GLODAPv2.2020 – the second update of GLODAPv2

GLODAPv2.2020 – the second update of GLODAPv2 Are Olsen , Nico Lange , Robert M. Key , Toste Tanhua , Henry C. Bittig, Alex Kozyr, Marta Álvarez , Kumiko Azetsu-Scott, Susan Becker , Peter J. Brown, Brendan R. Carter , Leticia Cotrim da Cunha , Richard A. Feely , Steven van Heuven , Mario Hoppema , Masao Ishii , Emil Jeansson , Sara Jutterström , Camilla S. Landa , Siv K. Lauvset , Patrick Michaelis, 5 Akihiko Murata , Fiz F. Pérez , Benjamin Pfeil , Carsten Schirnick , Reiner Steinfeldt , Toru Suzuki , Bronte Tilbrook , Anton Velo , Rik Wanninkhof , Ryan J. Woosley 23 1 Geophysical Institute, University of Bergen and Bjerknes Centre for Climate Research, Bergen, Norway 2 GEOMAR Helmholtz Centre for Ocean Research Kiel, Kiel, Germany 3 Atmospheric and Oceanic Sciences, Princeton University, Princeton, NJ, 08540, USA 10 4 Leibniz Institute for Baltic Sea Research Warnemünde, Rostock, Germany 5 NOAA National Centers for Environmental Information, Silver Spring, MD, USA 6 Instituto Español de Oceanografía, A Coruña, Spain 7 Departement of Fisheries and Oceans, Bedford Institute of Oceanography, Dartmouth, Nova Scotia, Canada 8 UC San Diego, Scripps Institution of Oceanography, San Diego CA 92093, USA 15 9 National Oceanography Centre, Southampton, UK 10 Joint Institute for the Study of the Atmosphere and Ocean, University Washington, Seattle, Washington, USA 11 Pacific Marine Environmental Laboratory, National Oceanic and Atmospheric Administration, Seattle, Washington, USA 12 Faculdade de Oceanografia, Universidade do Estado do Rio de Janeiro, Rio de Janeiro (RJ), Brazil 20 13 Centre for Isotope Research, Faculty of Science and Engineering, University of Groningen, Groningen, the Netherlands 14 Alfred Wegener Institute Helmholtz Centre for Polar and Marine Research, Bremerhaven, Germany 15 Oceanography and Geochemistry Research Department, Meteorological Research Institute, Japan Meteorological Agency, Tsukuba, Japan 16 NORCE Norwegian Research Centre, Bjerknes Centre for Climate Research, Bergen, Norway 25 17 IVL Swedish Environmental Research Institute, Gothenburg, Sweden 17 Research Institute for Global Change, Japan Agency for Marine-Earth Science and Technology, Yokosuka, Japan 18 Instituto de Investigaciones Marinas, IIM – CSIC, Vigo, Spain 19 University of Bremen, Institute of Environmental Physics, Bremen, Germany 20 Marine Information Research Center, Japan Hydrographic Association, Tokyo, Japan 30 21 CSIRO Oceans and Atmosphere and Antarctic Climate and Ecosystems Co-operative Research Centre, University of Tasmania, Hobart, Australia 22 Atlantic Oceanographic and Meteorological Laboratory, National Oceanic and Atmospheric Administration, Miami, USA. 23 Center for Global Change Science, Massachusetts Institute for Technology, Cambridge, Massachusetts, USA 35


Introduction
The oceans mitigate climate change by absorbing atmospheric CO 2 corresponding to a significant fraction of anthropogenic CO 2 emissions (Friedlingstein et al., 2019;Gruber et al., 2019) and most of the excess heat in the Earth System caused by the enhanced greenhouse effect (Cheng et al., 2020;. The objective of GLODAP (Global Ocean Data Analysis Project, www.glodap.info, last access: 25 May 2020) is to ensure provision of high quality 130 and bias-corrected water column bottle data from the ocean surface to bottom that document the state and the evolving changes in physical and chemical ocean properties, e.g., the inventory of the excess CO 2 in the ocean, natural oceanic carbon, ocean acidification, ventilation rates, oxygen levels, and vertical nutrient transports. The GLODAP core variables, which are quality controlled and bias corrected, are salinity, dissolved oxygen, inorganic macronutrients (nitrate, silicate, and phosphate), seawater CO 2 chemistry variables (dissolved inorganic carbon -TCO 2 , total alkalinity -135 TAlk, and pH on the total H + scale), and the halogenated transient tracers and CCl 4 .
Other chemical tracers are usually also measured on the cruises included in GLODAP. A subset of these data is distributed as part of the product but has not been extensively quality controlled or checked for measurement biases in this effort. For some of these variables, better sources of data may exist, for example the product by Jenkins et al. (2019) 5 The individual cruise data files were converted to WOCE exchange format: a comma delimited ASCII format for CTD and bottle data from hydrographic cruises. GLODAP deals only with bottle data and CTD data at bottle trip depths, and 450 their exchange format is briefly reviewed here with full details provided in Swift and Diggs (2008). The first line of each exchange file specifies the data type, in the case of GLODAP this is "BOTTLE", followed by a date and time stamp and identification of the person/group who prepared the file, e.g., "PRINUNIVRMK" is Princeton University, Robert M. Key.
Next follows the README section. This provides brief cruise specific information, such as dates, ship, region, method and quality notes for each variable measured, citation information, and references to any papers that used or presented the 455 data. The README information was typically assembled from the information contained in the metadata submitted by the data originator. In some cases, issues noted during the primary QC and other information such as file update notes are included. The only rule for the README section is that it be concise and informative. The README is followed by data column headers, units, and then the data. The headers and units are standardized and provided in Table 1 for the variables included. Exchange file preparation entailed units conversion in some cases, most frequently from milliliters per liter (mL 460 L -1 ; oxygen) or micromoles per liter (µmol L -1 ; nutrients) to micromoles per kilogram of seawater (µmol kg -1 ). The default procedure for nutrients was to use seawater density at reported salinity, an assumed measurement-temperature of 22 ºC, and pressure of 1 atm. For oxygen, the factor 44.66 was used for the milliliter to micromole conversion, while for the per liter to per kilogram conversion density based on reported salinity and draw temperatures was preferred, but draw temperature was frequently not reported and potential density was used instead. The potential errors introduced by any of 465 these procedures are insignificant. Missing numbers are indicated by -999, with trailing zeros to comply with the number format for the variable in question, as specified in Swift and Diggs (2008).
Each data column (except temperature and pressure, which are assumed "good" if they exist) has an associated column of data flags. For the exchange files, these flags conform to the WOCE definitions for water sample bottles and are listed in Table 2. If no such WOCE flags were submitted with the data, they were assigned by us. In any case, incoming files were 470 subjected to primary QC to detect questionable or bad data. This was carried out following  and , primarily by inspecting property-property plots. Outliers showing up in two or more different such plots were generally defined as questionable and flagged as such. In some cases, outliers were detected during the secondary QC; the consequential flag changes have then also been applied in the original cruise data files.

Secondary quality control 475
The aim of the secondary QC was to identify and correct any significant biases in the data from the 106 new cruises relative to GLODAPv2.2019, while retaining any signal due to temporal changes. To this end, secondary QC in the form of consistency analyses was conducted to identify offsets in the data. All identified offsets were scrutinized by the GLODAP reference group through a series of teleconferences during March and April 2020 in order to decide the adjustments to be applied to correct for the offset (if any). To guide this process, a set of initial minimum adjustment 480 limits was used (Table 3). These are set according to the expected measurement precision for each variable, and are the same as those used for GLODAPv2.2019. In addition to the magnitude of the offset, factors such as its precision, persistence towards the various cruises used in the comparison, regional dynamics, and the occurrence of time trends or other variations were considered. Thus, not all offsets larger than the initial minimum limits have been adjusted for. A guiding principle for these considerations was to not apply an adjustment whenever in doubt. Conversly, in some cases 485 where data and offsets were very precise and the cruise had been conducted in a region where variability is expected to be small, adjustments lower than the minimum limits were applied. Any adjustment was applied uniformly to all values for a 6 variable and cruise, i.e., an underlying assumption is that cruises suffer from either no or a single and constant measurement bias. Except where explicitly noted (Sect. 3.3.1), adjustments were not changed for data previously included 520 in GLODAPv2.2019.
Crossover comparisons, multi-linear regressions (MLRs), and comparison of deep-water averages were used to identify offsets for salinity, oxygen, nutrients, TCO 2 , TAlk and pH (Sect. 3.2.2 and 3.2.3). In contrast to GLODAPv2 and GLODAPv2.2019, evaluation of the internal consistency of the seawater CO 2 chemistry variables was not used for the evaluation of pH (Sect. 3.2.4). New to the present version is the more extensive use of  predictions for the evaluation of offsets in nutrients and seawater CO 2 chemistry data (Section 3.2.5). For the halogenated transient tracers, examination of surface saturation levels and the relationship among the tracers were used to assess the data consistency (Sect. 3.2.6). For salinity and oxygen, CTD and bottle values were merged into a "hybrid" variable prior to the consistency analyses (Sect. 3.2.1).

Merging of sensor and bottle data 530
Salinity and oxygen data can be obtained either by analysis of water samples (bottle data) and/or directly from the CTD sensor pack. These two types are merged and presented as a single variable in the product. The merging was conducted prior to the consistency checks, ensuring their internal calibration in the product. The merging procedures were only applied to the bottle data files, which commonly include values recorded by the CTD at the pressures of the upcast when the water samples are collected. Whenever both CTD and bottle data were present in a data file, the merging step 535 considered the deviation between the two and calibrated the CTD values if required and possible. Altogether seven scenarios are possible, where the fourth (see below) never occurred during our analyses, but is included to maintain consistency with GLODAPv2: 1. No data are available: no action needed. The number of cases encountered for each scenario is summarized in Sect. 4.1.

Crossover analyses
The crossover analyses were conducted with the MATLAB toolbox prepared by Lauvset and Tanhua (2015) and with the 550 GLODAPv2.2019 data product as reference. In areas where a strong trend in salinity was present, the TAlk and TCO 2 data were salinity normalized before crossover analysis, following Friis et al. (2003).
The toolbox implements the 'running-cluster' crossover analysis first described by . This analysis compares data from two cruises on a station-by-station basis and calculates a weighted mean offset between the two and its weighted standard deviation. The weighting is based on the scatter in the data such that data that have less scatter have 555 larger influence on the comparison than data with more scatter. Whether the scatter reflects actual variability or data 7 precision is irrelevant in this context as increased scatter regardless decreases the confidence in the comparison. Stations that are compared must be within 2° arc distance (~ 200 km) of each other, and only deep data are used. This minimizes effects of natural variability. As default, we used 1500 dbar as the upper depth limit, but in regions where deep mixing or convection occurs (such as the Nordic, Labrador, and Irminger seas) a more conservative limit of 2000 dbar was applied.
The deeper limit was also applied to the majority of the northern Pacific cruises on the RV Keifu Maru II and RV Ryofu 585 Maru III due the great abundance of deep data of the new-and reference cruises. As an example, the crossover for TCO 2 measured on the two cruises 49UP20160109 and 49UP20160703 is shown in Fig. 3. For TCO 2 the offset is determined as the difference. This is also the case for salinity, TAlk, and pH. For the nutrients, oxygen, and the halogenated transient tracers, ratios are used. This in accordance with the procedures followed for GLODAPv2. The TCO 2 values from 49UP20160109 are higher, with a weighed mean offset of 3.62 ± 2.67 µmol kg -1 compared to those measured at 590 49UP20160703.
For each of the 106 new cruises, such a crossover comparison was conducted against all cruises possible in GLODAPv2.2019, i.e., all cruises that had stations closer than 2° arc distance to any station for the cruise in question.
The summary figure for TCO 2 at 49UP20160109 is shown in Fig. 4. The TCO 2 data measured at this cruise are high when compared to the data measured at all nearby cruises included in GLODAPv2.2019, by 3.68 ± 0.83 µmol kg -1 . This is 595 slightly less than the initial minimum adjustment limit for TCO 2 of 4 µmol kg -1 (Table 3), but the offset is present against all cruises and there is no obvious time trend (particularly important for TCO 2 ), and as such qualifies for an adjustment of the data in the merged data product. In this case -3 µmol kg -1 was applied, in order to bring the TCO 2 data from 49UP20160109 into consistency with GLODAPv2.2019.
Two exceptions to the above-described procedure exist: In the Japanese Sea six new cruises were added. In this region, 600 there are only data from two cruises in GLODAPv2.2019. Therefore, all eight cruises were compared against each other and strong outliers were adjusted accordingly, instead of adjusting the six new cruises towards the two existing. A similar approach was used for the 10 new Davis Strait cruises; in this region no data were available in GLODAPv2.2019. Due to the complex hydrography and differences in sampling locations it was very problematic to fully quality control these data, however, so most have been labeled -888, i.e., they are included in the product but with a secondary QC flag of 0 (Sect. 605 6).

Other consistency analyses
A few new cruises had no or very few valid crossovers with GLODAPv2 data. In that situation two other consistency analyses were carried out for salinity, oxygen, nutrients, TCO 2 , and TAlk data, namely MLR analyses and deep water averages, broadly following Jutterström et al. (2010). For the MLRs, the presence of bias in the data for the cruise in 610 question was identified by comparing the MLR generated with the measured values. These methods were useful in the data-sparse Arctic and Southern oceans. Both analyses were conducted on samples collected deeper than the 1500 or 2000 dbar pressure level to minimize the effects of natural variations, and both used available GLODAPv2.2019 data from within 2° of the cruise in question to generate the MLR or deep water average. The lower depth limit was set to the deepest sample for the cruise in question. For the MLRs, all of the above mentioned variables could be included among 615 the independent variables (e.g., for a TAlk MLR, salinity, oxygen, nutrients, and TCO 2 were allowed), with the exact selection determined based on the statistical robustness of the fit, as evaluated using the coefficient of determination (r 2 ) and root mean square error (RMSE). MLRs based on variables that were suspect for the cruise in question were avoided (e.g., if oxygen appeared biased it was not included as an independent variable). The MLRs could be based on 10 to 500 Are Olsen 31/7/2020 11:42 Deleted: Typically…s default, we used 1500 dbar ... [35] 8 samples, and the robustness of the fit (r 2 , RMSE) and quantity of fitting data were considered when using the results to 655 guide whether to apply a correction. The same applies for the deep-water averages (i.e., the standard deviation of the mean). MLR and deep-water average results showing offsets above the minimum adjustment limits were carefully scrutinized, along with any crossover and CANYON-B and CONTENT results that existed, to determine whether or not to apply an adjustment.

pH scale conversion and quality control 660
Altogether 82 of the 106 new cruises included pH data. For one of these, the pH data were not supplied on the total scale or at 25 °C and 0 dbar pressure, which is the GLODAP standard, and were thus converted. The conversion was conducted using CO2SYS  for MATLAB  with reported pH and TAlk as inputs, and generating pH output values at total scale at 25 °C and 0 dbar of pressure (named phts25p0 in the product). Missing TAlk data were approximated as 67 times salinity. The proportionality (67) is the mean ratio of TAlk to salinity in 665 GLODAPv2 data. This is sufficiently accurate for scale-temperature-pressure conversions. Data for phosphate and silicate are also needed, and were, whenever missing, determined using CANYON-B . The conversion was conducted with the carbonate dissociation constants of , the bisulfate dissociation constant of Dickson (1990), and the borate-to-salinity ratio of . These procedures are the same as used for GLODAPv2.2019 (Olsen et al., 2019) 670 Internal consistency of CO 2 system variables were not used for the secondary quality control of the pH data of the 106 new cruises, but only crossover analysis supplemented by CONTENT and CANYON-B (Sect. 3.2.5). This avoids uncertainties in the quality control owing to incomplete understanding of the thermodynamic constants, major ion concentrations, measurement biases, and potential contribution of organic compounds to alkalinity (Álvarez et al., 2020;Takeshita et al., 2020). However, this applies only to the new cruises. The pH data of 840 of the 936 cruises in 675 GLODAPv2.2020 were QC'd for GLODAPv2 and GLODAPv2.2019, and for these earlier products internal consistency of CO 2 system was used for secondary QC of pH. Therefore the level of consistency between these 936 cruises remains at 0.01 to 0.02 pH units, as more thoroughly discussed in

CANYON-B and CONTENT analyses
CANYON-B and CONTENT  were used to support decisions regarding application of adjustments (or 680 not) from the analyses described above. CANYON-B is a neural network for estimating nutrients and seawater CO 2 chemistry variables from temperature, salinity, and oxygen. CONTENT additionally considers the consistency among the estimated CO 2 chemistry variables to further refine them. These approaches were developed using the data included in the GLODAPv2 data product. Their advantage compared to crossover analyses for evaluating consistency among cruise data is, that effects of water mass changes on ocean properties are represented in the non-linear relationships in the underlying 685 neural network. For example, if elevated nutrient values are measured on a cruise but are not due to a measurement bias but actual aging of the water mass(es) that have been sampled and as such accompanied by a decrease in oxygen concentrations, the measured values and the CANYON-B estimates will be similar. Vice-versa, if the nutrient values are biased, the measured values and CANYON-B predictions will be dissimilar. Of course, we kept in mind that this relies on the accuracies of the T, S and O 2 data and of CANYON-B and CONTENT in themselves. Used in the correct way and 690 with caution this tool is a powerful supplement to the traditional crossover analyses. As an example, the CANYON-B/CONTENT analyses of the data obtained at 49UP20160109 are presented in Fig. 5 Deleted: 77 of the 116 new cruises included pH data. For about 30 % of these, the pH data were not 695 supplied on the total scale, and at 25°C and 0 dbar pressure, which is the GLODAP standard. These data were converted to total pH scale and temperature and pressure of 25°C and 0 dbar. The conversions were conducted by using CO2SYS 700  for MATLAB  with reported pH and TAlk as inputs, and generating pH output values at total scale at 25°C and 0 dbar of pressure (named phts25p0 in the product). Whenever TAlk data were missing, 705 these values were approximated as 67 times salinity. The proportionality (67) is the mean ratio of TAlk to salinity in the GLODAPv2 data. This is sufficiently accurate for scale-temperature-pressure conversions. Data for phosphate and silicate are also needed, and 710 were, whenever missing, determined using CANYON-B . The conversion was conducted with the carbonate dissociation constants of , the bisulfate dissociation constant of Dickson (1990), and the 715 borate-to-salinity ratio of . These procedures are the same as used for GLODAPv2 , except for the CANYON-B estimation of phosphate and silicate.
... [39] results confirmed the positive offset in the TCO 2 values revealed in the crossover comparisons discussed in Sect. 3.2.2.
The magnitude of the inconsistencies for the CANYON-B estimate was 3.4 µmol kg -1 , i.e., slightly less than that the weighted mean crossover offset of 3.7 µmol kg -1 , while the CONTENT estimate gave an inconsistency of 2.7 µmol kg -1 .
The differences between these consistency estimates owes to differences in the actual approach, the weighting across stations, stations considered (i.e., crossover comparisons use only stations within ~200 km of each other, while 725 CANYON-B and CONTENT considers all stations where necessary variables are sampled, and depth range considered (> 500 dbar for CANYON-B and CONTENT vs. >1500/2000 dbar for crossovers). The specific difference between the CANYON-B and CONTENT estimates is a result of the seawater CO 2 chemistry considerations by the latter. For the other variables, the inconsistencies are low and agree with the crossover results (not shown here but results can be accessed through the Adjustment Table) with the exception of pH. The pH results are further discussed in Sect. 4.2. 730 Another advantage of CANYON-B and CONTENT is that by considering the each data point in it self, primary QC issues has been revealed and corrected for some of the cruises.

Halogenated transient tracers
For the halogenated transient tracers and CCl 4 ; CFCs for short) inspection of surface saturation levels and evaluation of relationships between the tracers for each cruise were used to identify biases, rather 735 than crossover analyses. Crossover analysis is of limited value for these variables given their transient nature and low deep-water concentrations. As for GLODAPv2, the procedures were the same as those applied for CARINA Steinfeldt et al., 2010).

Merged product generation
The merged product file for GLODAPv2.2020 was created by correcting known issues in the GLODAPv2.2019 merged 740 file, and then appending a merged and bias-corrected file containing the 106 new cruises to this error-corrected GLODAPv2.2019 file.

Updates and corrections for GLODAPv2.2019
Several minor omissions and errors have been identified in the GLODAPv2 and v2.2019 data products since their release in 2016 and 2019, respectively. Most of these have been corrected in this release. In addition, some recently available 745 data have been added for a few cruises. The changes are: − For cruise 33RR20160208, the CFC-113 data of station 31 were found to be bad and have been removed.
Additionally, the flags for CFC-11, CFC-12, SF 6 and CCl 4 were replaced with new ones received from the Principal Investigator, and recently published data for δ 13 C and Δ 14 C have been added to the product file.
− For 18HU20150504, the pH data measured at stations 196, 200, and 203 were found offset by approximately +0.1 750 units, because such large offset points to general data quality problems, these data have been removed.
− For 32PO20130829, pH values of station 133 cast 1 were in the wrong order in the file. This has been amended.
Additionally, pH values from cast 2 at this station were deemed questionable and have been removed.
− For 33RR20050109, the δ 13 C values of station 7 bottle 32 and station 16 bottle 22 were found bad (values were less than -6 ‰) and have been removed from the product file. 755 − For 35MF19850224, the δ 13 C value of station 21 cast 3 bottle 4 was found bad and has been removed.
− For 74JC20100319 the δ 13 C value at station 37 bottle 7 was found bad and has been removed.
Are Olsen 31/7/2020 11:42 Formatted: Right: -0 cm, Tabs: 17.25 cm, Left − All δ 13 C values from the large volume Gerard barrels (identified by bottle number greater than 80) were removed from the product files as these often have poor precision and accuracy related to gas extraction procedures.
− For 33HQ20150809, temperatures of station 52 cast 1 were found bad (less than -2 °C) and have been removed, hence all other samples were removed for this cast as well (the same depths and variables were sampled at the 770 other casts, however). Temperatures for casts 2 and 8 were replaced with updated values; these changes are very minor, on the order of 0.001 o C.
− For cruises 33RO20110926, 33RO20150525, and 33RO20150410, δ 13 C and Δ 14 C data have become available and added to the product. − Discrete fugacity of CO 2 (fCO 2 ) data are now included in the product files whenever available. Discrete fCO 2 is 780 one of the four variables that describes seawater CO 2 chemistry, but is rarely measured and has not been included in GLODAP product files before, in particular as a result of apparent quality issues that were not fully understood during the secondary QC for GLODAPv1.1 . However, for some cruises fCO 2 data were included indirectly in both GLODAPv1.1 and GLODAPv2 as they had been used to calculate TAlk, in combination with TCO 2 . These calculated TAlk values were, however, not included in v2.2019. We have now 785 chosen to include the discrete fCO 2 values in the product files. This increases transparency and traceability of the product; the fCO 2 data are also highly relevant for ongoing efforts toward resolving recently identified inconsistencies in our understanding of the relationships among the four seawater CO 2 chemistry variables Fong and Dickson, 2019;Takeshita et al., 2020;Àlvarez et al., 2020). A total of 33924 discrete fCO 2 measurements from 34 cruises conducted between 1983-2014 are now included. All values were converted to 20 790 °C and 0 dbar pressure using CO2SYS for MATLAB . This was also used for the conversion of partial pressure of CO 2 (pCO 2 ) to fCO 2 for the 20 cruises where pCO 2 was reported. The procedures for these conversions, in terms of dissociation constants and approximation of missing variables, were the same as for the pH conversions (Sect. 3.2.4). These fCO 2 data have not been subjected to secondary QC. The inclusion of discrete fCO 2 data has led to some changes in the calculations of missing seawater CO 2 chemistry variables; these 795 are described towards the end of the next section.

Merging
The new data were merged into a bias-minimized product file following the procedures used for GLODAPv1.1 , CARINA , PACIFICA , GLODAPv2 , and GLODAPv2.2019 , with some modifications: 800 − Data from the 106 new cruises were merged and sorted according to EXPOCODE, station, and pressure.
GLODAP cruise numbers were assigned consecutively, starting from 2001, so they can be distinguished from the GLODAPv2.2019 cruises that ended at 1116. Deleted: The new data were merged into a biasproduct file following the procedures used for 825 GLODAPv1.1 , CARINA , PACIFICA , and GLODAPv2 , but with minor changes: 11 − For some cruises the combined concentration of nitrate and nitrite was reported instead of nitrate. If explicit nitrite concentrations were also given, these were subtracted to get the nitrate values. If not, the combined concentration 835 was renamed to nitrate. As nitrite concentrations are very low in the open ocean, this has no practical implications.
− When bottom depths were not given, they were approximated as the deepest sample pressure +10 dbar or extracted from ETOPO1 (Amante and Eakins, 2009), whichever was greater. For GLODAPv2, bottom depths were extracted from the Terrain Base (National Geophysical Data Center/NESDIS/NOAA/U.S. Department of Commerce, 1995). The intended use of this variable is only drawing approximate bottom topography for sections. 840 − Whenever temperature was missing in the original data file, all data for that record were removed and their flags set to 9. The same was done when both pressure and depth were missing. For all surface samples collected using buckets or similar, the bottle number was set to zero. There are some exceptions to this, in particular for cruises that also used Gerard barrels for sampling. These may have valuable tracer data not accompanied by a temperature, so such data have been retained. 845 − All data with WOCE quality flags 3, 4, 5, or 8 were excluded from the product files and their flags set to 9. Hence, in the product files a flag 9 can indicate not measured (as is also the case for the original exchange formatted data files) or excluded from the product; in any case, no data value appears. All flags 6 (replicate measurement) and 7 (manual chromatographic peak measurement) were set to 2.
− Missing sampling pressures or depths were calculated following UNESCO (1981). 850 − For both oxygen and salinity, CTD and bottle values were merged following procedures summarized in Sect.

3.2.1.
− Missing salinity, oxygen, nitrate, silicate, and phosphate values were vertically interpolated whenever practical, using a quasi-Hermetian piecewise polynomial. "Whenever practical" means that interpolation was limited to the vertical data separation distances given in Table 4 in Key et al. (2010). Interpolated values have been assigned a 855 WOCE quality flag 0.
− The data for the 12 core variables were corrected for bias using the adjustments determined during the secondary QC. For each of these variables the data product also has separate columns of secondary QC flags, indicating by cruise and variable whether ("1") or not ("0") data successfully received secondary QC. A 0 flag here means that data were too shallow or geographically too isolated for consistency analyses or that these analyses were 860 inconclusive, but that we have no reasons to believe that the data in question are of poor quality.
− Missing seawater CO 2 chemistry variables were calculated, whenever possible. The procedures for these calculations have been slightly altered as the product now contains four such variables; earlier versions of GLODAPv2  included only three, so whenever two were included the one to calculate was unequivocal. Four CO 2 chemistry variables gives more degrees of freedom in this respect, e.g., a 870 particular record may have measured data for TCO 2 , TAlk, and pH, and then a choice needs to be made with regard to which pair to use for the calculation of fCO 2 . We followed two simple principles. First, TCO 2 and TAlk was the preferred pair to calculate pH and fCO 2 , because we have higher confidence in the TCO 2 and TAlk data 12 than pH (given the issues summarized in Sect. 3.2.4) and fCO 2 (because it was not subjected to secondary QC).
Second, if either TCO 2 or TAlk was missing and both pH and fCO 2 data existed, pH was preferred (because fCO 2 has not been subjected to secondary QC). All other options involve only two measured variables. The calculations were conducted using CO2SYS  for MATLAB , with the constants set as for the pH conversions (Sect. 3.2.4). For calculations involving TCO 2 , TAlk, and pH, if the 910 number of measured values for a specific cruise were less than half the number of calculated, then all measured values were replaced by calculated values. Such replacements were not done for calculations involving fCO 2 , as this would tend to overwrite all measured fCO 2 values or would entail replacing a measured variable that has been subjected to secondary QC (i.e., TCO 2 , TAlk, or pH) with one calculated from a variable that has not been subjected to secondary QC (i.e., fCO 2 ). Calculated values have been assigned WOCE flag 0. 915 − The resulting merged file for the 106 new cruises was appended to the merged product file for GLODAPv2.2019.

Secondary quality control results and adjustments
All material produced during the secondary QC is available at the online GLODAP Adjustment Table hosted by GEOMAR, Kiel, Germany at https://glodapv2-2020.geomar.de/ (last access: 18 June 2020), and which can also be accessed through www.glodap.info. This is similar in form and function to the GLODAPv2 Adjustment Table (Olsen et 920 al., 2016) and includes a brief written justification for any adjustments applied. Table 4 summarizes the actions taken for the merging of the CTD and bottle data for salinity and oxygen. For 81 % of the 106 cruises added with this update, both CTD and bottle data were included for salinity in the original cruise data files and for all these cruises the two data types were found to be consistent. This is similar to the GLODAPv2.2019 results. 925

Sensor and bottle data merge for salinity and oxygen
For oxygen, only 25 % of the cruises included both CTD O 2 and bottle values; this is much less than for GLODAPv2.2019 where 50 % of the cruises included both. Having both CTD and bottle values in the data files is highly preferred as the information is valuable for quality control (bottle mistrips, leaking niskin bottles, and oxygen sensor drift are among the issues that can be revealed). The extent to which the bottle data (i.e., OXYGEN in the individual cruise exchange files) in reality is mislabeled CTD data (i.e., should be CTDOXY) is uncertain. Regardless, the large majority 930 of the CTD and bottle oxygen were consistent and did not need any further calibration of the CTD values (23 out of 25 cruises), while for two cruises no good fit could be obtained and their CTD O 2 data are not included in the product.

Adjustment summary
The secondary QC actions for the 12 core variables and distribution of adjustments applied are summarized in Table 5 and Fig. 6, respectively. A very small fraction of the data is adjusted for most variables. None of the salinity data are 935 adjusted, for oxygen and nitrate 1% of the data are adjusted, 2 % for TCO 2 , 5 % for TAlk, 7 % for phosphate, and 9 % for silicate. For the CFCs, data from one of 16 cruises with CFC-11 is adjusted, while the fractions are two of 21 for CFC-12, and one out of three for CFC-113. The adjustments for the variables are also fairly small, overall. Thus the tendency observed during the production of GLODAPv2.2019 remains, namely, that the data collected at the large majority of recent cruises are consistent with earlier releases of this product. 940 13 The quality control of pH data proved challenging for this version. The large majority had been collected in the northwestern Pacific, at the cruises conducted by the Japan Meteorological Agency. Figure 7 shows the distribution of pH crossover offsets vs. GLODAPv2.2019. Most of the pH values are higher, some by up to 0.02 units, which is considerable, particularly as the data that are compared are from deeper than 2000 dbar where no changes due to ocean acidification are expected. The challenging aspect lies in the fact that the data that are being added are comparatively 980 many (~ 70 cruises vs. ~ 130 already included in v2.2019) and also are more recent (2010-2018 vs. 1993-2016). As such they might be of higher quality given advances in pH measurement techniques over the years. Adjusting a large fraction of the new cruises down (by the adjustment limit of 0.01) is not advisable. We therefore chose to not adjust any pH data, but to exclude the most serious outliers from the product file (using a limit of |0.015|) and include the rest of the data as is.
This is the reason that the number of adjusted cruises for pH is zero (Table 5). We expect that a crossover and inversion 985 analysis of all pH data in the northwestern Pacific will provide more information on the consistency among the cruises, and such an analysis will be conducted for the next update. This might result in re-inclusion of these data, the formal decision for these are therefore "suspend" (Table 5). For now, some caution should be exercised if looking at trends in ocean pH in that region using these data.
For the nutrients, the adjustments were applied to maintain consistency with data included in GLODAPv2 and 990 GLODAPv2.2019. An alternative goal for the adjustments would be maintaining consistency with data from cruises that employed CRMNS to ensure accuracy of nutrient analyses. Such a strategy was adopted by Aoyama (2020) for preparation of the Global Nutrients Dataset 2013 (GND13), and is being considered for GLODAP as well. However, as it would require a re-evaluation of the entire data set, this will not occur until the next full update of GLODAP, i.e., GLODAPv3. For now, we note the overall agreement between the adjustments applied in these two efforts (Aoyama, 995 2020), and that most disagreements appear to be related to cases where no adjustments were applied in GLODAP. This can be related to the strategy followed for nutrients for GLODAPv2, where data from GO-SHIP lines were considered a priori more accurate than other data. CRMNS are used for nutrients on most GO-SHIP lines.
The improvement in data consistency is evaluated by comparing the weighted mean of the absolute offsets for all crossovers before and after the adjustments have been applied. This "consistency improvement" for core variables is 1000 presented in Table 6. The data for CFCs were omitted for previously discussed reasons (Sect. 3.2.6). Globally, the improvement is modest. Considering the initial data quality, this result was expected, but this does not imply that the data initially were consistent everywhere. Rather, for some regions and variables there are substantial improvements when the adjustments are applied. For example, Arctic Ocean phosphate, Indian Ocean silicate and TCO 2 , and Pacific Ocean pH data all show considerable improvements. 1005 The various iterations of GLODAP provide insight into initial data quality covering more than 4 decades. Figure 8 summarizes the applied absolute adjustment magnitude per decade. These distributions are broadly unchanged compared to GLODAPv2.2019 ( Fig. 6 in  For several variables improvement is evident over time. Most TCO 2 and TAlk data from the 1970s needed an adjustment, but this fraction steadily declines until only a small percentage is adjusted. This is encouraging and demonstrates the value of standardizing sampling and measurement practices (Dickson 1010, the widespread use of CRMs , and instrument automation. The pH adjustment frequency also has a downward trend; however, there remains issues with the pH adjustments and this a topic for future development in GLODAP, with the support from the OCB Carbonate System Intercomparison Forum (CSIF, https://www.us-ocb.org/ocean-carbonate-system-intercomparison-forum/, last accessed: 20 June 2020) working group (Álvarez et al., 2020). For the nutrients and oxygen, only the phosphate adjustment frequency decreases from decade to 1015 Deleted: Ocean silicate for the adjusted data is 11.1 % and that for salinity is 10 ppm (i.e., a salinity of 0.01). This can be ascribed to two cruises, 58GS20130717 and 58GS20160802, conducted in the Greenland Sea where an increasing presence of 1075 Arctic sourced deep waters generates changes in these properties (Blindheim and Rey, 2004;Olafsson and Olsen, 2010;Olsen et al., 2009)  Deleted: , the widespread use of CRMs , and instrument 1070 ... [45] 14 decade. However, we do note that the more recent data, from the 2010s, receive the fewest adjustments. This may reflect 1080 recent increased attention that seawater nutrient measurements have received through an operation manual  availability of CRMNS , and the SCOR working group #147, Towards comparability of global oceanic nutrient data (COMPONUT). For silicate, the fraction of cruises receiving adjustments peaks in the 1990s and 2000s. This is related to the 2 % offset between US and Japanese cruises in the Pacific Ocean that was revealed during production of GLODAPv2 and discussed in . For salinity and 1085 the halogenated transient tracers, the number of adjusted cruises is small in every decade.

Data availability
The GLODAPv2.2020 merged and adjusted data product is archived at NOAA NCEI under https://doi.org/10.25921/2c8h-sa89 (Olsen et al., 2020). These data and ancillary information are also available via our web pages https://www.glodap.info and https://www.nodc.noaa.gov/ocads/oceans/GLODAPv2_2020/ (last access: 22 1090 June 2020). The data are available as comma-separated ascii files (*.csv) and as binary MATLAB files (*.mat). Regional subsets are available for the Arctic, Atlantic, Pacific, and Indian oceans. There are no data overlaps between regional subsets and each cruise exists in only one basin file even if data from that cruise crosses basin boundaries. The station locations in each basin file are shown in Fig. 9. The product file variables are listed in Table 1. A lookup table for matching the EXPOCODE of a cruise with GLODAP cruise number is provided with the data files. In the MATLAB files 1095 this information is also available as a cell array. A "known issues document" accompanies the data files and provides an overview of known errors and omissions in the data product files. It is regularly updated, and users are encouraged to inform us whenever any new issues are identified. It is critical that users consult this document whenever the data products are used.
The original cruise files are available through the GLODAPv2.2020 cruise summary table (CST) hosted by NOAA 1100 NCEI: https://www.nodc.noaa.gov/ocads/oceans/GLODAPv2_2020/ (Last access: 22 June 2020). Each of these files has been assigned a doi, but these are not listed here. The CST also provides brief information on each cruise and access to metadata, cruise reports, and its Adjustment Table entry.
While GLODAPv2.2020 is made available without any restrictions, users of the data should adhere to the fair data use

Summary 1135
GLODAPv2.2020 is an update of GLODAPv2.2019. Data from 106 new cruises have been added to supplement the earlier release and extend temporal coverage by 2 years. GLODAP now includes 47 years, 1972-2019, of global interior ocean biogeochemical data from 946 cruises. Figure 10 illustrates the seasonal distribution of the data. As for previous versions there is a bias around summertime in the data in both hemispheres; most data are collected during April through November in the Northern Hemisphere while most data are collected during November through April in the Southern 1140 Hemisphere. These tendencies are strongest for the poleward regions and reflect the harsh conditions during winter months, which make fieldwork difficult. Figure 11 illustrates the distribution of data with depth. The upper 100 m is the best sampled part of the global ocean, both in terms of number (Fig. 11a) and density (Fig. 11b) of observations. The number of observations steadily declines with depth. In part, this is caused by the reduction of ocean volume towards greater depths. Below 1000 m the density of observations stabilizes and even increases between 5000 and 6000 m; the 1145 latter is a zone where the volume of each depth surface decreases sharply (Weatherall et al., 2015). In the deep trenches, i.e., areas deeper than ~ 6000 m, both number and density of observations are low.
Except for salinity and oxygen, the core data were collected exclusively through chemical analyses of individually collected water samples. The data of 12 core variables: salinity, oxygen, nitrate, silicate, phosphate, TCO 2 , TAlk, pH, CFC-11, CFC-12, CFC-113, and CCl 4 were subjected to primary quality control to identify questionable or bad data 1150 points (outliers) and secondary quality control to identify systematic measurement biases. The data are provided in two ways: as a set of individual exchange formatted original cruise data files with assigned WOCE flags, and as globally and regionally merged data product files with adjustments applied to the data according to the outcome of the consistency analyses. Importantly, no adjustments were applied to data in the individual cruise files.
The consistency analyses were conducted by comparing the data from the 106 new cruises to GLODAPv2.2019GLODAPv2. . 1155 Adjustments were only applied when the offsets were believed to reflect biases relative to the earlier data product release related to measurement, calibration, and/or data handling practices and not natural variability or anthropogenic trends The Adjustment Table at https://glodapv2-2020.geomar.de/ (last access: 18 June 2020) lists all applied adjustments and provides a brief justification for each. The consistency analyses rely on deep ocean data (>1500 or 2000 dbar depending on region), but supplementary CANYON-B and CONTENT analyses considers data below 500 dbar. Data consistency 1160 for cruises with exclusively shallow sampling was not examined.
Secondary QC flags are included for the 12 core variables in the product files. These flags indicate whether (1) or not (0) the data successfully received secondary QC. A secondary QC flag of 0 does not by itself imply that the data are of lower quality than those with a flag of 1. It means these data have not been as thoroughly checked. For δ 13 C, the QC results by  for the North Atlantic were applied, and a secondary QC flag was therefore added to this variable. 1165 The primary, WOCE, QC flags in the product files are simplified (e.g., all questionable and bad data were removed). For salinity, oxygen, and the nutrients, any data flagged 0 are interpolated rather than measured. For TCO 2 , TAlk, pH, and fCO 2 any data flagged 0 are calculated from two measured seawater CO 2 variables. Finally, while questionable (WOCE flag =3) and bad (WOCE flag =4) data have been excluded from the product files, some may have gone unnoticed through our analyses. Users are encouraged to report on any data that appear suspicious. 1170 Based on the initial minimum adjustment limits and the improvement of the consistency from the adjustments (Table 6), the data subjected to consistency analyses are believed to be consistent to better than 0.005 in salinity, 1 % in oxygen, 2 % in nitrate, 2 % in silicate, 2 % in phosphate, 4 µmol kg -1 in TCO 2 , 4 µmol kg -1 in TAlk, and 5 % for the halogenated transient tracers. For pH, the consistency among all data is estimated as 0.01-0.02, depending on region.  Tables 7 and 8 Are Olsen 31/7/2020 11:42 Deleted: and 5 % for the halogenated transient tracers. For TAlk the stated consistency for GLODAPv2 is 6 µmol kg -1 . We now believe this is better, 4 µmol kg -1 , not only for 1215 the 116 new cruises, but for all data in GLODAPv2 from 2016 as well. This is based on the global average absolute offset for TAlk in the adjusted ... [46] 7 Author contributions.
AO and TT led the team that produced this update. RMK, AK, and BP compiled the original data files. NL conducted the 1220 secondary QC analyses. HCB conducted the CANYON-B and CONTENT analyses. CS manages the Adjustment Table   e-infrastructure. AK maintains the GLODAPv2 webpages at NCEI/OCADS while CSL maintains www.glodap.info. PM prepared PYTHON scripts for the merging of the data. All authors contributed to the interpretation of the secondary QC results and decisions on whether to apply actual adjustments. Many conducted ancillary QC analyses. AO wrote the manuscript with input from all authors. 1225

Competing interests
The authors declare that they have no competing interests.

Acknowledgements
GLODAPv2.2020 would not have been possible without the effort of the many scientists who secured funding, dedicated time to collect, and shared the data that are included. Chief scientists at the various cruises and principal investigators for        With data  106  101  97  97  97  92  96  82  16  21  3  0   No data  0  5  9  9  9  14  10  24  90  85  103  106   Unadjusted a  89  85  82  73  75  68  67  65  12  17  2      shows the data below the upper depth limit (in this case 2000 dbar) as points and the interpolated profiles as lines. Non-interpolated data either did not meet minimum depth separation requirements ( Table 4 in Key et al., 2010) or are the deepest sampling depth. The interpolation does not extrapolate. Panel (e) shows the mean difference profile (black, dots) with its standard deviation, and also the weighted mean offset (straight, red) and weighted standard deviation. Summary statistics are provided in (c). The left shows the depth profile while the right shows the absolute difference between measured and estimated values divided by the 2785 CANYON-B/CONTENT uncertainty estimate, which is determined for each estimated value. A value below 1 indicates a good match between the two as it means that the difference between measured and estimated values is less than the uncertainty of the latter. The statistics in each panel are for all data deeper than 500 dbar and N is the number of samples; considered. A gain ratio and its interquartile range is given for the nutrients. For the seawater CO 2 chemistry variables the numbers on each panel are the median difference between measured and predicted values for CANYON-B (upper) and CONTENT (lower). Both are given with their 2790 interquartile range. Figure 6. Distribution of applied adjustments for each core variable that received secondary QC. Grey areas depict the initial minimum adjustment limits. The figure includes numbers for data subjected to secondary quality control only. Note also that the y-axis scale is set to render the number of adjustments to be visible, so the bar showing zero offset (the 0 bar) for each variable is cut off (see Table 5 for these numbers).

2775
2795 Figure 7. Distribution of pH offsets for the cruises from Japan Meteorological Agency added in GLODAPv2.2020. Figure 8. Distribution of applied adjustments per decade for the 946 cruises included in GLODAPv2.2020. Dark blue: not adjusted; light blue: absolute adjustment is smaller than initial minimum adjustment limit (Table 3); orange: absolute adjustment is between limit and 2 times the limit, red: absolute adjustment is larger than 2 times the limit.       (PACIFICA) synthesis , and data from 168 additional cruises. The additional cruises include many collected within the framework of the "repeat hydrography" program , instigated in the early 2000s as part of CLIVAR and since 2007 organized as the Global Ocean Ship-based Hydrographic Investigations Program (GO-SHIP). Both GLODAPv1.1 and GLODAPv2 data were released in three formats: (i) as submitted by the data originator but reformatted to WOCE exchange format  and subjected to primary quality control to flag outliers, (ii) as a merged data product with bias minimization adjustments applied, and (iii) as globally mapped climatological distributions. We refer to the first as the original data, to the second as the data product, and to the third as the mapped product.
The GLODAP products have been widely used. The first version formed the basis for the first data-based estimate of the global ocean inventory of anthropogenic carbon , and the descriptive paper on GLODAPv1.1  has been cited more than 800 times according to Web of Science (FAIR) principles (Wilkinson et al., 2016), and are largely adhered to by the oceanographic community.
Data are routinely made available on a per cruise basis through national and international data centers. variable units, such that the comparability between many data sets is poor. Standard operating procedures have been developed for some variables Hood et al., 2010; and certified reference materials (CRM) exist for seawater TCO 2 and TAlk measurements  and for nutrients in seawater (CRMNS; . A total of twelve years separated the release of the two versions of GLODAP. The urgency and complexity of modern climate change issues necessitate more frequent updates. Ocean carbon uptake responds quickly to annual-to-decadal changes in ocean circulation (Fröb et al., 2016;Landschützer et al., 2015), ocean acidification is progressing at unprecedented rates and already causing carbonate mineral undersaturation in some regions (Feely et al., 2008;Qi et al., 2017), oxygen minimum zones are rapidly expanding (Breitburg et al., 2018), and declining nutrient supply to the euphotic zone is potentially changing phytoplankton composition in certain large ocean regions (Rousseaux and Gregg, 2015). In addition, improvements in data management practices and increased computational resources are transforming approaches to, and expectations for, integrated data products. The Surface Ocean CO 2 Atlas (SOCAT) is a prominent example in this regard with annual releases and rapid use in global carbon budgets (Bakker et al., 2016;Bakker et al., 2014;Le Quéré et al., 2018;Pfeil et al., 2013). GLODAP is also becoming an important source of calibration and validation data for the biogeochemical sensors that are now deployed on autonomous platforms. Altogether, regular and rapid updates are important.

However,
This contribution documents the first such regular update of GLODAP, which adds data from 116 new cruises to the 724 included in GLODAPv2 and corrects errors and omissions in GLODAPv2. It also forms the basis for the documentation of future updates, adopting the Earth System Science Data "living data" format for evolving data sets.  Table A1 in the Appendix. total scale, and at 25°C and 0 dbar pressure, which is the GLODAP standard. These data were converted to total pH scale and temperature and pressure of 25°C and 0 dbar. The conversions were conducted by using CO2SYS  for MATLAB  with reported pH and TAlk as inputs, and generating pH output values at total scale at 25°C and 0 dbar of pressure (named phts25p0 in the product). Whenever TAlk data were missing, these values were approximated as 67 times salinity. The proportionality (67) is the mean ratio of TAlk to salinity in the GLODAPv2 data. This is sufficiently accurate for scale-temperature-pressure conversions. Data for phosphate and silicate are also needed, and were, whenever missing, determined using CANYON-B . The conversion was conducted with the carbonate dissociation constants of , the bisulfate dissociation constant of Dickson (1990), and the borate-to-salinity ratio of . These procedures are the same as used for GLODAPv2 , except for the CANYON-B estimation of phosphate and silicate.
The secondary quality control of the pH data also followed previous procedures, using a combination of crossovers and internal consistency calculations. The latter were conducted when a cruise had data for TCO 2 and TAlk, in addition to pH. Note that internal consistency was only considered for the secondary QC of pH, and not for the secondary QC of TCO 2 and TAlk. Hence, the adjustments applied for pH are not only a bias correction but also a seawater CO 2 chemistry consistency correction. This is one factor that makes the secondary quality control of pH data problematic, in particular with regard to the application of a uniform correction for an entire cruise or leg based on offsets in deep data. pH dependent offsets between pH determined spectrophotometrically with purified dyes and pH calculated from TCO 2 and TAlk have recently been found. For example, at a pH of 7.6 the calculated pH is higher by ~ 0.01 than measured pH . The causes of these discrepancies are not entirely clear, suggestions include deficiencies in dissociation constants used for the seawater CO 2 chemistry calculations, errors in the total boron-to-salinity ratio, and unknown protolytes affecting the TAlk Fong and Dickson, 2019). Such low pH values exist only in the deep North Pacific Ocean. Here, application of pH corrections based on seawater CO 2 consistency considerations could impact the correction. Broadly speaking, the pH data in GLODAP have been obtained using a variety of methods (e.g. potentiometric measurements, and spectrophotometric measurements with purified or impure dyes). The pH values produced by these different approaches have documented pH-dependent offsets from one another (Carter et al., 2013;Liu et al., 2011;Patsavas et al., 2015;Yao et al., 2007) that challenge the viability of the uniform adjustments applied . While we have continued to apply such uniform offsets for this update, we have chosen the higher initial minimum adjustment limit of 0.01, which is twice that used for GLODAPv2 Values for potential temperature and potential density anomalies (referenced to 0, 1000, 2000, 3000, and 4000 dbar) were calculated using Fofonoff (1977) and Bryden (1973). Neutral density was calculated using Sérazin (2011). Apparent oxygen utilization was determined using the combined fit in .
Whenever only two seawater CO 2 chemistry variables were reported, the third was calculated using CO2SYS    ; this fraction is therefore improving, but it is still too large. Our simple linear 1 % compared to 5 % for the 724 GLODAPv2 cruises), for the halogenated transient tracers (0 %-3 % adjusted, depending on variable, compared to 6 %-10 % for GLODAPv2), and for TCO 2 (two cruises, i.e., 2 % compared to 1 7% for GLODAPv2).
The distributions of the magnitude of adjustments applied are presented in Fig. 5 and Table 6. For salinity, oxygen, and silicate, adjustments between 1 and 2 times the initial minimum adjustment limit are most prevalent. For nitrate, phosphate, CFC-11, and CFC-12, adjustments equal to or larger than 2 times the limit are most prevalent. For the salinity and oxygen this reflects that any biases in the data tends to be between 1 and 2 times the limit, while for CFC-11 and CFC-12 it also likely reflects limitations in our ability to confidently identify small biases. These limitations are related to the strongly transient nature of the CFCs. For TCO 2 and TAlk, none of the adjustments are larger than 2 times the adjustment limit, and for both properties half of the adjustments applied are below the limit. For TAlk, this distribution of adjustments supports the lowered minimum adjustment limit of 4 µmol kg -1 (instead of 6 µmol kg -1 ); these data have sufficient precision to enable the identification of such small adjustments.
For TAlk, seven out of eight adjustments are positive (i.e., the data are biased low), for pH nine out of 10 adjustments are positive, and for oxygen six out of seven are positive. The adjustments for other variables were more distributed around zero. For TAlk, prevalence of a negative bias was also observed in the interlaboratory comparison reported by Bockmon and Dickson (2015), who suggested the cause being the use of end point titrations rather than the (preferred) equivalence point titrations. However, 6 out of 7 of the negative bias cruises were Japanese. A tendency for bias in Japanese cruises to be negative was also identified in GLODAPv2 and may be due to the use of internal reference material. We note that the TAlk data from 23 out of 29 Japanese cruises with viable deep crossover checks had no apparent deep offset, so the majority of new TAlk data from Japan were consistent with GLODAPv2 even with the lowered threshold.
The prevalence of positive pH adjustments may relate to the fact that at low pH (as is common in the deeper waters where crossover analyses are done), measurements made with purified dyes tend to be lower than pH determined using electrodes, using impure spectrophotometric dyes with older dye coefficients (Clayton and Byrne, 1993), or calculated from TCO 2 and TAlk . The latter three types of pH data constitute the bulk of the reference data for the consistency checks, so the prevalence of a modern negative bias may be a consequence of limitations in the approaches used for the secondary quality control of the pH data in GLODAP. As mentioned above, refining these should be a priority in the future.
Here, we acknowledge the issue and believe that a realistic estimate of the consistency of the pH data in the product is approximately 0.01-0.02.
Crossover comparison is conducted on deep-water samples so atmospheric exchange during sample collection on the new cruises is not a viable explanation for the trend of positive oxygen adjustments.
Atmospheric contamination would usually increase deep-water oxygen concentrations since deep oxygen levels are usually low. The data are not collected in any particular region, or associated with any specific laboratory, country, or method. Consequently, no particular explanation can be offered for the prevalence of positive adjustments. Ocean silicate for the adjusted data is 11.1 % and that for salinity is 10 ppm (i.e., a salinity of 0.01). This can be ascribed to two cruises, 58GS20130717 and 58GS20160802, conducted in the Greenland Sea where an increasing presence of Arctic sourced deep waters generates changes in these properties (Blindheim and Rey, 2004;Olafsson and Olsen, 2010;Olsen et al., 2009) that have not been corrected for. The impact of northern variability on the final consistency estimate can be determined for the Atlantic Ocean by excluding all data north of 50° N from the analysis. This gives a much better initial and final consistency, on par with that for the Indian and Pacific Oceans (Table 8).  , the widespread use of CRMs , and instrument automation. pH adjustment frequency also has a downward trend; however, the situation is far from ideal and a topic for future development in GLODAP. For the nutrients and oxygen, only  . We now believe this is better, 4 µmol kg -1 , not only for the 116 new cruises, but for all data in GLODAPv2 from 2016 as well. This is based on the global average absolute offset for TAlk in the adjusted GLODAPv2 data product of 2.8 µmol kg -1 ( Table 5 in ) and the use of the initial minimum adjustment limit of 4 µmol kg -1 for the cruises added with the present version. For pH on the other hand, the consistency among all data is likely not better than 0.01-0.02