MERCATOR OCEAN JOURNA: 
SEPTEMBER 2021 
than relying on seasonal forecasts; this facilitated the 
extension of ocean forecasts to 10 days and the introduction 
of some higher frequency hourly diagnostics which were a 
strong user requirement and are now some of the most 
downloaded datasets. In this context ‘weakly coupled’ means 
the assimilation background states for both ocean and 
atmospheric components are from the coupled model. 
However, the data assimilation codes are run independently 
(without any opportunity for atmosphere observations to 
directly affect ocean increments on the same assimilation 
cycle) before increments are applied back into the coupled 
model with an incremental analysis update (IAU) step. As a 
consequence, the ocean data assimilation can be treated in a 
very similar way to the ocean-only FOAM system used 
previously. It continues to use NEMOVAR, a variational (3D-var) 
scheme developed specifically for NEMO and further tuned 
for the 1/4° resolution global model. Key features of NEMOVAR 
are multivariate relationships (specified through a linearized 
balance operator) and the use of an implieit diffusion operator 
to model background error correlations. 
The use of a weakly coupled data assimilation system 
necessitated a change from 6-hour to 24-hour assimilation 
windows in the ocean for consistency with the atmosphere. 
At present the ocean data assimilation has not been 
specifically tuned to account for this, although there would 
be potential benefit from such work in future. Multiple 
update cycles’ are used to mimic the previous behaviour 
where a ‘best analysis’ (assimilating as many ocean 
2bservations as possible) is made available a day later 
than the near-real-time analysis used to initialise the daily 
forecasts. The complete consistency between ocean 
analysis and forecast in the upgraded system benefits 
users, particularly those who are concerned with short 
lead-time forecasts. 
Significant upgrades in marine observations assimilated 
during Copernicus 1 were motivated both by newly 
available observations (with the potential to improve 
product quality and system robustness) and changes to the 
technical characteristics of observational products. In both 
zases careful testing is required to understand the 
technical and scientific Implications prior to fina 
assimilation. For the most part it was not possible to 
demonstrate a 'step change’ in product quality at the point 
assimilation of these observations was initially activated. 
However, subsequent degradation or discontinuity of other 
assimilated products means that, without the Sentinel-3 
satellite observations in particular, the GLO-CPL product 
Juality would have gradually degraded. 
— (CPLDA 
— FOAM 
- © DSYZ 
Global Ocean temperature all best estimate -36 
n N 
Da 
on, 
FF 
nn 
x 
D 
9% 
sp 
8 Jan 
?016 
an... Mar 
A—-. 
01 Jan 
2D17 
SL 
01 Jan 
2018 
WEN. a 
01 Jan 
2319 
Een Em 
01 Jan 
20720 
ET 
——— 
01 Jan 
2931 
—— RAMS£E 
/igure 1: Class 4‘ temperature profile statistics for the ‘best analysis‘ against Argo for the current GLO-CPL system (labelled as CPLDA:; 
also shown are the Met Office FOAM and Mercator Ocean PSY4 systems for comparison); bias and root-mean-square deviation (RMSD) are 
ıaveraced over depth: note that prior to July 2017 the GLO-CPL analvsis was provided from the FOAM svstem