206 
P. Poli et al.: SVP-BRST: genesis, design, and initial results 
Ocean Sci., 15,199-214, 2019 
www.ocean-sci.net/15/199/2019/ 
Table 3. Individual calibration data for SST sensors from three HRSST-2 buoys that were fortuitously recovered. The mean error is the 
average difference, for several verification points, between the temperature reported by the sensor and the temperature of the calibration bath. 
Lab. no. 1 indicates the initial calibration and verification that was made then. The last column, showing temporal drift (in Kyear -1 ), is 
365.25 times the difference between the mean error assessed by lab. no. 2 (or 3) minus the mean error assessed by lab. no. 1, divided by the 
number of days elapsed. 
WMO id. 
Lab. no. 
Date 
Mean error 
Time interval since 
lab no. 1 (days) 
Temporal drift since 
lab no. 1 
4400871 
1 
02/10/2012 
-0.010 K 
0 
- 
2 
23/09/2016 
-0.063 K 
1452 
-0.013 Kyear -1 
3 
16/08/2017 
-0.043 K 
1779 
—0.007 K year -1 
4400608 
1 
16/10/2012 
-0.006 K 
0 
- 
2 
23/09/2016 
-0.055 K 
1438 
-0.012 Kyear -1 
3 
16/08/2017 
-0.037 K 
1765 
—0.006 Kyear -1 
6200552 
1 
01/09/2012 
0.031 K 
0 
- 
2 
23/09/2016 
-0.007 K 
1483 
—0.009 Kyear -1 
3 
16/08/2017 
+0.014 K 
1810 
—0.003 Kyear -1 
cept of uncertainty validation is presented in detail by Cor- 
lett et al. (2014). Briefly, the standard deviation of the satel 
lite/drifter differences is comprised of contributions from 
the satellite and drifter measurements, as well as terms to 
represent the spatial and temporal differences between the 
two measurements. Having used models to adjust the drifter 
measurement to be the same time and depth as the satellite 
SST, Corlett et al. (2014) showed the standard deviation of 
the satellite/drifter differences approximately reduces to two 
terms: the satellite SST uncertainty and the drifter SST un 
certainty as in Eq. (1). 
^Satellite minus drifter — -/^Satellite + ^drifter ( ^ ) 
Figure 4 shows a comparison between 1 October 2016 and 30 
June 2017 of satellite SST validation results for the dual-view 
three-channel retrieval from SLSTR for two sets of drifters: 
all drifters in Fig. 4a (15 551 matchups) and a subset of 
HRSST-1 and HRSST-2 drifters in Fig. 4b (625 matchups). 
In the figure, the green lines indicate the theoretical disper 
sion of uncertainties using Eq. (1) and a value of 0.20 K for 
drifter (an assumption between those of O’Carroll et al., 2008 
and Lean and Saunders, 2013). The blue lines indicate the 
calculated dispersion for each set of data and the red lines in 
dicate the standard error. If the assumptions are correct, then 
the dispersion of the blue lines should track the spread of the 
green lines, which we see is the case in Fig. 4a (all drifters). 
Where the dispersion does not match the expected spread, the 
large standard errors imply a low number of satellite/drifter 
differences in those bins. For the subset of HRSST drifters. 
Fig. 4b shows that the dispersion underestimates the spread, 
even for low standard error cases, meaning one assumption 
is incorrect in this case. 
With all other factors being equal, the distinction in the 
drifter type between Fig. 4a and b suggests the drifter uncer- 
(b) HRSST drifters 
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5 
SST uncertainty (K) SST uncertainty (K) 
(SLSTR 3-channel nighttime) (SLSTR 3-channel nighttime) 
Figure 4. SLSTR SST uncertainty validation plot for (a) all drifters 
and (b) a subset of HRSST-1 and HRSST-2 drifters, with uncer 
tainty bins of 0.01 K. An uncertainty of 0.20 K is assumed for the 
drifter SST. 
(a) All drifters 
Theoretical 
SD 
Median +/- SE 
. _ - " 
' ; 
--- 
tainty assumed (0.20 K) is inappropriate for the HRSST sub 
set. To verify this. Fig. 5 contains the same data as Fig. 4 but 
with the theoretical dispersion (green lines) calculated for a 
drifter uncertainty of 0.05 K. While the calculated dispersion 
does not track any more the expected spread for all drifters 
(Fig. 5a), the assumption of 0.05 K for the uncertainty of the 
HRSST drifter data gives a much better fit (Fig. 5b). This 
demonstrates the improved quality of HRSST drifter data for 
satellite SST validation. 
2.5 Influence of the drogue on drifter SST 
measurements 
This section investigates the effect of the sea anchor or 
drogue on drifter SST measurements. By exerting its own 
weight and by following currents centered at 15 m depth, the 
drogue pulls the float downwards via the tether. This main 
tains the float and its drogue aligned in the vertical, in wave 
troughs. When the drogue is lost, the float has more free 
dom to oscillate by roll and pitch, and the temperature probe 
can sometimes be exposed to waters closer to the surface.