The glider has been deployed at sea near the eastern coast of Petite-Terre since 17 September 2021 (12 km southwest of the Horseshoe area: 12°53.5^′ S, 45°19^′ E). It is operated for 14 d on average before being recovered at sea to collect the full data set. The glider is then immobilized on land for 1 night to recharge the battery and perform regular maintenance before being redeployed the next day.

In the course of more than 2 and a half years of deployment, six gliders have been used for 72 deployments (see Table A1 in the Appendix).

These gliders are equipped with a CTD (conductivity, temperature, depth) instrument, either a SeaBird GPCTD or an RBR LEGATO. Dissolved oxygen sensors are also deployed, namely an SBE43F from SeaBird coupled with the GPCTD and an AROD-FT from JFE coupled with the LEGATO. These two sets of sensors, while having small technology differences (pumped sensors for SeaBird and unpumped sensors for RBR and JFE), provide comparable data (see Table D1). The scientific payload also includes an 1 MHz acoustic Doppler current profiler (ADCP) from Nortek (Signature1000 ADCP specifications with a casing modified for glider integration), a METS (dissolved methane sensor) from Franatech and a MINICO2 (dissolved CO₂ sensor) from Pro-Oceanus (see Table A1).

To carry out the mission, a specific and unusual sampling strategy was implemented. Until August 2023, the glider was limited to a maximum immersion depth of 1000 m. In order to stay as near as possible to the seafloor, where magmatic fluid emissions occur and should be sampled, the glider's navigation consisted of a three-phase progression: a downward phase where the glider reached a depth of 1000 m; a forward navigation phase, with about 10 ascent–descent phases (i.e. yo) between 900 and 1000 m; and a final phase of ascent to the surface. Dives carried out in the Horseshoe area last on average 8 to 9 h, with 6 h on average being spent between 900 and 1000 m, covering a distance of around 6 km. This radial navigation strategy, consisting of navigation at a fixed heading toward the central point of the Horseshoe structure, enables objective sampling of the zone of interest, covering all of its quadrants equally, with maximum sampling effort at its centre, decreasing progressively with distance (Fig. 2).

For the first time, two glider prototypes with a maximum immersion capacity of 1250 m have also been deployed in the area since August 2023. A slightly different navigation method was chosen, opting for spirals instead of straightforward dives. The radius of these spirals is 1.5 km in order to cover the entire study area as nearly as possible over the duration of a deployment.

Initially, a wide mapping of the area was chosen until August 2023 to detect seafloor fluid emissions and to get an idea of the physico-chemical properties over a large area. With this new navigation method, sampling is then focused on better characterization of the active fluid sites (Fig. 2).

These spirals also consist of a downward phase to a depth of 1250 m, followed by several yo's between 800 and 1250 m, and finally an ascent to the surface. These types of dives last an average of 10 h. This sampling method was chosen to ensure a good quality of data from dissolved gas sensors by flushing the sensors (see Sect. 2.2.2 and 2.2.3) and to focus the navigation on known active sites and their immediate surroundings.

The data produced during the continuous acquisition at high sample rates between September 2021 and April 2024 represent a substantial amount (∼2.2 million measuring points per sensor, corresponding to ∼22000 dives).

The CTD and dissolved gas sensors mounted on the SeaExplorer glider acquire measurements at a frequency of 1 Hz, subsequently averaged into 30 s time series available through the SEANOE data centre (see “Data availability” section). While data are averaged to 30 s intervals, this corresponds to approximately 5 m of vertical resolution at a typical glider ascent or descent speed. To optimize power consumption, the ADCP operates at a lower sampling frequency of 0.1 Hz, with its data similarly being averaged into 30 s intervals. As a result, fine-scale vertical structures or sharp shear zones may be partially smoothed, although the resolution remains adequate for capturing broader vertical patterns of current variability.

Based on UNESCO's best oceanographic practices (https://repository.oceanbestpractices.org/handle/11329/413, last access: 15 February 2024), which are also used for Argo floats, an objective and automatic quality control (QC) process was applied. Quality flags (QFs) are composed of four quality values (Table 2).

The procedure allows us to flag outliers but may be deficient in identifying some erroneous data. Tests presented hereafter relate to the other variables and are based on published methods (Pouliquen et al., 2010; Schmechtig and Thierry, 2016) and are recommended by the scientific community through international programmes, such as the EGO quality control manual for CTD and BGC (BioGeoChemical) data (https://archimer.ifremer.fr/doc/00403/51485/92689.pdf, last access: 4 August 2024).

A gross filter is applied on observed data using a global range test first (Table 3), with min and max values taken from World Ocean Atlas 2018 documentation (Garcia et al., 2019).

Values that fail this test are flagged with a QF of 4. To our knowledge there is not yet an international recommendation for CH₄ and CO₂; thus, values were chosen based on in situ measurements from MAYOBS cruises.

For all parameters, data acquired when the CTD pressure is negative (i.e. in-air measurement) were flagged as bad (QF of 4).

Furthermore, a large difference between sequential measurements, where one measurement greatly differs from adjacent ones, was also flagged bad if failing the following algorithm (only used to identify spikes in temperature, salinity and O₂ profiles; EGO quality control manual, 2022): 5Test=|V2-(V3+V1)2|-|V3-V12|, where V2 is the measurement being tested as a spike, and V1 and V3 are the values preceding and ensuing. The V2 value is flagged when the test value exceeds the following:

TEMP – 6.0 °C for PRES < 500 db or 2.0 °C for PRES ≥ 500 db

Membrane-based sensors (CO₂ and CH₄) are also deficient at high glider speeds (lag cannot be compensated for correctly). Thus, CO₂ and CH₄ data acquired at glider speeds exceeding 0.25 m s⁻¹ were flagged and excluded. This usually occurred when the glider rarely ascended in an alarm state.

All data provided have been quality controlled. Therefore, a QF of 2 is not used in this data set.

Following these various objective and subjective tests, the remaining data volume of each data set is provided in Table E1.

Every 6 months, a reassessment of the processing chain (algorithm, QC) and delayed-mode adjustments (drift, offset) is proposed.

Hydrological data presented are adjusted and associated with a QF of 1.

Time series of temperature, salinity and potential density are presented in Fig. 3; averaged vertical profiles are presented in Fig. 4; and a temperature–salinity diagram exhibiting the different water masses is shown in Fig. 5.

As for temperature, vertical profiles exhibit a relatively warm upper layer (∼0–100 m, 26–30 °C). The seasonal thermocline (steep thermal gradient of ∼0.1 °C m⁻¹) is observed between ∼100–200 m, and the permanent thermocline is located at ∼500 m and is mainly composed of South Indian Central Water (SICW, Miramontes et al., 2019).

Below 500 m, temperatures are in the range of 5–10 °C, with a minimum reached below 1000 m. In this layer, both Red Sea Water (RSW), which enters into the Mozambique Channel from the north, and Antarctic Intermediate Water (AAIW), which enters from the south, are found (Miramontes et al., 2019).

The vertical distribution of the salinity is more complex. Overall, the upper layer is characterized by values starting from ∼34.7 and 35.5 PSU. A subsurface salinity maximum is observed at ∼200–300 m, with salinity values reaching up to 35.5 PSU. At around 600 m, a local salinity minimum (34.6–34.8 PSU) is observed but is followed by a slight increase to reach ∼34.85 below 1000 m.

Deeper in the water column (i.e. below ∼100 m), the variability in hydrological properties is lower. However, disruptions in the vertical distribution of temperature and salinity are visible, such as between June and July 2022 (between the dotted grey lines in Fig. 3) and can be attributed to the general circulation of the area or the mesoscale variability. Below 1000 m, the variations are ∼1 °C and ∼0.1 for temperature and salinity, respectively (Fig. 6).

Most of the temporal variability is observed in the surface layer, above the 1024 kg m⁻³ isopycnal. These changes are particularly obvious in the temperature–salinity diagram (Fig. 5) indicating the succession of two distinct water masses (Collins et al., 2016). Indeed, low salinity values are typical of the tropical surface water and contrast with the higher-salinity subtropical surface waters (Di Marco et al., 2002). Surface temperatures exceeding 29 °C (which are regularly observed from December to April, Figs. 4 and 7) can also be the signature of the influence on the South Equatorial Current that contains Pacific waters (Di Marco et al., 2002) or associated with the transient presence of mesoscale eddies. Finally, in accordance with Wyrtki (1971), seasonal processes (and episodically tropical storms) may also account for the observed variability. The highest sea surface temperatures and low salinities are generally observed during austral summer, while, in winter, colder and saltier waters dominate. This seasonal variability is observed in this part of the water column (Fig. 7), with warm (∼27–28 °C) and salty (∼35.3–35.4) surface waters from November 2021 to July 2022 and from October 2022 to July 2023 during the warm and humid austral summer in Mayotte.

The analysis of the data set also highlighted the importance of smaller-temporal-scale processes, particularly vertical fluctuations of potential density levels (and temperature and salinity) per hour (Fig. 7). Likely a result of internal tides, the sampling strategy chosen, consisting of deep multi-yo patterns, unfortunately does not allow their quantification (Sect. 2.2.4).

Similarly to the hydrological data set, data presented are adjusted and associated with a QF of 1.

Measured O₂, CO₂ and CH₄ concentrations by the glider are shown in Fig. 3. Typical vertical distributions are observed (Fig. 4) and can be explained by ubiquitous physical (e.g. dissolution, sea–air exchanges) and biological oceanic processes (photosynthesis and respiration).

High O₂ concentrations corresponding to oxygen saturation concentration are measured (O₂ concentrations of about 180–200 µmol kg⁻¹, apparent oxygen utilization between 0–20 µmol kg⁻¹, Figs. 3 and 4) at the surface layer (0–100 m) because of both dissolution from the atmosphere and O₂ production by phytoplankton. The study area, located off the coast of Mayotte, is generally characterized by oligotrophic conditions, typical of tropical oceanic regions (Ternon et al., 2014). Such conditions are associated with low concentrations of nutrients in surface waters, which constrain primary production and consequently limit biological oxygen supersaturation. Nevertheless, the presence of detectable surface O₂ saturation suggests that, despite low nutrient availability, the prevailing light conditions and water column stratification support moderate phytoplankton activity in the upper layer. As the distance to the surface increases O₂ generally declines due to O₂ removal by consumption of deep-water organisms and by the decomposition of organic material by bacteria (Hedgpeth, 1957). In the glider data set, minimum O₂ values (O₂ in the range of 70–100 µmol kg⁻¹) are observed below 1000 m. In spite of this decrease, O₂ contents rise to a subsurface maximum at ∼400–500 m, with values reaching up to 200 µmol kg⁻¹. This high O₂ core (>180 µmol kg⁻¹) is characteristic of the South Indian Central Water (Di Marco et al., 2002).

Regarding the CO₂ vertical distribution (Fig. 4), it is essentially the reverse of O₂, mainly because both gases are involved in the same biological processes in opposite ways. At the surface, photosynthesis consumes CO₂, and, thus, concentrations are low (∼15 µmol L⁻¹). In deeper waters, CO₂ concentration increases as respiration exceeds photosynthesis, and decomposition of organic matter adds additional CO₂ to the water. In this data set, minimum CO₂ concentrations are found in the surface layer (0 - 100 m), with concentrations measured to be between 15–20 µmol L⁻¹, and maximum CO₂ values are measured below 1,000 m depth, with concentrations generally higher than 40 µmol L⁻¹ and sporadically exceeding 50 µmol L⁻¹. Moreover, the signature of SICW, with its oxygen maximum at ∼400–500 m (Di Marco et al., 2002), is not matched by a CO₂ minimum at this depth.

The vertical distribution of CH₄ differs significantly from the ones of O₂ and CO₂ (Figs. 3 and 4). Almost no CH₄ is detected in the 0–600 m layer (values are below 10 nmol L⁻¹). This is not surprising because the ocean is supposed to be depleted in CH₄, save for specific areas (methanogenesis in marine sediments, natural seepage by volcanoes or hydrothermal vents, pollution; Oremland and Taylor, 1978). Most CH₄ increases occur in the 900–1250 m layer, where the sampling effort is at its maximum. In this layer, when CH₄ anomalies are detected (above the sensor detection limit), a gradient of increasing concentration with depth is observed. High values relative to the background are regularly observed, with a maximum recorded on 18 February 2022, when CH₄ reached 120 nmol L⁻¹ at a 1000 m depth. Although observations above 900 m are scarcer than below, several vertical profiles also show significant CH₄ concentrations up to 700 m.

There is also variability in dissolved gas concentrations, with periods (e.g. September 2022 to mid-October 2022 and mid-November 2022 to February 2023) of decreasing CH₄ or CO₂ concentrations that are not yet explained (Fig. 8).

Considering the amount of data produced, semi-automatic methods are thus required to reduce this data set to relevant information. Here we focus on parameters that track magmatic fluid emissions (CO₂ and CH₄) and define anomalies as observations that deviate significantly from the majority of the data.

Identifying anomalies (that refer to fluid emissions) is challenging since it requires decoupling between natural variability (e.g. water masses, seasonality) and changes induced by fluid emissions. In particular, CO₂ and CH₄ signals are characterized by slowly varying background values related to dissolved gas accumulation and flushing over a large area. The CH₄ baseline is low (<10 nmol L⁻¹) and has a magnitude of variability of about 6 nmol L⁻¹. On the other hand, the CO₂ baseline is more elevated (∼45 µmol L⁻¹ at a 1250 m) because of the natural presence of CO₂ in seawater and CO₂ anomalies below 900 m, but fluctuations around the mean values do not exceed ∼3 µmol L⁻¹.

Results also show that both CO₂ and CH₄ baselines have similar temporal evolutions, supporting the hypothesis that this variability is likely to be real. For the purpose of anomaly detection, the baseline was thus subtracted from the raw data: 6Gas,anomalydeep(t)=Gas,measureddeep(t)-Gas,baselinedeep(t). However, the cause of this low-frequency variability still remains to be clarified. At this time, it is not clear if baseline fluctuations are related to accumulation–dispersion processes or induced by changes in fluid emission rates, both compatible with the chosen anomaly definition.

Of the 22 000 profiles, 5 % were associated with significant CH₄ anomalies (greater than the sensor detection limit plus twice the standard deviation), and 2 % were associated with significant CO₂ anomalies related to dissolved gas emissions (same definition as for the CH₄).

Data show that CO₂ and O₂ have similar patterns. Such a co-variation is expected in the ocean and is related to biotic processes. Examining this relation at a depth > 700 m (i.e. below the STSW) indeed confirms a high CO₂ and O₂ correlation (linear correlation coefficient, R2=0.96). Several single measuring points deviate from the linear relationship (Fig. 8). They are all found in the upper curve, i.e. at a depth where CO₂ is high and O₂ low. Most of these points are also independently associated with high CH₄ concentrations, which strongly suggests that these CO₂ anomalies are related to a non-biotic CO₂ source.

ADCP-derived water currents show a large profile-to-profile variability that encompasses, in all likelihood, spatial and temporal variability. The strongest currents are measured in the surface (0–100 m layer), but velocities remain elevated down to 1250 m (Fig. 9).

Overall, eastward velocities do not exhibit clear patterns. Values oscillate with no preferential direction (Fig. 9). Conversely, northward velocities are characterized by a distinguishable temporal variability. From several weeks to several months, the current direction changes, with long periods of time when the direction of flow remains unchanged. The strongest currents appear to be aligned with the continental slope (north–south axis), which may be related to barotropic currents.

Strong deep currents (≈0.4–0.5 m s⁻¹) below 900 m are also present in the area and are locally highly variable, with strong interactions with the bathymetry and the tide (Fig. 9).

Backscatter data estimated from ADCP measurements and expressed as the backscatter index (BI) are represented in Fig. 4 as an averaged profile. This averaged profile primarily depicts water mass optical property changes which are determined by phytoplankton and zooplankton abundance and mineral particle concentrations (Mullison, 2017; Gentil et al., 2020). Thus, as can be expected, the BI shows maximum values in the surface layer (0–100 m), where most of the biological activity takes place. Deeper in the water column, the BI is generally lower, although sometimes peaking at the level of the South Indian Central Water and in the ∼900–1250 m layer when crossing dissolved gas plumes or moving around the seafloor.

Similarly, the BI variability in the surface mirrors changes in temperature and salinity to a certain extent, and several glider dives showed a BI increase at depth, when the glider approaches the continental shelf. This variability is not fully understood and is likely to have a multi-factor cause resulting from a decoupling of the surface, subsurface and deep dynamics. Whatever the processes envisioned, in all likelihood, a direct contribution of magmatic fluid emissions can be discarded.

On the other hand, BI profiles are noisy and variable with depth, and, despite this large variability, the signature of bubbles and/or droplets appears to be unambiguous, associated with a large density of positive spikes of great amplitude and associated vertical velocity anomalies of about 15 cm s⁻¹, which is the ascent velocity of millimetre-sized gas droplets in seawater (Rehder et al., 2009; Leblond et al., 2014, Fig. 10). The BI increases in the deep layer were considered to be related to bubble and/or droplet plumes if several consecutive BI values exceeded ∼50 dB in at least six cells of the ADCP in order to discard the few possible misdetections at this depth. This was used as a criterion for BI anomaly detection, but each dive was also visually inspected.

Of the 22 000 profiles, 457 (∼2 %) were associated with significant BI anomalies. The relatively low occurrence of BI detections indicates that bubble and/or droplet plumes are likely to be of limited spatial extension (∼ hundreds of metres) especially compared with dissolved and neutrally buoyant gas plumes.

Although associating BI detections directly with a fluid emission active site is complex (uncertainty in glider positioning, tilt of droplets and/or bubbles plumes of several hundred metres because of deep and tidal currents), data show that most of the known active sites were indeed identified by the glider; 95 % of BI detections are found within a radius of 700 m from an active site, and the remaining 5 % are always found at a distance of less than 1.6 km.

Repeated dissolved gases and BI anomalies in the 800–1250 m layer (Fig. 10) provide evidence that elevated CO₂ anomalies in the 900–1250 m layer are related to magmatic fluid emissions from the seafloor.

BI detections outside of the 95th percentile observed around the Horseshoe zone may arise from intermittent, small, unidentified sites or false detections. Further analysis is needed to confirm the nature of these detections.

Temporal variations in dissolved gas concentrations (CH₄ and CO₂) and BI anomalies are presumably caused by a complex array of factors, including spatial variability related to the glider pathway. However, on the basis of our current knowledge, we assume that a large part of the observed changes can indeed be related to the variability of seafloor fluid emissions in the Horseshoe area, as well as the orientation and direction of the current at depth.

Over the 30 months of deployments, values of CH₄ and CO₂ show some interesting patterns, with a large profile-to-profile variability observed in dissolved gas deep anomalies. High anomaly values exceeding the detection limit of the sensor plus 2 times the standard deviation (20 nmol/l for the CH₄ anomalies and 7 µmol L⁻¹ for the CO₂ anomalies) are observed throughout the time series. In the 900–1000 m layer (more than 90 % of the data set), the maximum CH₄ anomaly value is reached on 21 February 2022 (116.9 nmol L⁻¹), and the maximum CO₂ anomaly value is reached on 4 September 2022 (29.2 µmol L⁻¹).

Although a direct correlation between currents and gas anomalies is speculative, our data suggest a potential impact of mesoscale structures on gas concentrations in the 900–1250 m layer. The underlying processes are still poorly understood from the sole analysis of the glider data set, but several processes would be worth investigating (e.g. trapping of gas bubbles in stratified layers, lateral dispersion by mesoscale flows, vertical diffusion of dissolved gases, or upwelling induced by topographic features or internal waves). Furthermore, additional bottom current data could potentially be useful for this analysis. In order to assess the spatial distribution, anomalies are plotted on maps (Fig. 11). This provides a comprehensive view of the area impacted by fluid emissions over the 30-month duration of deployments. In Fig. 11, data are binned in seven discrete concentration intervals and superimposed from weakest to strongest with circles of decreasing size. Only data that exceed the criteria of detection are coloured, and the maxima of the colour bar are equal to the 99th percentile.

The highest gas anomalies were all observed in the centre and in the immediate vicinity of the Horseshoe edifice, and the magnitude of the anomalies progressively decreased as the glider moved away from the centre. The radius of gas anomaly detection is around 10 km, and the total area impacted by fluid emissions spread over about 300 km². These maps, which gather all data from 17 September 2021 onwards, also show that the distributions of gas anomalies are not isotropic. In particular, relatively high concentrations relative to the far field are observed northward, which can highlight a preferred export direction for these quantities of dissolved gas.

Many questions still remain with regard to our understanding of the underlying processes. We can, for example, mention the spatial decoupling between acoustic and dissolved gas plumes or the contribution of physical factors in modulating the extension, direction and intensity of the plumes (currents, internal tides).

To deal with these scientific challenges, the synergy between the glider with other observation and measurement tools (CTD casts, remotely operated vehicles (ROVs), moorings), numerical models, and satellite products is promising. Several studies in this direction have been initiated, and although clear results are not yet available, ongoing work is progressing steadily.

Reference data acquired during MAYOBS cruises allowed for a cross-calibration exercise of the dissolved gas sensors. An exercise was carried out during the MAYOBS25 cruise in September 2023 and during the MAYOBS30 cruise in September 2024 by harnessing two gliders on the CTD cast in order to compare dissolved gases from glider sensors and Niskin samples. The aim of these calibrations is to quantitatively calibrate dissolved gas sensors in order to calculate the fluxes of fluids emitted at the seabed. The database will be updated accordingly for the time period covered by this study once the CH₄ and CO₂ sensor calibrations are finalized. Additional data sets will also be made available in the future as the mission is ongoing. Since the project is funded by the French government, all validated data will be made publicly accessible.