At LOCEAN, we have mostly used glass-tinted bottles (volume 20 or 30 mL) with a hard cap including an internal rim to minimize water exchange through the cap (referred to later on as a “common” cap). No independent internal stopper or insert is used, and the bottles are not collected full. For some, but not all, cruises, the cap has been secured with parafilm after sample collection. When arriving in the laboratory, samples are commonly stored in a cold room or in a refrigerator at 4 ∘C, except when the analysis is expected within 3 months after the arrival of the samples. The analysis has commonly been done within 1 year to 18 months after collection, and for some subsets such as for SURATLANT (Reverdin et al., 2018a) the analysis was usually done within 3 months after collection. However, due to various changes at LOCEAN, there has been a long backlog at times, with some samples having been stored in the cold room for 5 years or more. The longest storage time was for OISO-18 data collected in 2010 and analyzed 9 years later in 2019. Storage time was also very long for most samples from cruises OISO-21, OISO-22, OISO-23, OISO-25, and OISO-26 (southern Indian Ocean, 2012 to 2016). Before analysis, samples are checked for obvious signs of evaporation, such as low water level or salt crystals around the bottle's neck.

We tested whether the samples in common cap bottles change during storage by aging three reference waters of the same deep equatorial Atlantic origin over 2 years in a laboratory room which is not air-conditioned and without securing the common caps with parafilm. Water is extracted every 3 months for isotopic analysis, which so far over 23 months has not revealed any significant drift, certainly not larger than 0.02 ‰ in δ18O and 0.1 ‰ in δ D. We expect that drifts would be even smaller when samples are stored at 4 ∘C or with parafilm if the caps are properly tightened.

In 2019, new caps were introduced which were not rigid and would often not provide a tight seal, with very large sample evolution over less than a year, sometimes reaching close to 1 ‰ in δ18O. This was the case in particular for the samples collected on M/V Nuka Arctica in April 2019, resulting in 32 % of samples with suspected water vapor exchange (indicated by unexpected low d-excess and high δ18O; we verified this hypothesis by aging water in bottles with this cap, which also showed large drifts after 3 months at room temperature).

Even for bottles with the common caps, issues of poor conservation have been suspected in some cases, in particular after long storage (typically, for 5 years or more). There is also the possibility that water vapor exchange has happened during transport, in particular when the samples have experienced very high temperatures, for instance for cruises ending in tropical ports and with long storage times in containers. This issue was probably the case for samples from the EUREC4A-OA cruise collected in February 2020 (Stevens et al., 2021) with an almost 2-month storage period in a container placed without sun shielding in Pointe-à-Pitre (Guadeloupe, France), for which close to 22 % of the bottles with no parafilm securing the cap are suspected to have signs of evaporation (during analysis, we noticed that the cap was often not tightly closed; their isotopic values also contrasted with the ones from special tightly closed nutrient vials pasteurized at 80 ∘C for 40 min after collection that did not present any anomalous d-excess). There are also other subsets with data presenting obvious evaporation issues. The extreme case is for samples collected on M/V Nuka Arctica in 2018 and 2019, for which we suspect evaporation for 20 % of the water samples. In this case, the water was transferred from salinity bottles during the salinity analysis to be stored in bottles with the common cap, where they stayed for close to 18 months before analysis.

The last large source of uncertainty is the value (on the VSMOW scale) attributed to the internal standards used. On the Isoprime IRMS, most internal standards were extracted from different batches of Eau de Paris (EDP) stored in a tank covered with paraffin, whereas since 2012, three internal standards have been regularly extracted from metal tanks in which they are kept for up to 5–6 years with a slight overpressure of dry air (following Gröning, 2018, TEL Technical Note No. 03). The internal standards have been calibrated using VSMOW and GISP (or GRESP), usually more than once, and some were also sent to other laboratories at different times to independently evaluate their characteristics. Comparisons were done in 2013 and 2014 for three internal LOCEAN standards with six laboratories for δ18O and four laboratories for δD, which, taken together, did not reveal an average bias larger than 0.01 ‰ for δ18O or 0.10 ‰ for δD. However, there seem to be differences for the individual standards (Table 1), with the one at -3.26 ‰ for δ18O and -21.32 ‰ for δD presenting an average positive difference of +0.029 ‰ for δ18O and +0.19 ‰ for δD, whereas the other two present a negative difference (i.e., LOCEAN standards seemed too low) smaller than or equal to -0.01 ‰ for δ18O and -0.19 ‰ for δD.

The next round of comparisons of the LOCEAN internal standards took place between 2019 and 2021 with five other European laboratories, and for two of them, there were two different setups for δ18O (most of those with IRMS, except for one with a PICARRO L2130-i CRDS). Thus, this includes seven comparisons for δ18O and five for δD. This set of comparisons (Table 1) was done for the three internal standards used in 2019–2021 and presents a large scatter between the different laboratories, with standard deviation on the order of 0.055 ‰ in δ18O and 0.7 ‰ in δD. As the differences between laboratories are similar for the three internal standards, this comparison suggests some systematic differences between laboratories. However, the large scatter implies that the average differences found are very uncertain. The differences in δ18O and δD found for the three internal standards used in 2019–2021 range between +0.029 ‰ in δ18O and +0.21 ‰ in δD for the less enriched standard to -0.010 ‰ in δ18O and +0.02 ‰ in δD for the most enriched one, respectively (Table 1). This comparison might indicate that we have a positive bias for two of our recent internal standards, which would also produce a small difference in the response slopes of the Picarro CRDS adopted since 2020. A set of four calibration runs done in November 2021 at LOCEAN using new VSMOW, GRESP, and three USGS standards with intermediate values confirmed a positive bias in the most negative internal standard (MIX2). This run, however, did not confirm the average biases in the other internal standards at LOCEAN suggested by Table 1, nor any major slope error. Therefore, the correction of a systematic bias has only been applied to the MIX2 value for analyses since August 2020. For some internal standards, we witnessed larger differences for measurements done in June 2020 after the L2130-i just returned from a cruise with long shipping and storage for more than 9 months. We assume that this anomaly is instrumental and did not last for a long time, as the anomaly was not reproduced during later tests in August 2020 or in November 2021.

The two storage methods used successively for internal standard waters were designed to minimize water vapor exchange. It is, however, possible that small isotopic drifts of the internal standards have taken place with time due to evaporation or possible oxidation of the tanks (rust was found in one nearly empty tank). As mentioned, based on different comparisons over time, sometimes on remnants of the tank waters, we could verify that these drifts have remained smaller than 0.02 ‰ in δ18O and 0.1 ‰ in δD. Finally, standards for the daily runs are temporarily stored for up to a month in glass bottles at 4 ∘C, which are briefly opened every day to extract water. Through its storage life, this water will slightly exchange with the outside air that penetrates when the bottle is briefly opened. Back-of-the-envelope estimates suggest that the effect should be less than 0.01 ‰ in δ18O and 0.05 ‰ in δD, even after a month.

Both oxygen and hydrogen isotope compositions are reported in parts per thousand (‰) on the VSMOW scale. One issue is that we analyze saline samples, while the internal standards are freshwater standards, and the method of analysis has changed over time. There is still a large uncertainty in the correction to be applied to account for the effect of salt on IRMS and CRDS seawater analyses. Here we have applied the corrections provided by Benetti et al. (2017c). Note that in some instances IRMS and CRDS analyses of the same seawater samples may yield similar values. For example, Walker et al. (2016) found very close δ18O values in unadjusted measurements of seawater samples from the same water mass done on different IRMS and CRDS instruments. We have adjusted LOCEAN CRDS and IRMS data on the concentration scale based on the study of Benetti et al. (2017c) as well as on complementary tests with the different wire meshes used more recently and between duplicated IRMS–CRDS samples. The values we report are thus internally consistent but could present differences with datasets processed in other institutions without this proposed adjustment or with other changes of scale of up to 0.10 ‰ in δ18O and 0.20 ‰ in δD, as indicated in Benetti et al. (2017c). We thus expect that adjusted LOCEAN CRDS δ18O data would be higher (more enriched in heavy isotopes) than these other CRDS and more common IRMS data.

In regions where there is enough information in the LOCEAN dataset to establish an average relationship between d-excess and salinity (Benetti et al., 2017b), a large breathing of a sample during storage can be detected using its d-excess value, which is then too low compared to the expected relationship. This was recently checked on a set of 10 water samples originating from salinity bottles collected in the surface North Atlantic in 2021 on MV Tukuma Arctica that did not have the usual plastic insert and thus had evaporated as witnessed by the comparison of salinity with thermosalinograph records. These samples indeed present higher practical salinity (S), d-excess lower than expected, and δ18O and δD higher than the expected values, estimated by average linear fits of d-excess versus salinity and δ18O versus S for this region. The average values of the deviations are ΔS=+0.29, Δδ18O =+0.15 ‰, ΔδD=+0.33 ‰, and Δd-excess =-0.82 ‰. The deviations from these expected values present a loose relationship with the deviation in δ18O (Δδ18O) on the order of -20 % of the deviation of d-excess (Δd-excess) (Fig. 2). This relationship is close to the one used by Benetti et al. (2017b) based on other data in the Labrador Sea, where Δδ18O =-1/7 Δd-excess, ΔδD=+2Δδ18O, and Δd-excess =+0.34ΔS. On the other hand, the correlation between Δd-excess and ΔS is not significantly different from 0, which might be caused by uncertainties in sampling time causing errors in estimating salinity deviation.

In cases when breathing was not too large (resulting in an increase of less than +0.11 ‰ in δ18O), we used the deviation from the expected d-excess relationship to S to estimate an adjusted δ18O and δD (Benetti et al., 2017b). When this method is used, δ18O and δD data are flagged to “probably good” and d-excess to probably bad, as these data are certainly not as accurate as the data with no “correction”, with the adjustment adding an uncertainty on the order of 0.05 ‰ in (δ18O and 0.10 ‰ in δD). For larger suspected evaporation, δ18O and δD data are not adjusted and flagged as “probably bad”. Altogether, we have flagged 12.3 % of the CRDS-measured samples, most of which (11.3 %) correspond to data with anomalously low d-excess and thus suspected evaporation. There is of course also the possibility that for some samples (for 1 % of the samples), d-excess that is too low or too high might just result from an occasional large uncertainty in the analysis.

We recently tested the effectiveness of applying this adjustment for 32 samples collected during cruise OVIDE2018 (North Atlantic Ocean in 2018; Lherminer, 2018) which were stored in two sets of bottles. One set of bottles was analyzed by CRDS at LOCEAN and the other set by IRMS for δ18O at Geozentrum Erlangen. Among the 32 LOCEAN samples, 11 show indications of breathing and have been slightly adjusted based on their negative d-excess deviation. The comparison between the 32 LOCEAN and Geozentrum Erlangen isotopic values suggests that the adjustment we applied to 11 of the LOCEAN data results in decreasing the standard deviation of the δ18O differences between the two sets from 0.060 ‰ to 0.041 ‰. The adjustment of the 11 LOCEAN samples also decreased the standard deviation in the differences between d-excess and d-excess estimated from the d-excess versus S relationship derived for the entire LOCEAN dataset from 0.25 ‰ to 0.15 ‰. As a comparison, when the set is restricted to the 21 non-adjusted LOCEAN samples, the corresponding standard deviations for the δ18O differences between LOCEAN and Geozentrum Erlangen values and the d-excess differences to the expected d-excess versus S relationship were 0.043 and 0.14 ‰, respectively. These values are very close to what is found for the set of 32 LOCEAN samples including the 11 adjusted samples, suggesting some homogeneity in the adjusted dataset.

For earlier IRMS analyses at LOCEAN, we base the identification of possible evaporated data on excessive scatter in the δ18O versus S scatter plots or between successive data compared to what we have previously measured in regions with repeated cruises, and outliers (6 %) are flagged as probably bad. The smaller (by half) proportion of flagged IRMS analyses than for the CRDS analyses suggests either that this validation missed some evaporated IRMS samples, that these earlier data had evaporated less than the more recent ones (some were analyzed sooner after collection), or that the IRMS runs had smaller uncertainties than the latter CRDS runs.

Figure 3 presents the spatial distribution of the LOCEAN-analyzed data close to the surface, with the largest data collection being in the North Atlantic (Fig. 3a) (in particular, with OVIDE cruises since 2002 and the SURATLANT ship of opportunity dataset since 2011), the tropical Atlantic (in particular, the EGEE and PIRATA cruises since 2005), and the southern Indian Ocean (Fig. 3b) (OISO cruises since 1998).

Table 3 reports the number of valid data points by depth range, which indicates that the emphasis in this set has been on near-surface data (58 % of the δ18O data above 40 m depth, 13 % between 40 and 200 m depth, and only 12 % at 1000 m or deeper). There is less valid δD than δ18O data, the difference corresponding to IRMS-measured data, which correspond to 25 % of the total number of water samples in the database. There is even less valid d-excess than δD (by 10 %), the difference corresponding to samples for which an adjustment for slight evaporation was done on δ18O and δD data. The database contains fewer deep samples since the transition to CRDS than before because of a recent emphasis on sampling the upper ocean.

We illustrate the dataset with time series of June (or July) data between 50 and 55∘ N in the eastern North Atlantic subpolar gyre (NASPG) collected mostly during the OVIDE cruises (Fig. 4). This scatter plot of cruise-averaged S and δ18O indicates a near alignment of the values. It is striking that the strongest negative (fresher and lighter) anomalies in 2016 fit rather well on the regression line (in red) for water samples in the southwestern NASPG. This regression line is derived from data from the 47–55∘ N, 30–49∘ W region, excluding very low-salinity data from seasonal sea ice melt or from shelf waters, and is very similar to the distribution in Frew et al. (2000). Thus, this reinforces the hypothesis of Holliday et al. (2020) that the strong freshening present in the eastern subpolar gyre in 2016 originated from the transport of Arctic fresh water from the western boundary current into the eastern basins and not from local rainfall, which would have likely resulted in higher δ18O at the same “low” salinity such as depicted by the black line (Frew et al., 2000; Camille Risi, personal communication, 2021).

The SURATLANT surveys provided a seasonal sampling of water isotopes between late 2011 and 2019 along the western flank of the Reykjanes Ridge in the central part of the gyre (53–56∘ N, 38–44∘ W). Annual summaries of these data are provided in Fig. 5a. There is less alignment of the interannual values on the average southwestern NASPG linear regression line than for the OVIDE surveys (Fig. 4). However, there is some aliasing of the seasonal cycle in the annual averages (see Reverdin et al., 2018a), which contributes to the scatter, as well as noise in the data and natural variability. On this plot the freshest year appears to be 2017, in agreement with an analysis using a much more complete salinity dataset (Reverdin et al., 2018b). 2017 is also one of the lighter δ18O years. The corresponding d-excess versus S diagram (Fig. 5b) presents yearly anomalies that are fairly aligned with the average regression between southwestern NASPG d-excess and salinity data. Error bars are large, but nevertheless, low-salinity waters exhibit high d-excess, as described in Benetti et al. (2017a, b).