The CTD-Oxygen in the WMED (CTD-O2WMED) dataset comprises 1382 dissolved oxygen profiles collected with CTD probes across 25 CNR research cruises (Table 1). Figure 1 shows the spatial distribution of these profiles, illustrating broad coverage throughout the northern WMED and along key hydrographic transects. The majority of measurements are concentrated in the eastern portion of the WMED, including the Ligurian and Tyrrhenian seas and the Tunisia–Sicily–Sardinia region.

Spanning the period from 2004 to 2023, the dataset provides robust temporal coverage, particularly from 2004 to 2015 (see Fig. 2a). Notably, the years 2005, 2006, 2010, and 2012 comprise the highest number of CTD stations, coinciding with monthly surveys (Fig. 2). While the temporal distribution remains consistent between 2004 and 2015 (except for 2014), the number of sampled stations decreases significantly after 2016.

CTD-O2, which stands for sensor-based dissolved oxygen, was measured using Sea-Bird SBE43 sensors mounted on the CTD rosette frame. For each cruise, all variables were checked and converted to standard units to ensure uniformity. Dissolved oxygen was the converted from milliliters per liter (mLL-1) to micromoles per kilogram (µmolkg-1) using potential density and the conversion factor 44.66 µmolO2L-1, ensuring consistency across datasets. In addition, discrete samples collected at various depths from Niskin bottles were analyzed via Winkler titration on board. This dataset focuses exclusively on CTD sensor data, so the discrete data are not included in the final product.

The Gibbs Sea Water (GSW) toolbox (https://www.teos-10.org/pubs/gsw/html/gsw_contents.html, last access: June 2024) was used to compute conservative temperature (CT), absolute salinity (SA), and potential density to provide an accurate representation of seawater properties during the first quality check process.

In the second step, the CTD-O2 data from cruises listed in Table 1 were post-calibrated using Winkler discrete data following standard protocols (Grasshoff et al., 1983, 1999; Langdon, 2010) and accounting for sensor drift and hysteresis in line with the procedures by Janzen et al. (2007) and Uchida et al. (2010). We followed the Sea-Bird Electronics Application Note 64-2 (SBE 43 DO Sensor calibration and data correction, https://www.seabird.com, last access: January 2024). Residuals between the Winkler (O2 bottle) and sensor CTD (O2 sensor) measurements, matched based on pressure, were evaluated after post-calibration, following Uchida et al. (2010) (Fig. 3). When more than one SBE43 sensor was deployed during a cruise, the sensor with the lowest residuals relative to Winkler samples was used in the further assessment.

Figure 3a shows the residuals (O2 bottle – O2 sensor) plotted against pressure, with cruises color-coded by start date. Differences of up to ±10 µmolkg-1 are observed especially in the upper 800 dbars. Figure 3b summarizes residual distributions for each cruise using whisker plots, where it is easy to identify cruises with high variability (standard deviation of the mean residual >7 µmolkg-1): 48UR20041006, 48UR20050412, 48UR20080905, and 48QL20171023. For cruise 48UR20080905, only five Winkler samples were available, which limited the post-calibration quality. In cruise 48UR20050412, which comprised two legs, discrete samples from only one leg were used for the entire cruise calibration, further affecting quality of the CTD-O2 data. Figure 3c summarizes the cruise-level agreement using mean residuals and the percentage of residual values within ±2 µmolkg-1. Following Uchida et al. (2010), residuals within this threshold are considered acceptable. Cruises were categorized as follows: green encompass ≥40 % of residuals within ±2 µmolkg-1 (good agreement), blue 19 %–39 % of residuals within ±2 µmolkg-1 (moderate or uncertain agreement), and gray <19 % of residuals within ±2 µmolkg-1 (poor agreement and probably systematic bias in CTD-O2 data). 8 cruises showed good agreement, 11 were moderate, and the remainder exhibited systematic positive or negative biases, as pointed to by the mean value of the residuals (Fig. 3c). These deviations may reflect sensor drift, post-calibration issues, or bottle-handling errors.

An external reference dataset for dissolved oxygen discrete measurements was used to compare the CNR CTD-O2WMED data. Discrete Winkler measurements on those reference cruises (Table 2, Fig. 4) were performed following the GO-SHIP (Global Ocean Ship-Based Hydrographic Investigations Program) protocol ensuring robust O2 data quality to better than 1 µmolkg-1 (Langdon, 2010). Among these reference, cruises 06MT20011018 and 06MT20110405 (Hainbucher et al., 2014) are significant surveys, contributing to the GLODAPv2 data product (Olsen et al., 2016, 2020). They underwent full quality control with no bias correction applied to the original data. Similarly, cruises 48UR20070528, 29AH20140426, and 06M220180302, which are being assembled into the consistent carbon and ancillary dataset CARIMED (CARbon, tracer, and ancillary data in the MEDsea; Álvarez et al., 2019), have been rigorously quality controlled. Cruises 29AJ20160818 and 11BG20220517 conducted in 2016 and 2022 (Tanhua, 2019a, b; Jullion, 2016; Schroeder, 2022; Schroeder et al., 2024) followed GO-SHIP protocols under the Med-SHIP (Mediterranean Sea repeat hydrography) framework, which emphasizes the collection of high-quality hydrographic and biogeochemical data for long-term climate studies (Schroeder et al., 2015, 2024). These seven reference cruises were selected based on data quality and geographic overlap with the CTD-O2WMED dataset. Figures 1 and 5 show the regional distribution of the reference data, which aligns well with the CTD-O2WMED cruise tracks, particularly in the Tyrrhenian Sea and Algerian Basin, which are the two most frequently sampled subregions.

Following unit conversion and post-calibration with Winkler measurements (Sect. 2.2), each cruise underwent an outlier screening process. Data quality flags were assigned following the World Ocean Circulation Experiment (WOCE) standards: flag 2 for acceptable values, flag 3 for questionable values, flag 4 for bad values, and flag 9 for missing or not measured data. Flagging was tailored to the expected accuracy and precision of CTD-O2 data for each individual cruise, which depends on the available measurements for Winkler oxygen and the stability of the CTD-O2 sensor. Note that metadata information was scarce for some cruises. Property–property plots was analyzed for each region, and CTD-O2 values identified as outliers in multiple plots were flagged as questionable. Approximately 0.2 % of the CTD-O2 data were flagged as outliers (flag 3). The first QC is inherently subjective, relying on the expertise of the analyst reviewing the data.

To illustrate the regional oxygen distribution, Fig. 5a presents the vertical CTD-O2WMED profiles by pressure across 10 WMED subregions defined in Fig. 5b according to Manca et al. (2004). Gray lines show the full dataset, while blue lines indicate the range of oxygen concentrations in each subregion, revealing clear spatial and vertical patterns. The vertical distribution of dissolved oxygen in the WMED reflects a balance between air–sea gas exchange, biological activity, and regional circulation. Surface concentrations remain near atmospheric saturation due to gas exchange and photosynthesis activity. As organic matter sinks and is remineralized, oxygen is consumed at depth, generating vertical gradients. Unlike other ocean basins where pronounced OMZs develop, the WMED remain relatively well oxygenated thanks to episodic deep convection in the Gulf of Lion. This process ventilates the WMDW, which then spreads across subregions including the Algerian Basin and Ligurian Sea, known to be among the best-ventilated areas (Schneider et al., 2014).

In the WMED, a recurrent feature is the intermediate OML, typically found between 300 and 600 db. This layer coincides with the core of the Eastern Intermediate Water (EIW), which is warmer, saltier, and consistently lower in oxygen than surrounding waters (Tanhua et al., 2013; Coppola et al., 2018; Mavropoulou, 2020). The OML's depth, thickness, and intensity vary by region and year, depending on remineralization rates, mixing, and circulation (Coppola et al., 2018). In the Tyrrhenian subregion (DT3, DT1), the oxygen increases again below the OML, suggesting the influence of deeper, more oxygenated waters. In contrast, the lowest oxygen levels in the WMED are found in the Sicily Channel (DI3) and Tyrrhenian Sea (DT3, DT1), where the EIW is prominent and deep ventilation is absent. The Alboran (DS1) and Balearic (DS2) seas, by contrast, show relatively well oxygenated profiles throughout the water column. This reflects both the presence of WMDW and enhanced vertical transport due to mesoscale and the submesoscale processes that facilitate the downward movement of oxygen-rich surface waters (Middleton et al., 2025).

In order to assess the internal consistency and precision of the CTD-O2WMED data across CNR cruises, we computed for each cruise the median and median absolute deviation (MAD) of deep oxygen values flagged as good (depth >800 dbar, see Fig. 6 and Table S3). This approach minimizes the influence of atmospheric forcing, mesoscale variability, and residual outliers. Figure 6a illustrates the spatial variability in deep-water oxygen concentrations across the WMED. A pronounced east–west gradient is evident, with lower oxygen levels (blue tones) in eastern subregions such as the Tyrrhenian Sea, Sardinia Channel and Sicily Channel, and higher oxygen levels (green to yellow) in western regions, where deep convection processes in the Gulf of Lion enhance ventilation. The MAD was used as a proxy for precision, and overall MAD values ranged from 0.1 to 7.5 µmolkg-1. In well-sampled subregions (≥5 cruises) high MAD values or anomalous medians were used to identify potentially biased CTD-O2 cruises.

Below we summarize the findings by subregion:

Ligurian Sea (DF3): of the five cruises, cruise #3 (48UR20050412) and #21 (48UR20121108) had the highest MAD. Cruise #2 (48UR20041006) had a median consistent with the others, but that of cruise #3 was ∼10 µmolkg-1 lower, and that of cruise #6 was 16 µmolkg-1 higher despite a low MAD, suggesting possible bias or calibration errors.

Following an approach adapted from Olsen et al. (2016), large MAD values combined with anomalous medians and limited spatial coverage were used to identify cruises with low internal precision and potential systematic biases:

Cruise #3 (48UR20050412), recurring multiple subregions, consistently showed high MAD or biased medians.

To support the regional assessment of data quality, a summary table (Table 3) compiles the number of cruises per subregion, highlights those with high MAD, and identifies cruises exhibiting anomalous median values. This table facilitates the identification of cruises with potential CTD-O2 bias issues and helps assess the overall initial internal consistency of the CTD-O2 dataset across the WMED.

The secondary quality control (second QC) procedure involves comparing the CTD-O2 data from CNR cruises (Table 1) with selected reference cruises (Table 2). These reference datasets are assumed to be accurate and precise with low temporal variability, particularly in the deep ocean. However, this assumption may not fully hold for recent cruises due to strong spatial gradients and potential long-term trends in dissolved oxygen concentrations in the Mediterranean Sea.

Global synthesis efforts like GLODAP (https://www.glodap.info, last access: June 2024) (Key et al., 2004; Olsen et al., 2016, 2019; Lauvset et al., 2024) and CARINA (Key et al., 2010) typically adopt a 1 % threshold in crossover analysis to identify measurement biases and ensure inter-cruise consistency for oxygen. However, applying this same threshold to the Mediterranean may not be appropriate due to its unique oceanographic variability. Here, we assess the validity of reference cruise data, determine appropriate consistency thresholds, and identify the most temporally stable depth range for the crossover analysis in the WMED.

This section describes the obtained crossover results and discusses the correction factor and the final adjustments applied, if needed, to each CNR WMED cruise to finally probe the increased consistency of the CTD-O2 WMED data product released. Each crossover result was evaluated, somehow subjectively, considering its quality, which would depend on the number of available stations, the calculated standard deviation of the offset, the depth range covered, and the regional variability. A total of 265 crossover results for each cruise including regional clusters were carefully inspected, and only the appropriate ones, following the conservative approach, were considered to obtain the correction factors (Fig. 10). The final correction factors applied are summarized in Table 4 and only applied when the offset exceeded the ±2 % threshold limit. Overall, only a few cruises exhibited deviations outside the ±2 % limit and thus required adjustments (Fig. 10). The analysis indicated that deep oxygen values from cruises #2 (48UR20041006), #8 (48UR20060928), #211 (48UR20130604), and #222 (48QL20151123) required upward correction to align with the reference dataset. Conversely, cruises #3 (48UR20050412), #5 (48UR20051116), and #24 (48QL20171023) showed slightly higher values compared to their respective reference and were adjusted downward. 14 cruises did not require any correction as their values were consistent with the reference dataset. Cruises #25 (48DP20180918), #27 (48DP20221015), and #28 (48DP20230324) were excluded from the crossover analysis due to an insufficient number of deep stations below 800 dbar, but their data remain part of the final dataset. The reader is encouraged to compare the descriptions with the corresponding plots in Fig. 10 and the crossover summary figures in Figs. S3–S9.

After applying the correction factors listed in Table 4, the offsets were recalculated to validate the adjustments. As shown in Fig. 10, the corrected values (in blue) reduced the recalculated offsets and improved consistency. To assess the overall consistency of the adjusted CTD-O2WMED dataset, we computed the WM of the crossover offsets after adjustments (Fig. 11). The internal consistency of the final CTD-O2WMED dataset was estimated to be 0.998. These adjustments reduced potential biases linked to errors related to methodological discrepancies, resulting in improved coherence across the dataset.

This subsection provides a rationale for the proposed correction factors in Table 4. It includes interpretations of significant offsets even in cases where no final correction was applied but poor data quality was observed. Cruises not mentioned below were found to be consistent with the reference data and required no further review.

Cruise #2 (48UR20041006): two crossovers were found in the Alboran Sea, five in the Algerian Basin, and four in the Tyrrhenian Sea. A consistent mean offset of 0.96±0.005 across all regions suggests CTD-O2 values were ∼ 4 % lower than the reference cruises (Fig. 12). This is also supported by high residuals between Winkler and sensor data (>2 µmolkg-1, Fig. 3) and by the unusually low regional deep averages (Fig. 6). Given that deep WMED ventilation prior to 2004 was still stable, this is a 3.2 % increase. An adjustment of 1.032 is applied.

To resume, the CTD-O2WMED original version includes all cruises that have passed the first levels of quality checks but have not yet been corrected through crossover analysis (i.e., secondary quality control), so no adjustments were applied in the original version. However, the adjusted version of the dataset CTD-O2WMED adjusted builds upon the original dataset by incorporating the secondary quality control adjustments.