Nong et al. 
Argo Data 1999-2019 
50. + 
0.100 
0.080 
> 0.060 
N 
2 0.040 
% 9.020 
< 
0.000 
—0.020 
3 
40. 
3] 
1] 
230. 
20. 
1AS!| > 0.005 psu 
‚AS | > 0.010 psu 
1AS| > 0.020 psu 
A 
nr 
a 
wrnl 
a 
Kan 
wM0=5904627 
a 
10. 
ya 
| 
D. 
KT AT TTTT—— 
40 B0 120 160 200 240 280 
Cycle 
an 
1 7 ’ — 
40 60 80 100 120 
Cycle 
FIGURE 10 | Percentage of salinity adjustments vs. the number of cycles, 
"om 10.048 Arao floats. 
FIGURE 11 | Two examples of SBE CTDs that showed salty sensor drift: a 
“slow” drift (WMO ID 5904629, in red), and a “fast” drift (WMO ID 5905247, in 
5lack}. 
float salinity data with about 0.01 PSS-78 uncertainty. In most 
cases, when the magnitude of sensor drift exceeds 0.05 PSS-78, 
the data will become erratic or will exhibit significant vertical 
variations in the amount of sensor drift. These salinity data are 
flagged as bad and not adjustable in delayed-mode. 
Problems Encountered 
The calibration drift of salinity sensors over time is a common 
problem in oceanography. Shipboard CTDs are recalibrated 
regularly to maintain their stability and accuracy, but this 
is obviously not possible for floats. Early float deployments 
that used the FSI inductive-style conductivity sensors with a 
dissolvable biocide coating showed that the cells tended to drift 
toward fresher values. SBE CTDs use an enclosed pumped system 
with the electrical conductivity of the seawater measured directly. 
This method of inferring salinity from conductivity produces 
highly accurate salinity estimates, but it relies on the geometry 
of the conductivity cell remaining stable and uncontaminated 
(Riser et al., 2008). Biocide is used in the pumped loop of the 
SBE conductivity cells to mitigate biological fouling on the cells. 
Occasionally the biocide can leak onto the cell, causing a fresh 
offset in salinity, but that usually gets washed away within a few 
sampling cycles and the salinity measurements return to being 
in calibration. An additional measure to prevent biofouling is 
to shut off the CTD pump before the instrument reaches the 
sea surface. As noted earlier, with the current use of Iridium 
telemetry, the time spent on the sea surface, where floats are 
most susceptible to biofouling and other hazards, is reduced 
(Roemmich et al., 2004). 
Overall the SBE CTD design has worked well over the years, 
with only a minority of conductivity cells showing mild sensor 
drift over time. However, starting around 2015, a larger than 
average number of SBE CTDs in the serial number band 6000- 
7100 developed a drift toward higher salinities within 2-3 years 
of deployment (Figure 11). Many of these SBE CTDs were still 
active as of the time of this writing and, as a result, a higher than 
normal portion of Argo real-time salinity data were subject to 
errors that were larger than 0.01 PSS-78. The best estimate was 
that, at the time of this writing, about 25% of real-time profiles 
might be subject to this salinity error. In the real-time data 
stream, the Argo national DACs have flagged salinity from the 
5BE CTDs in the serial number range 6000-7100 as questionable 
data. In the delayed-mode data stream, the adjustment of these 
affected SBE CTDs has been treated with high priority. As a result, 
the residual salinity bias in the Argo dataset due to this sensor 
drift is now small. The cause of this conductivitv drift is presently 
still under investigation. 
Assessment of Argo Pressure and Salinity 
Bias Against GO-SHIP 
5Shipboard CTD systems, used with water sampling bottle 
salinities and recently calibrated sensor sets, deliver the highest 
possible accuracy data presently available. Here we have 
attempted to quantify any possible bias in Argo pressure and 
salinity by comparison with data from the Global Ocean Ship- 
Based Hydrographic Investigations Program (GO-SHIP). GO- 
SHIP grew out of the WOCE Hydrographic Program and 
adopted and built upon WOCE data standards. CTD data from 
280 post-2000 GO-SHIP cruises were selected. Nearby Argo 
and GO-SHIP profile buddies, or pairs, were defined as profiles 
deeper than 1,300 dbar that were collected within 300 km and 30 
days of each other. Individual Argo and GO-SHIP profiles could 
(and typically did) appear in multiple pairs, but each pairing was 
unique. In total, 294,373 Argo/GO-SHIP pairs were found with 
these search criteria, involving 31,056 unique Argo profiles. Only 
the highest-quality data have been selected (WOCE QC flag “2”; 
Argo QC flag “1”). The Argo profiles consist of both real-time 
profiles and delayed-mode profiles. 
Argo/GO-SHIP differences were then examined on a density 
surface: specifically, potential density anomalies relative to 1,000 
dbar (01). In order to account for the wide variation in 01 
stratification across latitudes, a separate 01 grid was defined for 
each 20° latitude band (—-90° to —70°, —70° to —50°, etc.). These 
co, grids were determined by averaging GO-SHIP o1 profiles 
in each latitude band and on 10-dbar pressure levels from the 
surface to 2,000 dbar. Salinity and pressure from each Argo/GO- 
SHIP profile pair were then interpolated onto the respective 07 
grid based on their locations. 
rontiers in Marine Science | www.frontiersin.orı 
Zantemhbhear 2020 1 Valıme 7 1 Article Z0