Wong et al. 
firmware. As a result, profile data from these SOLO-W floats are 
offset upward by one or more pressure levels, resulting in a cold 
bias at depth from these instruments (Willis et al., 2009). Data 
from the affected instruments have been identified and flagged as 
bad in the Argo dataset, and account for about 1% of all Argo 
CTD profiles. These affected SOLO-W floats are no longer active. 
Salinity: Accuracy and Issues 
Manufacturer Static Calibration 
The SBE-41/41CP are calibrated as a complete unit such 
that the conductivity calibration is run concurrently with the 
temperature calibration. During the calibration process, an SBE- 
4 conductivity sensor is used as the reference sensor in the 
calibration bath. At the 24.0°C calibration, the bath salinity is 
checked with an Autosal laboratory salinometer standardized to 
[nternational Association for the Physical Sciences of the Oceans 
(IAPSO) standard seawater. The conductivity ratio of the SBE- 
4 reference to the Autosal is used to correct the conductivity 
reference over the calibration range. This procedure is repeated 
3-5 times in order to assess sensor stability. Static accuracy 
from the calibration process is 0.0003 Siemens per meter for 
conductivity, which corresponds to about 0.0035 PSS-78 in 
salinity accuracy at 2°C and 2,000 dbar. 
Sensor Response Correction 
Attaining the most accurate salinity from conductivity, 
temperature, and pressure measurements requires considerable 
processing and a number of corrections for various sensor 
response issues (e.g., McTaggart et al., 2010). For 1Hz or 
more frequently sampled data, the mismatch between the 
).5s response time of the SBE-41CP thermistor and the faster 
response of the conductivity cell must be taken into account. 
The combined effect of the difference in sampling time between 
conductivity and temperature by the CTD, plus the time required 
for water to flow from the thermistor into the cell must also be 
accounted for (e.g., Johnson et al., 2007; Martini et al., 2019). 
However, for bin-averaged data (on order of 10s per dbar) or 
spot-sampled data, these adjustments, which amount to fractions 
of a second, are not possible. They could be done within the CTD 
onboard the float prior to bin-averaging and transmission, but 
those corrections have not yet been implemented internally on 
the SBE-41/41CP. 
The conductivity cell thermal mass error (e.g., Johnson et al., 
2007) represents a longer (multi-second) time-scale error. The 
error results from the fact that the conductivity cell and its 
surrounding protective jacket (the covering of the conductivity 
cell) both store substantial amounts of heat, which they exchange 
among themselves, with the water outside the float (in the case 
of the jacket), and the water flowing through the conductivity 
cell. When the CTD is moving through a vertical temperature 
gradient, this can mean that the temperature of the water in the 
conductivity cell is not the same as the temperature measured 
by the thermistor. Since conductivity is a strong function of 
temperature, the temperature of the water in the cell must be 
estimated (and used) to attain an unbiased salinity measurement. 
Although the most obvious manifestation of this error is a “spike” 
at the base of the mixed laver, this error, left uncorrected, also 
rontiers in Marine Science | www.frontiersin.or. 
Argo Data 1999-2019 
causes a bias in the thermocline, and can exceed 0.01 PSS-78 in 
some cases. 
The conductivity cell thermal mass error can be corrected in 
a statistical sense, in spot-sampled data and 2-dbar bin-averaged 
data, assuming the temperature gradients are well-characterized 
at the telemetered data resolution, if the ascent rate of the float is 
known (Johnson et al., 2007). The correction coeflicients depend 
on the CTD type, with different coefficients for the SBE-41 and 
the SBE-41CP because of their different pumping strategies. The 
SBE-41CP pumps slowly and continuously when operated in 
CP mode, whereas the SBE-41 pumps faster but intermittently, 
turning on only for spot samples. Coefficients for the SBE-41CP 
in spot-sample mode have not been determined, and work is 
ongoing to better characterize this error (e.g., Martini et al., 
2019). 
Long-Term Sensor Stability 
The long-term stability of float salinity data is evaluated in 
delayed-mode by comparing time series of data from each float 
with nearby high-quality reference data on potential temperature 
surfaces. The differences between float-measured and reference 
values over several years are treated by statistical methods and 
represented by a piecewise linear fit to discern any observable 
trends over time (Wong et al., 2003; Bohme and Send, 2005; 
Dwens and Wong, 2009; Cabanes et al., 2016). The observed 
trends are then evaluated by oceanographic experts to determine 
whether they are due to sensor drift or due to ocean variability. 
'f the observed trends are determined to be due to sensor drift, 
then the salinity data from the affected floats are adjusted to 
the reference data according to the piecewise linear fit over 
time. The salinity adjustment is computed as a multiplicative 
correction in potential conductivity, which is equivalent to an 
additive correction in salinity, with slight variations as a function 
of pressure due to the non-linearity of the equation of state 
for seawater. This model assumes that the changes in reported 
salinities are due to changes in the measurement volume of the 
conductivity cell (Lueck, 1990). In practice, this model works well 
for any salinity sensor drifts that can be adjusted with an offset 
correction with no significant vertical variations. 
The effectiveness of this statistical method relies on availability 
of contemporaneous reference data and/or the existence of water 
masses that have stable temperature-salinity characteristics for 
comparison. In Argo delayed-mode salinity analysis, two separate 
reference databases are used: a first one that is based on shipboard 
CTD data, and a second one that is based on Argo profiles 
that have been judged as accurate and needing no adjustment. 
Both databases are updated periodically to include more recently- 
acquired reference data, so as to account for temporal changes in 
the global ocean. 
Analysis of delayed-mode salinity data from 10,048 Argo 
floats showed that for the first 2 years after deployment 
(about 72 cycles), < 10% of the floats required any kind of 
sensor drift adjustment (Figure 10). After 280 cycles, about 40% 
required salinity adjustments > 0.01 PSS-78 in magnitude, while 
30% required adjustments > 0.02 PSS-78 in magnitude. With 
adequate reference data and stable water masses, the statistical 
technique used in delayed-mode can usually produce adjusted 
Qanteambear 2020 1 Valııme 7 | Article Z01