which gradually mixes with the resident Mediterranean water (Tintoré et al., 1991; Mojica et al., 2018; Sánchez-Garrido and Nadal, 2022).

In the Alboran Sea, the SDs crossed two chlorophyll fronts (Chl-a >1.5 mg m^-3, Figure 4A), as evidenced by the O₂ supersaturation (Figure 4C). This influenced both the pCO2sw concentrations, which in the region was mostly driven by non thermal processes (Figure 4D), and CO2 flux (Figure 4E). Indeed as observed by Roobaert et al. (2024) and Frazão et al. (2025) phytoplankton blooms can rapidly draw down CO₂, creating transient CO₂ sinks during productive seasons especially in temperate ocean where spring blooms can induce sharp declines in pCO_2sw over a few days to weeks, often exceeding the influence of temperature during the bloom peak (Lévy et al., 2024).

The Alboran Sea is one of the most productive areas of the Mediterranean Sea according to estimates based on satellite data by Uitz et al. (2012). The Chl-a concentration in boreal spring along the northern flank of the Alboran Sea is influenced by the El Niño Southern Oscillation (ENSO) that is responsible for the development of anomalous distribution patterns of Chl-a. Furthermore, ENSO influences the western Mediterranean by altering the winds over the Alboran region and, consequently, the wind-driven upwelling that occurs along the southern Spanish coast in spring. This, in turn, modifies the surface Chl-a concentration: a high (low) Chl-a concentration in the northern Alboran Sea during spring is expected under La Niña (El Niño) conditions in the tropical Pacific during the preceding winter (Lopez-Parages et al., 2022). Also in the past elevated Chl-a concentrations have been reported for the western Alboran Sea (Minas et al., 1991; Rodríguez et al., 1998; Lopez-Parages et al., 2022), while particulate primary production can reach up to 632 mg C m^–2 d^–1 along a coast-to-coast gradient crossing the western Alboran Sea gyre (Moran and Estrada, 2001) and 700 mg C m^–2 d^–1 in the northern part of the eastern gyre (Videau et al., 1994).

The SDs arrived in the Ligurian Sea at the beginning of April 2020 after crossing the Gulf of Lion (Figure 5A). The pCO_2sw fluctuation was high and the pCO_2sw concentration low in the northwestern Mediterranean at this time of year. The area was characterised by low SST and undersaturated O₂ concentrations highlighting the presence of intense vertical mixing (see Figure 5A in Martellucci et al., 2024). Convection can be intense in February and March in the Gulf of Lion due to cold and dry winds and preconditioning events (Schroeder et al., 2008; Houpert et al., 2016; Testor et al., 2018; Many et al., 2021), which refers to the set of oceanic conditions that favour deep convection, such as the presence of a weakly stratified water column and the accumulation of dense water from previous winters. These conditions induce an increase of surface water density sensitive to the winds blowing. The convection process is less pronounced in the Ligurian Sea where wind forcing and air-sea heat flux are usually less intense (Coppola et al., 2018). A recent study from Ulses et al. (2023) demonstrated that the deep convection zone in the northwestern Mediterranean Sea served as a moderate CO₂ sink for the atmosphere in the years 2012 to 2013, when convection was particularly strong. Between the beginning of March and the end of April 2020, the pCO_{2sw_N} dominated the pCO_2sw signal (Supplementary Figure S2, Supplementary Material). From the Alboran Sea to the Ligurian Sea, respiration possibly prevailed over photosynthesis, increasing the pCO_{2sw_N} while O₂ was consumed. In the western Mediterranean, the end of deep vertical mixing favours the onset of a spring phytoplankton bloom with a peak in primary production and phytoplankton concentration at the surface which generally occurs in April in this region (Mayot et al., 2017; Kessouri et al., 2020). The 2020 spring bloom produced a strong increase in Chl-a between March and April 2020 (Supplementary Figure S4, Supplementary Material) resulting in a pronounced uptake of CO₂ from the water (Figure 5E), decreasing pCO_{2sw_N}, while O₂ production increased during the bloom event (O_sat > 106, Figure 5C).

On 9 April 2020 the SDs left the area off Toulon moving towards the W1M3A site and sailing about 40 km from the coast. The area is characterized by the Northern Current (NC) that is a key boundary current in the western Mediterranean (Berta et al., 2018). This current originates in the Ligurian Sea from the convergence of the Western and Eastern Corsica currents. At surface, it transports the old Atlantic Water (AW) southwest ward, influencing the circulation of the northwest Mediterranean. The NC extends about 40 km off the coast and its variability is influenced by winds and freshwater inputs. This pattern was clearly detected by the SDs until 15th of April (Figure 5B). Until then the pCO_2sw values ranged from 360 to 400 µatm and the SST varied by 2 °C. This can be ascribed to the presence of strong mixing processes, driven by northwesterly winds, that influenced the pathways of the NC current forming a frontal zone near the coast, separating warm coastal waters from colder mixed waters of the Gulf of Lion (Prieur et al., 2020).

From mid-April 2020 SST increased as the SDs moved eastwards (Figure 5A), and pCO_2sw increased from 360 µatm to 400 µatm. This increase was consistent with the values reported by Coppola et al. (2020) for the seasonal cycle in the Ligurian Sea. Here, pCO_2sw values usually range from 300 to 550 μatm, as already reported for the DYFAMED site. The glider deployed during the mission helps to spatially characterize the observed area and to generate predicted carbonate system variables using the CANYON-MED neural network (see Section 2.5). The glider moving from Nice to Calvi measured SST similar to those measured by the two SDs at matching locations.

The O₂ concentration decreased from 260 µmol kg^-1 to almost 230 µmol kg^-1 as the SDs left the Ligurian Sea. pCO_2sw was comparable between the platforms.

In the Ligurian Sea, the main processes governing the pCO_2sw at the sea surface on an annual scale, after removing the temperature effect, are vertical mixing and biological production (Copin-Montégut et al., 2004; Merlivat et al., 2018; Coppola et al., 2023). However, during the studied period, in the Ligurian Sea the non-thermal component of pCO_2sw prevailed over the thermal one (R<1; Figure 5D) suggesting that biological activity might have influenced the pCO_2sw variability in this area during April 2020 (Supplementary Figure S4, Supplementary Material).

In early May 2020, the pCO_2sw started to increase (from 350 µatm to 450 µatm) crossing the Tyrrhenian Sea, concurrently with temperature increase (Figure 2A). The pCO_{2sw_TD} reflected the change in SST, being predominant in the overall pCO_2sw signal in the region.

The neural network (CANYON-MED) derived estimates of pCO_2sw closely matched the SD measurements (SD 1030, R² = 0.95, p<0.05; SD 1030, R² = 0.88, p<0.05), with only a small offset of 15–20 µatm observed as the SD moved along its path (Figure 6, Supplementary Figure S6). This offset fell within the expected error range of the neural network-derived pCO_2sw estimates (see section 2.5) and aligned with data collected by the W1M3A fixed ocean station. However, it is important to note that between 28 April and 1 May 2020, when the SDs circled the W1M3A station, sea conditions shifted from calm to very rough (significant wave height of 2.5 m), leading to water mixing that may have influenced the measurements. By 4 May 2020, sea conditions had returned to calm.

Further analysis of discrepancies between pCO_2sw measurements from W1M3A and the SD revealed that differences primarily stemmed from the total gas stream pressure measurement provided which was used to calculate pCO_2sw. In contrast, better agreement was observed when comparing xCO_2sw measurements, with an absolute mean error of 3.9 ppm. Figure 6 illustrates that CANYON-MED predictions for pCO_2sw are consistent with direct measurements from SD 1030 and sensors deployed on fixed buoys (DYFAMED and W1M3A).

Occasionally, discrepancies in pCO_2sw between CANYON-MED predictions and in situ measurements are seen (Figure 6). These are mainly due to the neural network’s limited spatial resolution, which cannot capture sub-mesoscale variability. This is especially evident in the northern Adriatic Sea, where complex coastal influences (e.g., river inputs, groundwater discharges) fall outside the neural network original training scope (open sea in the Mediterranean Sea), leading to notable mismatches with data from SDs and fixed stations like PALOMA and MIRAMARE. Despite these limitations near the coast, CANYON-MED performs well in offshore regions such as the Ligurian, Ionian, and southern Adriatic Sea, where its predictions align closely with SD measurements. This suggests that CANYON-MED can effectively complement the in situ measurements and help fill observational gaps where direct pCO_2sw measurements are unavailable due to sensor failure or maintenance.

Starting from mid-June to the beginning of July 2020, the SDs crossed the southern Adriatic Sea (Figure 7A). This area exhibits a strong variability triggered by the inflow of cold and less salty water from the northern Adriatic and warm and salty water entering the basin through the Otranto strait in the south (Ingrosso et al., 2017; Cantoni et al., 2024; Martellucci et al., 2024). This, drives the formation of small scale eddies (e.g., baroclinic instability) mixing these different water masses. The SD pCO_2sw showed an increasing trend over the entire period (25 June - 1 July 2020) which was similar to the measurements at the E2M3A site (Figure 7D). This was likely due to the increasing SST (Figure 7B) as confirmed by the analysis of the thermal and non-thermal components of pCO_2sw suggesting a strong influence of temperature rather than biological processes (R>1; Figure 7D). This was also highlighted by very low fluctuation in the O₂ concentration Figure 7C over the period that could have been related to biological processes, as the southern Adriatic is an oligotrophic area. Pronounced day-night variations of SD pCO_2sw were also evident in correspondence with SST relative maxima. The CO₂ fluxes (Figure 7E) showed positive values over the entire period indicating CO₂ outgassing from the sea. While Ingrosso et al. (2017) reported that dense water formation was among the primary drivers of CO₂ uptake in the area in winter, the SD measurements highlighted an opposite pattern during summer. In addition, the CO₂ fluxes calculated from SD measurements and the fluxes calculated from the pCO_2sw data of E2M3A show again good agreement in their general patterns (R₂ = 0.72, p ≤ 0.05). SST and SSS measured by the two SDs between 25 June and 1 July 2020 showed similar values at distance less than 10 km apart and also a good agreement with the E2M3A data (R₂ = 0.72, p < 0.005; in contrast to the other fixed stations, the SD surrounded E2M3A for a sufficient amount of time to perform reliably statistics). However, during 26–28 June 2020, when the SDs sailed eastward along the Bari-Dubrovnik track, the SSS decreased from that measured at E2M3A, evidencing that the SDs entered the less salty coastal waters.

The carbonate balance and dynamics in the shallow (mean depth is approx. 30 m) northern Adriatic sea is strongly influenced by river discharges. The Po river is the major source of freshwater and nutrients in the area (e.g., Cozzi and Giani, 2011), but also smaller rivers influence the carbon chemistry of the region (Giani et al., 2023). Between 13 and 14 July 2020, the SDs crossed a plume of the Po river in the northern Adriatic (Figure 8A). The plume extended eastward from the Po delta on the western Italian coast as evidenced by the lower SSS and higher SST measured by the SD (Figure 8B). The eastward extension of the plume was caused by the northeasterly katabatic Bora wind blowing in the area from 12 to 15 July (Supplementary Figure S6, Supplementary Material). Chl-a in the plume was higher (Supplementary Figure S7, Supplementary Material) as a response to the river borne nutrients, and the corresponding phytoplankton bloom caused a marked decrease in pCO_2sw reaching its lowest values of about 370 µatm which resulted in a CO₂ influx (Figures 8D, E). The CO₂ drawdown by phytoplankton can be seen by the increase in the pCO₂ ratio (green line in Figure 8D, Supplementary Figure S2) due to the decrease of the non thermal pCO₂. This was the only occurrence of CO₂ influx in the Adriatic Sea during the demonstration mission (Figure 2B).

On 16 July 2020, before entering the Gulf of Trieste, the SDs crossed the Isonzo river plume (Figure 8A) as marked by a decrease in SSS (~ -1.6) and increase in SST (~ +1.1°C) (Figure 8B). However, differently from the crossing of the Po river plume, the SDs did not record marked pCO_2sw changes. The pCO_2sw increase due to the thermal effect was compensated by the non-thermal decrease presumably due to the CO₂ drawdown caused by primary productivity as also suggested by the increase of O₂ saturation (Figure 8C). Successively, from 17 to 18 of July 2020 a new outburst of the Bora wind was associated to an increase in pCO_2sw, despite the decrease in SST (~ -1.0 °C) and a switch from thermal (R = 1.14) to non thermal (R = 0.95) control. Strong positive CO₂ fluxes from the sea, up to 80 mmol m^-2d^-1, were measured (Figure 8E) turning the Gulf of Trieste from a weak (FCO₂ = 2.6 mmol m^-2d^-1 on average) to a strong (FCO₂ = 30.7 mmol m^-2d^-1 on average) CO₂ source. Contrary to what might be expected due to CO₂ outgassing, this increase in pCO_2sw can be explained by the mixing with bottom waters induced by northeasterly winds, which promote coastal upwelling. When the SDs entered the Gulf of Trieste, the water column was strongly stratified, with bottom waters depleted in O₂ (down to 132 µmol kg^-1; 56% O₂ sat) and enriched in CO₂ (up to 641 µatm in the middle of the gulf; Supplementary Figure S8, Supplementary Material). The increase of DIC and pCO_2sw and the decrease of pH and O₂ concentration below the pycnocline is typically observed in the Gulf of Trieste (Cantoni et al., 2012; Ingrosso et al., 2016a, b) during summer when the microbial respiration prevails over the primary production (Fonda Umani et al., 2012). The results reported here are consistent with previous findings by Turk et al. (2010); Cossarini et al. (2015) and Urbini et al. (2020) and further strengthens the role of this area as a CO₂ source during summer and highlights the interplay of multiple factors in controlling sea surface pCO_2sw variability. This phenomenon further strengthens the role of this area as a CO₂ source during summer and highlights the interplay of multiple factors in controlling sea surface pCO_2sw variability.