MERCATOR OCEAN JOURNA:
SEPTEMBER 2021
Although the system described above for implementation
in 2022 includes an interactive ocean in the ensemble
part, it does not involve the generation of a realistic ocean
ensemble. To account for this, SST perturbations seen by
the atmosphere are generated, and applied, in exactly the
same way as when running atmosphere-only models.
However, in preparation for a later upgrade, a global
ocean and sea-ice ensemble system has been developed.
This is based on the ocean-only FOAM system with
ensemble members forced by different members of an
atmospheric ensemble. Ensemble spread is also driven
by perturbed observation locations and values (linking
with work in the GENOA service evolution R&D project),
stochastic model perturbations (which will benefit from
the NEMO code development in the SCRUM2 project) and
an ensemble inflation scheme. Results from tests of
hybrid ensemble/3DVar data assimilation (where the
NEMOVAR code is used to combine the existing
representation of background error covariances with a
localised estimate of the daily-varying sample error
zovariances coming from the ensemble) have been very
promising. Figures 3 and 4 show an example for sea level
anomaly assimilation statistics: the hybrid DA significantly
improves the deterministic model, and there are further
benefits from using the ensemble mean instead of the
unperturbed member.
sta_BiasCorrfb
107
a
106
1.05
O4
0.03
— <xn_ENOOO
— C_eN5mean
— en50a END
— da enSTeaN
02
01
(0) -
ı
' DZ
—
8
Sn |
a f
A
Ba
„r.
ze
a u.
En
Ba?
‚igure 3: Global sea level anomaly root-mean-square deviation (RMSD) compared with altimeter observations before their assimilation. The
black line shows the error in the standard 3DVar unperturbed member (equivalent to the current FOAM system). The blue line shows the error
in the ensemble mean using standard 3DVar. The red line shows the error in the hybrid DA unperturbed member and the green line shows the
arror in the ensemble mean using the hybrid DA.
