Nong et al. 
x“ 
Bat ac 
Argo Data 1999-2019 
100 
500 
kam 
MO 
800 
v 1000 
1200 
1400 
L60C 
1800 
-20 -10 0 10 -20-10 0 10 -20-10 0 10 -20-10 0 
A Pressure (dbar) 
700 
SR 
SPF“ 
10 
z0 -10 0 106 
400 
600 
> 800 
om 
[4 
S 
5 1000 
2.1200 
£ 
1400‘ 
1600 
1800 
2000 + 
-0.02 -0.01 0 0.01 0.02 -0.02 -0.01 0 0.01 
A Salinity (PSU) A Salinity (PSU) 
0.02 -0.02 -0.01 0 20.01 0.02 
A Salinity (PSU} 
FIGURE 12 | Bias (Argo minus GO-SHIP} on o+ surface as a function of pressure (dbar). The thick white line is the median for all pairs in a cohort, and the thin white 
'\ines are the medians across the 4 hemispheres: east, west, north, and south. Colors qualitatively indicate the fraction of data in a bias bin for any given pressure (high 
’action in red, low fraction in blue). (A) Median pressure bias (Argo minus GO-SHIP), for buddy pairs averaged by pressure sensor makers. Black dashed lines show 
‚he manufacturer accuracy specification on deployment of 2.4 dbar. The number of floats in each buddy cohort is marked in the top left-hand corner. (B) Median 
salinity bias (Argo minus GO-SHIP}, for budav pairs averaged bv CTD types. Grav dashed lines show expected accuracv of 0.01 PSS-78 
For each pair, a pressure difference (AP) and a salinity 
difference (AS) were computed from the interpolated values on 
the 01 level, where GO-SHIP values were subtracted from Argo 
values. Differences between Argo and GO-SHIP buddy profiles 
were due to short time- and space-scale ocean variability (such 
as mixed layer and eddy variability) and instrument error. In this 
analysis, we assumed the ocean variability was random and thus 
averaged to near zero across large numbers of pairs. Non-zero 
averaged differences were assumed to be due to instrumental bias. 
Assessment of Pressure Bias 
Profile pairs were analyzed in cohorts based on pressure sensor 
manufacturer. To illustrate statistical repeatabilityv, we compared 
rontiers in Marine Science | www.frontiersin.or 
averages of AP as a function of pressure from 4 hemispheres: east, 
west, north, and south (Figure 12A). Argo profiles with Ametek 
and Paine pressure sensors had too few GO-SHIP buddies to 
deliver a statistically stable result, as medians from different 
hemispheres were divergent. By far the most abundant pressure 
sensor represented in the profile pairs was the Druck sensor. Our 
results showed that a slight high-pressure bias existed near 1,000 
dbar, but its magnitude was within the manufacturer’s stated 
sensor accuracy. At levels deeper than 1,200 dbar, the results 
were not stable statistically, as indicated by a lack of agreement 
between hemispheric averages. Kistler sensor results from this 
analysis were also noisy but suggested a slight high-pressure bias 
that was also near the manufacturer’s stated sensor accuracy; 
Qanteambear 2020 1 Valııme 7 | Article 701