Wong et al. 
drifting at park pressure and the time and location when the 
float stops drifting and begins to descend in preparation for 
the ascending profile to the sea surface. Unfortunately, these 
positions and times are, in many cases, not well-known. The 
only portion of a float profile where positions are known is 
at the sea surface, where satellites are used to determine the 
float’s location. Early APEX and SOLO floats transmitted some 
timing information, but the transmitted data were insufficient 
to determine both the times when the float reached the surface 
and when it began its descent. The first global subsurface velocity 
data product based on trajectories of Argo floats, YoMaHa’07, 
employed the last location and time on the sea surface from 
one cycle and the first location and time on the sea surface 
from the next cycle in order to make a subsurface velocity 
estimate (Lebedev et al., 2007). The YoMaHa’07 velocity product 
began with 290,247 cycles in 2007 and continued to be updated 
regularly. Ollitrault and Rannou (2013) used a similar method 
as YoMaHa07, but with improved estimation of ascent end 
time and descent start time, and 600,000 deep displacements 
based on ARGOS and GPS fixes from floats prior to January 
2010 to create the ANDRO Atlas, which continued to be 
updated yearly°. Using the ANDRO Atlas, Ollitrault and Colin 
de Verdiere (2014) provided a gridded field of geostrophic 
velocities at 1,000 dbar. Gray and Riser (2014) estimated surface 
arrival and departure times and positions and, together with 
geostrophic shear estimates from profile data, created gridded 
absolute geostrophic velocity fields for a number of levels in the 
upper 2,000 dbar of the global ocean. 
There are two main sources of errors in these velocity 
estimates: (i) unknown surface drift prior to the first and 
after the last location for ARGOS floats, and (ii) horizontal 
displacement when descending and ascending due to velocity 
shear. YoMaHa’07 estimated the global mean error due to both 
these sources to be 0.53 cm s7!. Knowing that floats experience 
much higher currents at the surface than at depth, Park et al. 
(2005) tried to reduce the error due to surface drift and improve 
the accuracy of the subsurface velocity by using a combination 
of linear and circular motion at the sea surface with ARGOS 
float locations, along with surface arrival and departure times 
to estimate the corresponding positions of surface arrival and 
descent. They demonstrated a velocity uncertainty of the order 
of 0.2 cm s7! in the Sea of Japan by using this method. 
In all of these efforts, the common difficulty in estimating 
velocities from Argo trajectory data results from a lack of timing 
information from the floats. In addition, for floats that use 
ARGOS communications, the locations at surfacing and descent 
are not well-known; the floats wait for unknown amounts of time 
at the surface prior to connecting with ARGOS satellites passing 
overhead in order to define a position, and then again wait for 
an undetermined amount of time after the last position before 
the float begins its descent. Newer float models that use Iridium 
communications return more timing information throughout 
the float mission, typically with a GPS fix at the beginning of the 
surface interval and a second GPS fix just prior to descending. 
Sdoi: 10.17882/47077 
trontiers in Marine Science | www.frontiersin.or 
Argo Data 1999-2019 
While drifting during their park phase (typically about 9 days 
in duration), some floats collect discrete samples of temperature, 
salinity, and other biogeochemical parameters. These underway 
data, available in the trajectory data files, have the same accuracy 
as the vertical profile data and have proven to be useful for 
studying high-frequency phenomena such as internal gravity 
waves (Hennon et al., 2014) and eddy diffusivity at 1,000 dbar 
(Roach et al., 2018). 
HOW TO CITE ARGO DATA: THE DYNAMIC 
DOI STRUCTURE 
The citation of Argo data used in scientific studies is a challenging 
subject since the Argo dataset is “dynamic,” evolving and growing 
in time. Dynamic data citation is an area of active research. To 
allow reproducibility of scientific studies that use Argo data, a 
snapshot of the entire dataset at the GDACs is preserved each 
month. The snapshot contains all the Argo data available at the 
time of the snapshot creation. To manage citation of this dynamic 
dataset, Argo adopted a Digital Object Identifier (DOI) format 
that gives a single DOI to track data usage, but that also allows 
users to cite specific time snapshots (Merceur, 2016). The Argo 
DOT takes the form http://doi.org/10.17882/42182#<nnnnn>, 
where <nnnnn> is the unique identifier for the specific time 
snapshot being used. Each snapshot identifier is appended to the 
DOI with a “#” character to delimit the suflix from the DOI. Based 
on this format, the Argo dataset can be cited in two ways: 
l. The Argo GDAC as a whole (without data reproducibility) 
should be cited as follows: Argo Data Management Team 
2019). Argo float data and metadata from Global Data 
Assembly Centre (Argo GDAC). SEANOE. https://doi.org/10. 
17882/42182. 
An Argo snapshot (enabling data reproducibility) should be 
cited as follows: Argo Data Management Team (2019). Argo 
float data and metadata from Global Data Assembly Centre 
(Argo GDAC)—Snapshot of the Argo GDAC as of September 
6th, 2019. SEANOE. https://doi.org/10.17882/42182#66797. 
FUTURE CHALLENGES 
2. 
While the Argo Program has made monumental progress in 
the past two decades on the technical problems relating to the 
collection of CTD data by profiling floats, work on these issues 
continues to this day. Individual floats now provide quality data 
over many years, sending megabytes of data including basic CTD 
parameters and a myriad of other types of observations to the 
GDACs, followed by adjustment of the data in a finely tuned 
delayed-mode process. Yet there remains room for improvement 
in each of these areas. 
First, while many present floats provide excellent data for 
more than 5 years, there are still too many that fail in half of that 
time. While there are a number of reasons for these early failures, 
an all-too-often cause is the lack of adequate pre-deployment 
checks on the part of some float groups. A central lesson from 20 
years of Argo is that there is no substitute for vigilance in making 
sure floats are operating properly before they are deployed. Argo 
Qanteambear 2020 1 Valııme 7 | Article 701