Nong et al. 
34.65 
Salinity on 9 = 2.8°C 
34.6 
a 
3 | 
534.55 
34.5 
& 
DD 
34.45 
Argo 
— GO-SHIP P18 
-20 0 
Latitude (degree) 
-40 
20 
FIGURE 13 | Salinity on the potential temperature surface of 2.8°C along the 
218 GO-SHIP occupation from 2016. P18 lies roughly along 110°W in the 
Pacific Ocean. Black line is salinity from the GO-SHIP CTD casts, and the gray 
shading shows a 0.01 PSS-78 envelope around the GO-SHIP measurements, 
ndividual Argo salinity estimates from within 1° of the GO-SHIP stations are 
shown in blue. 
in aggregate, across all pairs, the results were similar to those 
from the Druck cohort. Overall, we found no evidence of a large 
pressure bias for Druck and Kistler profiles, though a small bias 
might exist near the boundary of the historical manufacturer’s 
stated sensor accuracy of 2.4 dbar. 
Assessment of Salinity Bias 
A similar analysis was done for salinity differences AS on the 
o1 surface across the pairs. For the cohorts of CTD type for 
which there were enough pairs (SBE-41/41CP), any bias in the 
dataset was much smaller than 0.01 PSS-78 (Figure 12B). Across 
most of the water column, the bias was about 0.001 PSS-78 
for the Druck and Kistler pressure sensor cohorts. There was 
a small fresh bias that peaked around —0.002 PSS-78 in the 
lower thermocline (400-800 dbar), but it was not evident in 
all the hemispheres. While GO-SHIP profiles do contribute to 
Argo’s reference database used to assess salinity sensor drift, their 
small number, along with the fact that only about 15% of Argo 
profiles are adjusted, means that they likely do not dominate these 
estimates. Thus, it is remarkable that, in aggregate, Argo profiles 
show such small salinity bias compared to the contemporaneous 
GO-SHIP surveys. This result is also consistent with the small 
pressure bias analyzed above. For example, a pressure bias of 10 
dbar will manifest as a salinity bias of 0.005 PSS-78 near 2,000 
dbar, which is not evident in Figure 12. 
Another way to assess float salinity accuracy is by comparing 
Argo salinity estimates on a deep potential temperature surface 
found in the ancient water masses of the deep Pacific with that 
measured by the GO-SHIP program (Figure 13). In the tropics 
and subtropics, the P18 line samples waters at 2.8°C that are 
low in oxygen and high in carbon isotopes, suggesting their great 
age and the relative absence of surface forced influences. On this 
rontiers in Marine Science | www.frontiersin.orı 
Argo Data 1999-2019 
isotherm, GO-SHIP salinities show very low variance between 
stations (< 0.003 PSS-78) north of 20°S. In comparison, the Argo 
salinity estimates vary much more, but largely within 0.01 PSS-78 
of the GO-SHIP values. The Argo values can be clumped above 
or below the GO-SHIP estimates, and these are associated with a 
single float record, suggesting that float salinity can be biased at 
the 0.01 PSS-78 level. 
A similar study was done by Riser et al. (2008), which 
compared salinities from 142 floats with shipboard CTD data 
collected along 32°S in the South Pacific. On the 2.4°C potential 
temperature surface, it was found that float-derived salinities 
agreed with shipboard data to within 0.01 PSS-78. This salinity 
accuracy is in accordance with the experience of the Argo 
delayed-mode teams and their ability to remove sensor drift 
or offsets. 
Positions, Subsurface Velocities, and Other 
Park-Phase Data 
Positions 
The ARGOS system uses the Doppler shift of received 
transmissions to estimate positions. As a result, its positioning 
accuracy depends on the number of satellites within range and 
the configuration of the constellation at the time the messages are 
received. ARGOS positions have four levels of accuracy ranging 
from better than 250-m radius to over 1,500-m radius. Some 
ARGOS position estimates are accompanied by an error ellipse, 
which gives a more exact error on individual positions than the 
broad horizontal error associated with the location classes. 
Floats that employ the Iridium satellite system for data 
communications use the Global Positioning System (GPS) to 
establish their positions. GPS tracking is more accurate than 
ARGOS tracking, with a typical GPS horizontal accuracy being 
about 8m (with a 95% confidence interval). Additionally, the 
[ridium satellite system itself can provide positions based on data 
{rom their satellites that are within range of the float. However, 
'ridium positions are of a much lower accuracy than GPS or 
ARGOS positions. Uncertainty in Iridium fixes is roughly 3 km in 
the meridional direction and about 20 km in the zonal direction; 
any individual Iridium fix can have much larger errors. Hence, 
'ridium positions are only used as a backup when GPS fixes 
cannot be obtained. 
Many floats operating in the Southern Ocean are equipped 
with an ice-avoidance algorithm to prevent the floats from 
reaching the surface when sea ice is inferred to be present 
(see section Geographical Coverage). These under-ice profiles 
are stored in the memory of the floats, but they are without 
any satellite-derived positions. If they do not have underwater 
acoustic positioning capability, their positions are estimated, 
most commonly by linear interpolation between known positions 
from ice-free periods. Chamberlain et al. (2018) estimated that 
maximum position uncertainty over an 8-month period was 116 
+ 148 km in the Weddell Sea, which was equivalent to about 1° 
in latitude and about 3° in longitude at 70°S. 
Subsurface Velocities and Other Park-Phase Data 
Computation of subsurface velocities from floats should ideally 
be based on the time and location of the float when it begins 
Qanteambear 2020 1 Valııme 7 1 Article Z01