_e Menn et al. 
necessary (O’Carroll et al., 2008). Only after validation, the 
resulting SST retrievals can be used to generate global datasets 
with spatio-temporal consistency (e.g., Titchner and Rayner, 
2014). 
In situ SST measurements go back at least 200 years (Kennedy, 
2014). They have been collected for several purposes and with 
varying instruments. The first measurements were made from 
seawater collected by buckets, and after by seawater circulating 
through the steam condenser of the engine room inlets on 
ships. Since the 1970s, oceanographic and vessels of opportunity 
are equipped with hull thermometers. In quiet sea states, 
they measure temperature at 5 or 6m under the surface and, 
for the last 10 years, many have been equipped with high- 
resolution and stable Sea-Bird Electronics (SBE)-38 sensors 
(e.g., Gaillard et al., 2015). Argo profiling float temperatures 
are also used for comparisons. Since January 2005, they offer 
comprehensive ocean coverage (Hausfather et al., 2017). Argo 
products provide temperatures at different depths: 2.5, 5, 10, 
20, 30 m, or deeper levels with an initial accuracy close to 2 
mK. Sensors are generally stopped several meters below the 
surface to avoid the fouling of the conductivity cell by surface 
contaminants. A few floats are equipped with SBE STS (Surface 
Temperature Salinity) sensors which sample the final meters 
up to the surface, with a degraded accuracy in salinity, but 
most of Argo temperatures exploited as SST are measured at 
5m under the surface (Roemmich and Gilson, 2009). While 
only the initial accuracy has been guaranteed so far, first efforts 
have been made to recover Argo floats, in order to document 
potential changes in trueness (BIPM, 2012) over time (e.g., 
Oka, 2005). 
Generally, measurements made in the upper 10m of the 
acean are considered as SST measurements. However, satellite 
infrared radiometers measure radiations emitted from the upper 
few tens of microns (skin temperatures) or millimeters (subskin 
temperatures) for microwave radiometers (Donlon et al., 2004). 
Therefore, surface drifting buoys observations are preferred 
for comparisons with satellites data, as their sensors are at a 
nominal depth of between 10 and 20 cm (Merchant et al., 2012). 
According to the Data Buoy Cooperation Panel (DBCP) about 
1,500 drifting buoys cover nowadays the seas of the globe and 
according to (Kennedy, 2014), they provide about 90% of in situ 
SST data. 
Designed in the 1980s to study ocean currents in the context 
of the Surface Velocity Program (SVP; World Climate Research 
?rogramme, 1988) and for meteorological purposes, these buoys 
had to be inexpensive, easy to deploy and reliable during at 
least 18 months. The design specifications of SVP drifters were 
standardized in 1991. In 1993, it became possible to equip a 
SVP drifter with a barometer port to measure sea-level air 
pressure. The result was called a SVP-B drifter. SVP drifters 
were also equipped with SST sensors. This sensor should have 
an accuracy of 0.1K with a stability better than 0.1 K/year 
(World Climate Research Programme, 1988). There were other 
documented requirements, though less stringent, with 0.5K 
requested in the range from —5 to 30°C (EGOS, 2002). Of 
note in the SVP-B design manual (Sybrandy et al., 2009), is the 
requirement that a thermal isolation be included to ensure that 
rontiers in Marine Science | www.frontiersin.or. 
SVP-BRST Fiducial Reference Network 
the solar heating of the top of the surface float does not impact 
the SST measurement. The sensor should be accurate to better 
than 0.1 K when the inside of the float is 1 K warmer than the 
zea surface. 
In his publication, Kennedy (2014, Table 2, p. 8) cites 10 
references dealing with estimates of measurement errors or 
uncertainties of drifting buoys (with no clear distinction between 
error and uncertainty). They range from 0.12 to 0.67K. He 
discusses also the possibility to separate observation errors 
or uncertainties into random and systematic components, 
particularly for drifters, from two earlier publications (Kennedy 
et al., 2011a,b) and from a publication by Kent and Berry 
2008). They find similar results with estimated random 
components of (respectively) 0.56 and 0.6K and systematic 
components of (respectively) 0.37 and 0.3 K. These values are 
zlose, for example, to the expected accuracy of the Advanced 
Along-Track Scanning Radiometer (AATSR) launched in 
March 2002. It is designed to produce SST retrievals to 
better than 0.3K accuracy, with a long-term stability of 
vetter than 0.1K per decade (Lewellyn-Jones et al., 2001). 
Therefore, the corresponding drifting buoys SST measurements 
collected so far cannot be considered as references from a 
metrological point of view. Neither can they be considered 
as references for the more recent EUMETSAT-operated 
Copernicus Sentinel-3A, the first in a new generation of satellites 
designed to collect and monitor long-term climate and ocean 
data with metrological specifications equivalent to AATSR 
(Donlon et al., 2012). 
Separating systematic and random components is not 
an easy task for SST measurements, because the data from 
several authors (see Kennedy, 2014 or Castro et al., 2012) 
suggest a dependency on the time period considered. 
I£ random components come from the variability in 
time and space of the thermal and dynamical states of 
the sea, in the case of SVP drifters, the biggest part of 
systematic components can come from the buoy and sensor 
conception and from the unknown temporal drift of their 
SST sensors. 
This short review underlines the need to develop a new 
concept of surface drifting float which would be characterized 
in metrology laboratory. Its design has to comply with the 
requirements of satellite SST measurement validation and 
must allow the link through comparisons of its measurements 
to the Systeme International d’unites (SI). This need was 
described in a EUMETSAT tender, the goal of which was 
to build a Fiducial Reference Measurements (FRM) network 
of 100 high-resolution SST drifting buoys for the Copernicus 
Sentinel satellites validation. The development of this network 
echoes also, for the ocean surface, the need raised by Immler 
et al. (2010) for upper-air measurements, to constitute an 
independent infrastructure based on a different measurement 
principle and for which uncertainties are defined. Beyond the 
needs underlined by the review, this development answers the 
necessity of assuring long-term stability of references (World 
Meteorological Organization, 2016), the uncertainties of which 
are fully characterized by a metrological approach, for climate 
change studies. 
Zantembear 2019 I Valıme A 1 Article S7£