CARINA alkalinity data in the Atlantic Ocean

CARINA alkalinity data in the Atlantic Ocean A. Velo, F. F. Perez, P. Brown, T. Tanhua, U. Schuster, and R. M. Key Instituto de Investigaciones Marinas, CSIC, Eduardo Cabello, 6, 36208 Vigo, Spain School of Environmental Sciences, University of East Anglia, Norwich, NR4 7TJ, UK Leibniz-Institut für Meereswissenschaften, Marine Biogeochemie, Kiel, Germany Princeton University, Program in Atmospheric and Oceanic Science, Forrestal Campus/ Sayre Hall, Princeton, NJ 08544, USA Received: 25 June 2009 – Accepted: 3 July 2009 – Published: 3 August 2009 Correspondence to: A. Velo (avelo@iim.csic.es) Published by Copernicus Publications.

For a complete list of parameters for the CARINA data base, see Key et al. (2009).Note the different names for the parameters in the Exchange files (the individual cruise 1 Introduction CARINA is a database of carbon and carbon relevant data from hydrographic cruises in the Arctic, Atlantic and Southern Oceans.The project was formed as an essentially informal, unfunded project in Kiel, Germany, in 1999, with the main goal to create a database of carbon relevant variables in the ocean to be used for accurate assessments of oceanic carbon inventories and uptake rates.Not only the collection of data, but also the quality control of the data has been a main focus of the project, with both primary and secondary quality control (QC) of the data having been performed.The CARINA database consists of essentially two parts: the first part consists of the individual cruise files where all the measured data, and their quality flags, are stored.These files are in WHP (WOCE Hydrographic Program) exchange format where the first lines consist of the condensed metadata.There are essentially no calculated nor interpolated values in the individual cruise files, with the exceptions of pressure calculated from depth and some bottle salinities that were taken from ctdsal.No adjustments have been applied to any of these values, with the exception that all pH measurements were converted to the seawater pH scale at 25 • C.
The second part of CARINA consists of three merged quality controlled and adjusted data files; one each for the Atlantic Ocean, Arctic Mediterranean Seas and Southern Ocean regions.These files contain all the CARINA data and include: 1) interpolated values for nutrients, oxygen and salinity if those data are missing and if interpolation could be made according to criteria described in Key et al. (2009); calculated carbon parameters (e.g. if total dissolved inorganic carbon (TCO 2 ) and Total Alkalinity (A T ) were measured, pH can be calculated).Calculated and interpolated values have the quality flag "0".All the values in the merged data file have been adjusted according to the values in Table 2 and described in Sect. 5.In many cases there are more reported parameters in the individual cruise files than has been included in the secondary QC, such as 14 C, 13 C and SF 6 .This report describes the consistency analysis of alkalinity measurements of the Atlantic Ocean part of the CARINA database (CARINA-ATL).A more comprehensive description of the complete CARINA data base can be found in Key et al. (2009), for an overview of the North Atlantic CARINA data, see Tanhua et al. (2009a).
The Total Alkalinity in sea water is defined as "...the number of moles of hydrogen ion equivalent to the excess of proton acceptors (bases formed from weak acids with a dissociation constant K ≤10 −4.5 , at 25 • C and zero ionic strength) over proton donors (acids with K >10 −4.5 ) in one kilogram of sample."(Dickson, 1981).A T is one of the four basic related parameters of the carbon dioxide CO 2 system in seawater, with the others being total dissolved inorganic carbon (C T ), the fugacity of dissolved CO 2 (f CO 2 ), and pH.If at least two of these are known, then the remaining parameters can be calculated and the entire carbonate system determined using thermodynamic constants for a given temperature, salinity and pressure.High quality seawater carbon data are critical for detecting small changes in the carbonate system.Both A T and C T are used in a number of methods for calculating the anthropogenic CO 2 signal, most specifically those that employ back-calculation techniques.A T is a key in both determining changes in C T produced by CaCO 3 dissolution (Feely et al., 2002) and establishing concentrations of C T in surface waters at equilibrium with the atmosphere, whether at historical, present, or future CO 2 levels Furthermore, many measurements of the CO 2 system in seawater have been performed using the A T -pH pair for determining C T .For these reasons, this is a key parameter for the CARINA objectives.

Data Provenance and Structure
As derived by its definition, the salt content of seawater is a determining factor on the A T .For this reason, salinity is a good parameter to explain some of the A T variability.In Fig. 1a, alkalinity is shown against salinity for the full CARINA-ATL dataset.As expected, the parameters correlate well, giving a R 2 regression coefficient of 0.77.Near surface formation of water masses is the main process that affects A T variability.In Fig. 1b, the variation of A T versus salinity for the subset of surface waters (depth < 200 m) is plotted.The correlation for this data set is higher than in Fig. 1a, giving a R 2 of 0.94 due to most of the variability in both salt and alkalinity occuring in the upper few hundred meters, where there is a strong salt-alkalinity correlation.This enforces the idea of the interdependence of both parameters.Alkalinity data included in the CARINA dataset originates from a multitude of international research groups using a number of different analysis methods.Whilst most of the A T data was determined using closed cell potentiometric titrations (Dickson and Goyet, 1994), many measurements were also made by potentiometric titration at end-point in open cell (Mintrop et al., 2000).
In total, data from 98 cruises were included in the dataset for CARINA-ATL, of which 75 reported A T data, totaling 65531 unique alkalinity measurements from 4774 stations.After the 1st QC check, 59793 data points (4580 stations from 73 cruises) remained flagged as good.However, crossover comparisons were performed on only 2562 stations (15 706 data points) as the remainder had insufficient deep water samples located sufficiently close together to enable station-station analyses.The results of the comparison exercise (2nd QC) are used for a total of 46593 A T data points from 4696 stations sampled on 51 cruises, ensuring their quality regardless of the application of a correction or not.This is equivalent to 71% of the original data.
Of the 51 cruises included in the analysis, alkalinity data from 31 had been generated using certified reference materials (CRMs) to test or calibrate the titration system.The remaining 21 did not use CRM at all, instead using an alternative solution for standardization (DOE, 1994).The most typical reported analytical error was around 1%, ∼2.5 µmol kg −1 .However, similar crossover exercises (Key et al., 2004;Sabine et al., 2005;Wanninkhof et al., 2003) performed on older data estimated an overall accuracy of ±5 µmol kg −1 .
The overall objective of this work was to assess of the quality of the Atlantic Carina alkalinity data in order to generate a mutually consistent database, by checking ESSDD 2, 2009

CARINA alkalinity data in the Atlantic Ocean
A. Velo et al. for apparent offsets for each cruise considering all the information reported by different sources (Tanhua et al., 2009b) For A T , the minimum adjustment to be applied to the final dataset following the identification of an additive offset was established as ±6 µmol kg −1 , as was the case for the previous crossover comparison study for the North Atlantic (Wanninkhof et al., 2003).

Methods
The methods and techniques applied here are described in detail in (Tanhua et al., 2009b).In summary, the procedure essentially entails the comparison of data from separate cruises whose tracks cross or at least come close to each other, so-called crossover analyses.Only data that were flagged "good" during the primary QC procedure (Key et al., 2009) are considered in this process, with the application of various software packages (Tanhua et al., 2009b) generating statistical and objective information about the offsets between pairs of cruises, as well as the graphics needed to visually verify the computer generated offsets.For each crossover analysis, A T data from samples deeper than 1500 m were compared on sigma-4 density surfaces generating an offset and a standard deviation of this difference, as well as totaling the number of contributing stations and samples.For alkalinity, additive offsets (and adjustments) were determined.
In this work, the semi-automated crossover procedure was run for all possible pairs of Atlantic cruises.Next, the crossover results were then visually inspected in order to ensure quality and to check the analysis had run correctly.Only "good" quality crossovers were selected, and those results were used for subsequent cruise adjustment calculations.Good crossovers had enough sample data to yield a reasonably uniform additive offset over the entire zone of analysis, leading to parallel cruise profiles.Standard deviations for individual cruise data and difference profiles were also used to provide more information on crossover quality.
In crossover (on the left) and a bad one (on the right) can be easily appreciated, as well as the lack of data and dispersion.The quality of the offset is assessed when the standard deviation of the offset is compared (±2.7 versus ±7.9 µmol kg −1 ) After this first iteration, automated procedures described by Tanhua et al. (2009b) weighted the crossover quality by statistical parameters for later adjustment calculations for all cruises.This process was followed for all available cruise crossovers in the Atlantic.In total, 337 individual crossovers were obtained for A T .
The final offsets and statistics were used as the input for an inverse least squares procedure (Tanhua et al., 2009b).Inversion results generated a set of suggested corrections for all cruises included in the analysis that minimized the differences.Figure 3 shows the offsets for all A T crossovers in the Atlantic region before (pink dots), and after (blue dots) the adjustments were applied to each cruise.The convergence to values inside the bounds can be easily appreciated for values after inversions.The standard error for the original crossovers is 10.5 µmol kg −1 whereas the standard error for the adjusted ones is 4.7 µmol kg −1 .
In order to ensure the highest quality results from the inversion and to help get a more accurate and consistent solution to the system, a small subset of cruises were a priori defined as "core".These were chosen according to their geographical extent (i.e.covering a large distance) and expected high data quality (i.e.WOCE/CLIVAR quality), and were agreed upon by the Carina Atlantic group.Offsets identified towards "core" cruises received a higher weighting in the inversion minimization process (Tanhua et al., 2009b).Once the full result of A T offsets for each cruise had been generated, only the suggested corrections that exceeded a predefined limit of ±6 µmol kg −1 were applied.These subsets of high values were used as starting point to establish the final adjustment values to be proposed.The decision on final values was made manually and by consensus among the CARINA collaborators.For this decision, subjective factors such as the location, use of CRMs, technique, date, quantity and quality of crossovers, or some particularly relevant crossovers were used.Only those offsets that were strongly supported by the analysis were finally adopted and subsequently applied to the measured results.The corrections that were actually applied to the data product are, in following, referred to as an adjustment.
After the need for an adjustment and its magnitude was established for each cruise, the values were applied to the original database, and the full process of crossovers and inversions repeated.The result of this second iteration was a very useful way of validating the proposed offsets.Figure 4 shows the corrections values for each cruise obtained after the first inversion procedures (values in black), and the corrections obtained after the second iteration (values in red).The results from the second inversion clearly show that the remaining offsets are lower, and most of them fit within the chosen minimum error boundary

Overall accuracy
The offsets for the crossovers applied to the data product were used to estimate the overall accuracy of the alkalinity data, Fig. 5.The weighted mean (W M) was calculated for alkalinity by using the absolute value of the offset (D) of the L crossovers with the uncertainty (σ): Based on this analysis we have estimated the accuracy of the CARINA-ATL alkalinity data to 3.3 µmol kg −1 The following data is presented in the table:

Results are summarized in
-Region: Location of the cruise.All cruises in Table 1 belong to Atlantic Ocean (NA), but some overlap with the Arctic Mediterranean Seas (AMS) or Southern Ocean (SO) areas.
-CRM: Indicates whether or not CRMs were used during instrumental analysis -WLSQ adj: Result of the inversion process through the Weighted Least Squares method (Tanhua et al., 2009b).
-WDLSQ adj: Result of the inversion process through the Weighted Damped Least Squares method (Tanhua et al., 2009b).
-Adjustment: Adjustments applied for the cruises in the merged data product.All adjustments are fully supported by the CARINA group and no adjustments smaller than ±6 µmol kg −1 are applied.
-Core cruises are highlighted in bold types.
Figure 6 is a comparison between the original corrections and the final adjustments applied.Blue dots represent offsets obtained by applying full solution (all corrections); pink dots represent offsets obtained with the final adjustments applied to the data product (Table 1).The relation for the applied adjustments (pink dots) has only a slightly lower correlation coefficient than for the full solution.
In the following paragraph a set of figures and comments are presented for each cruise summarizing all crossover offsets with their standard deviation.Each figure shows the following information: -Green dots: "Offsets".These values are the offsets taken directly from each crossover.The standard deviation is shown as error bars on these dots.
-Yellow line indicates the additive correction calculated by inversions for the cruise.Note that the correction and offsets are of opposite sign.-Blue squares: "Predicted offset" shows the calculated offset that would be obtained by applying all inversion corrections to the cruises.
-Red dots: These are the residuals between the "Offsets' (Green dots) and Predicted Offsets"' (Blue squares) c suffix in the upper X axis labels stands for Core Cruises.

Cruises
In this section, an assessment and description of the adjustments applied to cruises for CARINA-ATL database is made.Carina identifiers for the cruises are the numbers indicated between the parentheses.

Cruise 06MT19941012 (12)
This is the WOCE leg A02 cruise, with 53 stations and 24 sampling levels using a rosette system.A closed cell potentiometric titration method was used on measurements, and CRM Batch #22 was used as reference.The report indicates an esti-

Cruise 18HU19970509 (44)
This is AR07Wh, a CCHDO cruise with 13 stations using a Rosette system with 24 ten-liter bottles.CRM analyses were used for standardization and precision was 2.7 µmol kg −1 .There are eight crossovers giving a fitted correction of 6.2 µmol kg −1 .The suggested adjustment for this cruise is therefore 6 µmol kg −1 .Most of the residuals fits very close to 0 and keep inside ±5 µmol kg −1 after offsets applied.There are very good fits with four core cruises.Based on this evidence, an adjustment of 6 µmol kg −1 was applied to the alkalinity data.

Cruise 29CS19930510 (52)
This is WOCE line AR16e, also called 29CSMORENA 1.It has 92 stations and 24 sampling levels using a rosette system.CRMs were not used for these measurements.
There are nine crossovers.The fitted correction of inversions is 7.1 µmol kg −1 .The suggested adjustment is therefore 7 µmol kg −1 .Crossovers with three core cruises shows very low residuals after offset applied.
Based on this evidence, an adjustment of 7 µmol kg −1 was applied to the alkalinity data.

Cruise 29GD19840218 (55)
This cruise has 33 stations in a hydrocast with 1.7 liter Niskin bottles.A precision of 0.1% and accuracy of 1.4 µmol kg −1 is reported.No CRMs were used during the analysis.There are nine crossovers giving a fitted correction of 6.5 µmol kg −1 .The suggested adjustment is +6 µmol kg −1 .Almost all residuals fit very close to 0 and stay within ±5 µmol kg −1 after offsets applied.There are very good fits with three core cruises.Based on this evidence, an adjustment of 6 µmol kg −1 was applied to the alkalinity data.

Cruise 29GD19860904 (57)
This cruise has 50 stations in a hydrocast with 1.7 liter Niskin bottles.A precision of 0.1% and accuracy of 1.4 µmol kg −1 are reported.No CRMs were used.There are nine crossovers giving a fitted correction of +4.3 µmol kg −1 .The suggested adjustment is +6 µmol kg −1 as t-student statistical checks show that the proposed offset is indistinguishable from 6 µmol kg −1 .Almost all residuals fit very close to zero and keep inside ±5 µmol kg −1 after offsets applied.There are very good fits with three core cruises.
Based on this evidence, an adjustment of 6 µmol kg −1 was applied to the alkalinity data.

Cruise 29HE20010305 (61)
This is the Hesperides FICARAM II cruise HE073.It is an A17 repeat section in the western South Atlantic region.The cruise has 29 full depth stations and 24 sampling levels using a rosette system.CRM batches 41 and 51 were used.Uncertainty is 1.4 µmol kg −1 according to cruise report.There are 9 crossovers, two of them in the Southern Ocean giving a fitted correction of −5.1 µmol kg −1 .Analysis of cruise documentation seems to suggest a correction of −5 µmol kg −1 .The proposed adjustment is −6 µmol kg −1 , as it is between the offset error and is better supported than no adjustment.All residuals fit very close to 0 and keep inside ±5 µmol kg −1 .There are very good fits with five core cruises.Based on this evidence, an adjustment of −6 µmol kg was applied to the alkalinity data.

Cruise 316N19971005 (65)
This is the Knorr cruise KN154/2 on the WOCE AR24b section.It has 162 stations and 24 sampling levels using a rosette system.No reference to the use of CRM was reported.There are 15 crossovers giving a fitted correction of +8.1 µmol kg suggested adjustment is +8 µmol kg −1 .Excepting one, all inversion residuals fit inside ±6 µmol kg −1 .The cruise has a good fit with three core-cruises.Based on this evidence, an adjustment of 8 µmol kg −1 was applied to the alkalinity data

Cruise 31AN19890420 (64)
This cruise has 51 stations taken with a rosette system.Values seem to be about 5 to 10 µmol kg −1 too high in relation to OACES93 and A16N2003.CRMs were not used.The fitted correction is −9.6 µmol kg −1 .The inversions suggest an adjustment of −10 µmol kg −1 .Crossovers with core cruises support the proposed offset very well.In addition, GLODAP 317519930704 supports this adjustment.
Based on this evidence, an adjustment of −10 µmol kg −1 was applied to the alkalinity data

Cruise 323019940104
This is the CITHER2 cruise along the WOCE A17 section.In the cruise report, an offset for alkalinity of −8 µmol kg −1 was stated.There are eight crossovers that give a fitted correction of −8.9 µmol kg −1 .The suggested adjustment is −8 µmol kg −1 because it agrees with both the fitted result and the cruise report.Most of the inversion residuals fit inside ±5 µmol kg −1 .There are very good fits with four GLODAP cruises.
Based on this evidence, an adjustment of −8 µmol kg −1 was applied to the alkalinity data.

Cruise 35TH19990712 (106)
This is EQUALANT99 cruise, in the Equatorial Atlantic region.It has 102 stations and 24 sampling levels using a rosette system..Precision is reported as 1.7 µmol kg −1 .A comparison with WOCE A15 data implies offset values as high as 12.7 µmol kg −1 .
There are six crossovers, giving a fitted correction of -6.8 µmol kg Based on this evidence, an adjustment of −6 µmol kg −1 was applied to the alkalinity data

Cruise 64TR19890731 (153)
The cruise has 73 stations and 12 sampling levels using a rosette system.Nine stations have been flagged 3.There are nine crossovers giving a fitted correction of 13.8 µmol kg −1 .The suggested adjustment is 14 µmol kg −1 .All inversion residuals are close to zero and fit inside the ±5 µmol kg −1 boundary.There are very good fits with two GLODAP cruises.Based on this evidence, an adjustment of 14 µmol kg −1 was applied to the alkalinity data.

Cruise 64TR19900417 (154)
This is a meridional section from 30 • N to 60 • N along 20 • W. This cruise has 23 stations with multiple casts on most of them.Method of Bradshaw et al. (1981), and calculations with constants of Goyet and Poisson (1989).Values are about 20 µmol kg −1 low relative to CLIVAR A16N-2003 and have twice the scatter.There are 14 crossovers.The fitted correction is +14.4 µmol kg −1 .The suggested adjustment is +14 µmol kg −1 .After fitted corrections are applied, 10 inversion residuals fit inside ±5 µmol kg −1 .There are good fits with two GLODAP and one core cruise (06GA20000506).Based on this evidence, an adjustment of 14 µmol kg −1 was applied to the alkalinity data.

Cruise 74AB19910501 (160)
This is the 74AB058 VIVALDI cruise.It has 34 deep stations and 24 sampling levels using a rosette system.Values are low in relation to GLODAP by about 15 µmol kg −1 .
Good precision is noted.Laboratory-made borax standards were used but no CRMs.Full Screen / Esc

Printer-friendly Version
Interactive Discussion There are 16 crossovers giving a fitted adjustment of 5.2 µmol kg −1 .The suggested adjustment is 5 µmol kg −1 as a student t-test shows that the proposed correction is indistinguishable from 6. Crossovers with six core cruises also support the proposed offset.Furthermore, GLODAP 317519930704 supports this adjustment, as very low residuals are present after the offset has been applied.Based on this evidence, an adjustment of 5 µmol kg −1 was applied to the alkalinity data.

Cruise 74DI19900612 (170)
This cruise has 20 stations and 12 sampling levels using a rosette system.Comparing with one GLODAP station, values are about 18 µmol kg −1 lower.There are six crossovers giving a fitted correction of 14.9 µmol kg −1 .The inversions suggest an adjustment of 15 µmol kg −1 .Inversion residuals are inside a ±5 µmol kg −1 boundary after corrections are applied.Based on this evidence, an adjustment of 15 µmol kg −1 was applied to the alkalinity data.

Cruise 74DI19980423 (172)
This is the 74DI233 cruise, a meridional section along 20 • W from 20 • N to 60 • N. Data is generally good, with a few data high relative to neighbors and GLODAP.CRMs were used.There are 24 crossovers giving a fitted correction of −8.5 µmol kg −1 .The suggested adjustment is −9 µmol kg −1 .Most of the residuals fit very close to zero and keep inside ±5 µmol kg −1 after corrections are applied.There are very good fits with core and GLODAP cruises.Based on this evidence, an adjustment of −9 µmol kg −1 was applied to the alkalinity data.

Cruise OMEX2 (188)
There are nine crossovers giving a fitted correction of −8.1 µmol kg ±6 µmol kg −1 .There are many uncertainties.There are three crossovers with core cruises that appear to suggest an even larger offset.Based on this evidence, an adjustment of −8 µmol kg −1 was applied to the alkalinity data.

Data quality evaluation
In order to make an overall evaluation of alkalinity data quality, a Multi-Linear Regression (MLR) was done.
For improving the quality of the evaluation, the MLR analysis was applied in four density layers.Density at 1000 db (σ 1 ) was used to divide the ocean in four layers.The upper thermocline was set by σ 1 <32.25 kg m −3 ; intermediate waters (depths from about 1000 to 2000 m) were defined by second layer (32.25<=σ 1 <32.39 kg m −3 ); water depths between ∼2000 to 3000 m were defined by 32.39<=σ 1 <32.53 kg m −3 -corresponding to North Atlantic Deep Waters (NADW); and finally, the fourth layer attends for depths to bottom waters, where the presence of Antarctic Bottom Waters (AABW) dominates.This last layer is set by σ 1 >32.53 kg m −3 .The surface layer with depths <200 m was removed for this evaluation.
Using the MLR analysis, alkalinity residuals were calculated by the following Eq.( 2): where X i stand for Theta, Salinity, Latitude, AOU, Nitrate, Phosphate, Silicate, and a constant term.This procedure was done with the CARINA-ATL corrected database, and also for the database without any alkalinity adjustments applied.
The alkalinity residuals for each density layer are shown in Fig. 23.This shows a box plot of alkalinity residuals cruise data for each cruise.The width of individual boxes attends to the number of samples for that cruise.The best fit (R 2 =0.95) is obtained for the shallower waters, with a mean standard deviation of 6.2 µmol kg −1 .The other layers have a slightly lower mean standard deviation (5.4,5.9 µmol kg −1 respectively), except for the bottom layer with 6.5 µmol kg −1 .The median of mean standard deviations for all cruises is 4.1 µmol kg −1 units for each of three deepest density layers.The lower panel in the figure stands for the joined alkalinity residuals of the four density layers.
As can be seen, alkalinity residuals are lower when using the corrected database, in comparison with the uncorrected original ones.Most of the cruises have the alkalinity residuals median inside of the ±6 µmol kg −1 boundary.In addition to the alkalinity measurement errors, there are two other sources that increase the variability of the alkalinity residuals: firstly, the MLR is not able to explain all of the observed variability of alkalinity; and secondly, the predictor parameters have their own inherent measurement errors.2,2009

CARINA alkalinity data in the Atlantic Ocean
A. Velo et al.

CARINA alkalinity data in the Atlantic Ocean
A. Velo et al.
Fig. 2 two examples of A T crossovers are shown; the contrast between a good ESSDD

-
Black stars indicate the correction calculated by inversions for the other cruises that intersect this cruise.
mated precision of ±3 µmol kg −1 and an accuracy ±6 µmol kg −1 .This cruise has 19 crossovers.The inversions suggest a correction of −25.1 µmol kg −1 .Most of the residuals fit within ±10 µmol kg −1 , with half part of them inside ±5 µmol kg −1 after corrections applied.No good fit exists with the core-cruises however.The crossovers with 74DI19970807 and 06GA20000506 suggest a minor offset, but the other four core cruises (two of them GLODAP), suggest a correction of around −28 µmol kg −1 .Based on this evidence, an adjustment of −25 µmol kg −1 was applied to the alkalinity data.

Fig. 5 .Fig. 6 .
Fig. 5. Sorted offsets calculated for the crossovers in the CARINA-ATL data after adjustments have been applied.W L: the weighted mean of the offsets (see text); F : the percentage of offsets indistinguishable from 1 within their uncertainty; L: the number of crossovers.

Table 1 .
This table shows the results for cruises in this exercise with A T data in Atlantic Ocean, i.e. cruises with at least two acceptable crossovers.
. All inversion residuals fit inside ±6 µmol kg −1 after corrections are applied.There are very good fits with two GLODAP cruises.
A. Velo et al.