QUOVADIS_NUM19_OND_2014

Quo Va Dis ? Quarterly Ocean Validation Display #19, OND 2014
April 2015
contact: qualif@mercator-ocean.fr
QuO Va Dis?
Quarterly Ocean Validation Display #19
Validation bulletin for October-November-December (OND) 2014
Edition:
Charly Régnier, Marie Drévillon, Bruno Levier, Charles Desportes, Coralie Perruche,
(MERCATOR OCEAN/Production Dep./Products Quality)
Contributions :
Eric Greiner (CLS)
Jean-Michel Lellouche, Olivier Le Galloudec (MERCATOR OCEAN/ Production Dep./R&D)
Credits for validation methodology and tools:
Eric Greiner, Mounir Benkiran, Nathalie Verbrugge, Hélène Etienne (CLS)
Fabrice Hernandez, Laurence Crosnier (MERCATOR OCEAN)
Nicolas Ferry, Gilles Garric, Jean-Michel Lellouche (MERCATOR OCEAN)
Stéphane Law Chune (Météo-France), Julien Paul (MERCATOR OCEAN), Lionel Zawadzki (CLS)
Jean-Marc Molines (LGGE), Sébastien Theeten (Ifremer), Mélanie Juza (IMEDEA), the DRAKKAR and
NEMO groups, the BCG group (Météo-France, CERFACS)
Bruno Blanke, Nicolas Grima, Rob Scott (LPO)
Information on input data:
Christine Boone, Gaël Nicolas (CLS/ARMOR team)
Abstract
This bulletin gives an estimate of the accuracy of MERCATOR OCEAN’s analyses and forecast
for the season of October-November-December 2014. It also provides a summary of useful
information on the context of the production for this period. Diagnostics will be displayed for
the global 1/12° (PSY4), global ¼° (PSY3), the Atlantic and Mediterranean zoom at 1/12°
(PSY2), and the Iberia-Biscay-Ireland (IBI) monitoring and forecasting systems currently
producing daily 3D temperature, salinity and current products. Surface Chlorophyll
concentrations from the BIOMER biogeochemical monitoring and forecasting system are also
displayed and compared with simultaneous observations.

Table of contents
I Executive summary ............................................................................................................4
II Status and evolutions of the systems ................................................................................5
II.1. Short description and current status of the systems..................................................5
II.2. Incidents in the course of OND 2014.........................................................................10
III Summary of the availability and quality control of the input data..................................10
III.1. Observations available for data assimilation.........................................................10
III.1.1. In situ observations of T/S profiles.....................................................................10
III.1.2. Sea Surface Temperature...................................................................................10
III.1.3. Sea level anomalies along track .........................................................................10
III.2. Observations available for validation ....................................................................11
IV Information on the large scale climatic conditions..........................................................11
V Accuracy of the products .................................................................................................13
V.1. Data assimilation performance .................................................................................13
V.1.1. Sea surface height..............................................................................................13
V.1.2. Sea surface temperature....................................................................................18
V.1.3. Temperature and salinity profiles......................................................................22
V.2. Accuracy of the daily average products with respect to observations.....................38
V.2.1. T/S profiles observations....................................................................................38
V.2.2. SST Comparisons ................................................................................................49
V.2.3. Drifting buoys velocity measurements ..............................................................50
V.2.4. Sea ice concentration.........................................................................................56
V.2.5. Closer to the coast with the IBI36V2 system: multiple comparisons ...............59
V.2.5.1. Comparisons with SST from CMS ...................................................................59
V.2.5.2. Comparisons with in situ data from EN3/ENSEMBLE for OND 2014 .............60
V.2.5.3. MLD Comparisons with in situ data................................................................62
V.2.5.4. Comparisons with moorings and tide gauges ................................................63
V.2.6. Biogeochemistry validation: ocean colour maps...............................................66
VI Forecast error statistics....................................................................................................68
VI.1. General considerations..........................................................................................68
VI.2. Forecast accuracy: comparisons with T and S observations when and where
available ...............................................................................................................................68
VI.2.1. North Atlantic region......................................................................................68
VI.2.2. Mediterranean Sea.........................................................................................70
VI.2.3. Tropical Oceans and global.............................................................................71
VI.3. Forecast accuracy: skill scores for T and S.............................................................74
VI.4. Forecast verification: comparison with analysis everywhere ...............................76
VII Monitoring of ocean and sea ice physics.........................................................................77
VII.1. Global mean SST and SSS.......................................................................................77
VII.2. Surface EKE.............................................................................................................78
VII.3. Mediterranean outflow .........................................................................................79
VII.4. Sea Ice extent and area..........................................................................................79
I Annex A ............................................................................................................................81
I.1. Table of figures..........................................................................................................81

II Annex B.............................................................................................................................85
II.1. Maps of regions for data assimilation statistics........................................................85
II.1.1. Tropical and North Atlantic................................................................................85
I.1.1. Mediterranean Sea.............................................................................................86
I.1.2. Global ocean.......................................................................................................87
III Annex C.............................................................................................................................88
III.1. Quality control algorithm for the Mercator Ocean drifter data correction (Eric Greiner)
88
III.2. Algorithm of the Lagrangian verification of the Mercator Ocean surface currents
forecast.....................................................................................................................................89

I Executive summary
T & S
The Mercator Ocean monitoring and forecasting systems are evaluated for the period
October-November-December 2014. The 1/12° global system providing products for
MyOcean V3 global MFC, and the other Mercator “off shore” systems (global ¼° and
Atlantic+Mediterranean 1/12°) display similar performance in terms of water masses
accuracy. The system’s analysis of the ocean water masses is very accurate on global
average and almost everywhere between the bottom and 200m. Between 0 and 500m
departures from in situ observations rarely exceed 1 °C and 0.2 psu (mostly in high
variability regions like the Gulf Stream or the Eastern Tropical Pacific). At the global scale
PSY4 is fresh (up to 0.02) at the surface and warm in the whole water column (below 0.05°C)
whereas PSY3 is salty (up to 0.01) in the 0-300m layer and globally cold (below to 0.05). PSY3
seems to be close to PSY4 but most accurate in the Pacific and Atlantic basins. PSY4 is the
best in the Indian basin. The evolution of this difference between the global systems will be
monitored in the next issues.
The temperature and salinity forecast have significant skill in many regions of the ocean in
the 0-500m layer, but the signal is noisy.
Surface fields: SST, SSS, SSH, currents
A warm SST bias of 0.1 °C on global average is diagnosed this quarter, mainly due to the
tropics and the ACC. The SST bias is cold in the Arctic.
Especially in the high resolution systems PSY2 and PSY4, a strong fresh bias is diagnosed in
the tropics, in link with the overestimated convective precipitations in the ECMWF
atmospheric fields.
The monitoring systems are generally very close to altimetric observations (global average of
6 cm residual RMS error). Biases persist locally that correspond to known uncertainties in
the mean dynamic topography (for instance in the Indonesian region).
For the high resolution system PSY4, the surface currents are overestimated in the mid-
latitudes and for the Equatorial Pacific currents (South Equatorial Current and North
Equatorial Counter Current) with respect to in situ measurements of drifting buoys (drifter
velocities are corrected of windage and slippage with a method developed by Mercator
Ocean). He also overestimates the Kuroshio region, the Gulf of Alaska and the Southern
oceans. PSY3 underestimates the South Pacific Equatorial Current, the Western Subarctic
gyre and the South Indian Ocean. Both overestimate the Southern oceans and the Gulf
Stream region.
The underestimation ranges from 20% in strong currents up to 60% in weak currents. On the
contrary the orientation of the current vectors is well represented. Lagrangian metrics show
that after 1 day, 80% of the virtual drifters trajectories performed with Mercator Ocean
forecast velocities stay within a 25 km distance of the actual drifters observed trajectories.
Regional North East Atlantic

The high resolution North East Atlantic at 1/36° (IBI36V1) with no data assimilation is
accurate on average. Tidal and residual sea surface elevations are well represented. Zones
of intense tidal mixing are less accurate. The mixed layer is too shallow in the Bay of Biscay
(the thermocline is too diffusive). The upwelling along the Iberian coasts is underestimated.
Sea Ice
Sea ice observations are not yet assimilated, nevertheless sea ice concentrations are
realistic. In OND 2014, there is an underestimation of modelled sea ice concentration with
observed sea ice concentration near the marginal zones and an overestimation in the
Canadian Archipelago. This could be partly explains by to much melting during the summer
and a bad resolved sea ice circulation. In the Antarctic the summer sea ice concentration is
underestimated on average by both PSY3 and PSY4, On the contrary it is overestimated
along the sea ice edge in the Bellingshausen Sea and in the south of Kerguelen Island.
Biogeochemistry
The new version of BIOMER4 seems to be closer to the observations, the serious
discrepancies in the tropical band are corrected. Due to a bad timing of the blooms, some
discrepancies are still present in the Peru-Chili coast and in the Ziapola region.
II Status and evolutions of the systems
II.1. Short description and current status of the systems
A bit of history
PSY3V3 (global ¼°) and PSY2V4 (Altantic and Mediterranean 1/12°) systems have been
operated at MERCATOR OCEAN since 2010 December, 15th
. These systems provide the
version 1 (PSY3V3R1/PSY2V4R1, see QuOVaDis? #2) and version 2 (PSY3V3R1/PSY2V4R2, see
QuOVaDis? #5) products of the MyOcean global monitoring and forecasting centre. As
reminded in table 1 (and illustrated for PSY2V2 in Figure 1) the atmospheric forcing is
updated daily with the latest ECMWF analysis and forecast, and a new oceanic forecast is
run every day for both PSY3V3R1 and PSY2V4R2.
The PSY3V3R1 system is started in October 2006 from a 3D climatology of temperature and
salinity (World Ocean Atlas Levitus 2005) while the PSY2V4R2 is started in October 2009.
After a short 3-month spin up of the model and data assimilation, the performance of
PSY3V3R1 has been evaluated on the 2007-2009 period (MyOcean internal calibration
report, which results are synthesised in QuOVaDis? #2).
The PSY4 system (global 1/12°) is delivering operational products since the beginning of
2010, and was developed in 2009. Its first version PSY4V1 did not benefit from the scientific
improvements of PSY3V3R1 and PSY2V4R2, developed in 2010 and 2011. This system
delivers 7-day forecast (and not 14-day like PSY3V3R1 and PSY2V4R2).
An upgrade of the data assimilation systems was performed in March 2012 in order to
assimilate MyOcean V2 altimetric observations and in situ observations (instead of
respectively AVISO and CORIOLIS observations, corresponding to MyOcean V0 observations).

In consequence, more in situ observations are assimilated by all systems in the European
seas since March 2012.
Current global systems
The whole Mercator Ocean global analysis and forecasting system (including PSY4, PSY2 and
PSY3) has been updated in April 2013 (MyOcean products version 3). The current versions
names are PSY3V3R3, PSY2V4R4 and PSY4V2R2. Results from these current systems only
will be described in this issue. For more simplicity they will be called PSY3, PSY2 and PSY4
in this document. A description of most updates, as well as the evaluation process, can be
found in Lellouche et al (2013)1
. With respect to this article, several additional modifications
were made in order to stabilize the performance of the system (see table 1). A specific
paragraph is dedicated to the evaluation of these new systems: in Quo Va Dis?#12.
North east Atlantic
The IBI36 system is described in QuOVaDis? #5 and #6 (see also table 1 and Figure 1). The
nominal MyOcean production unit for IBI36 is Puertos Del Estado (Spain) while Mercator
Océan produces the back up products. The Mercator Océan IBI36V1 system was officially
operational in June 2011. The version IBI36V2 of the system was operated from December
2011 to April 2014 and was very similar to IBI36V1 except it used realistic river runoffs from
SHMI and Prévimer instead of climatological runoffs. The version IBI36V4 of the system is
operated since April 2014: the only difference with IBI36V2 is the switch of the NEMO code
version, from 2.3 to 3.4. There are however minor code upgrades. The leapfrog time
stepping scheme of 3d variables has been modified according to the work of Leclair and
Madec (2009)2
; this should have a small impact on the results. The computation of coriolis
terms in the barotropic equations has been corrected in order to be fully compliant with the
3d calculation. The locally enhanced bottom friction in the Gulf of Cadiz implemented in IBI-
V3 has been suppressed, since this leads to a relatively minor improvement of
Mediterranean overflow.
1
J.-M. Lellouche, O. Le Galloudec, M. Drévillon, C. Régnier, E. Greiner, G. Garric, N. Ferry, C.
Desportes, C.-E. Testut, C. Bricaud, R. Bourdallé-Badie, B. Tranchant, M. Benkiran, Y. Drillet,
A. Daudin, and C. De Nicola, Evaluation of global monitoring and forecasting systems at
Mercator Océan, Ocean Sci., 9, 57-81, 2013, www.ocean-sci.net/9/57/2013/,
doi:10.5194/os-9-57-2013
2
M. Leclerc, G. Madec, A conservative Leap-Frog time stepping method, Ocean Modelling
30(2-3), 88-94, 2009, http://dx.doi.org/10.1016/j.ocemod.2009.06.006

Figure 1: schematic of the operational forecast scenario for IBI36Q (green) and PSY2Q (blue). Solid lines are
the PSY2 weekly hindcast and nowcast experiments, and the IBI36 spin up. Dotted lines are the weekly 14-
day forecast, dashed lines are daily updates of the ocean forecast forced with the latest ECMWF atmospheric
analysis and forecast. The operational scenario of PSY3, PSY4 and PSY3Q, PSY4Q is similar to PSY2’s scenario.
In the case of PSY4, only weekly hindcast, nowcast and 7-day forecast are performed.
Biogeochemistry
The BIOMER4 system is described in QuOVaDis? #18 (see also table 1). It is a global hindcast
biogeochemical model forced by physical ocean fields. The biogeochemical model used is
PISCES. The coupling between ocean physics and biogeochemistry is performed offline. The
physical fields from PSY3 have a 1/4° horizontal resolution and 7-day time resolution.

Table 1: Main characteristics and latest updates of the Mercator Ocean global analysis and forecasting systems. The systems studied in Lellouche et al (2013) include the main characteristics (in
black) plus the updates in blue. The 2013 systems (in red) include the main characteristics (in black) plus the updates in blue and red. In the legend below one can find a description of the updates
referred to as “mix”, “colour”, etc…
System name domain resolution Physical Model Assimilation Assimilated observations Inter dependencies Status of production
PSY4V1R3
(operational in OND
2013)
PSY4V2R2
(operational in AMJ
2013)
Global 1/12° on the
horizontal,
50 levels on
the vertical
ORCA12 LIM2 NEMO 1.09
Bulk CLIO
24-h atmospheric forcing
LIM2 EVP NEMO 3.1
Bulk CORE
mix, colour, iceberg, EMP
init WOA09: oct 2012
SAM2V1 (SEEK) + IAU
3D-Var bias correction
coast error, shelf error
new MDT, radii
Increase of Envisat error
new QC, SST bulk corr
RTG-SST, MyOcean SLA
along track,
MyOcean T/S vertical
profiles
AVHRR-AMSR SST,
new MDT
Sea Mammals T/S
profiles in CUTRR
Black Sea SLA files
Weekly 7-day
forecast
Weekly 7-day
forecast
Daily update of
atmospheric forcing
for daily 7-day
forecast
PSY3V3R1
(operational in OND
2013)
PSY3V3R2 (described
in Lellouche et al,
2013)
PSY3V3R3
(operational in AMJ
2013)
Global 1/4° on the
horizontal, 50
levels on the
vertical
ORCA025 LIM2 EVP NEMO
3.1
Bulk CORE
mix, colour, iceberg, EMP
flux corr
no flux corr
current in wind
init WOA09:oct 2006
SAM2V1 (SEEK) + IAU 3D-
Var bias correction
MDT error adjusted
first update of radii
new QC
radii, SST bulk corr
RTG-SST, MyOcean SLA
along track,
MyOcean T/S vertical
profiles
AVHRR-AMSR SST,
new MDT
Sea Mammals T/S
profiles in CUTRR
Black Sea SLA files
Weekly 14-day
forecast
Daily update of
atmospheric forcing
for daily 7-day
forecast
PSY2V4R2
(operational in OND
2013)
PSY2V4R3 (described
in Lellouche et al.,
2013)
PSY2V4R4
(operational in AMJ
2013)
Tropical
North Atlantic
Mediterranean
1/12° on the
horizontal,
50 levels on
the vertical
NATL12 LIM2 EVP NEMO 3.1
Bulk CORE
mix, colour
flux corr
no flux corr
current in wind
init WOA09:oct 2006
SAM2V1 (SEEK) + IAU 3D-
3D-Var bias correction
first update of radii
QC on T/S vertical profiles
radii, SST bulk corr
Larger weight of Bogus OBC
on TSUV
AVHRR-AMSR SST,
MyOcean SLA along
track , MyOcean T/S
vertical profiles
new MDT
Sea Mammals T/S
profiles in CUTRR
OBC from
PSY3V3R1
OBC and SMEMP
from PSY3V3R2
OBC and SMEMP
from PSY3V3R3
Weekly 14-day
forecast
Daily update of
atmospheric forcing
for daily 7-day
forecast
BIOMER4
upgrade in
September 2014
Global 1/4° on the
horizontal, 50
levels on the
vertical
PISCES, NEMO 3.2, offline none none Two weeks
hindcast, nowcast
with
PSY3V3R3 1/4° phy
Weekly 14-day
analysis
Weekly 7-day
forecast
IBI36V4
upgrade in AMJ 2014
North East
Atlantic and
West
Mediterranean
Sea (Iberian,
Biscay and
Ireland) region
1/36° on the
horizontal, 50
levels on the
vertical
NEATL36 NEMO 3.4 3-hourly
atmospheric forcing from
ECMWF, bulk CORE, tides,
time-splitting, GLS vertical
mixing, corrected
bathymetry, river runoffs
from SMHI & Prévimer
none none Two weeks spin up
initialized with
PSY2V4R4 and OBC
from PSY2V4R4
Weekly spin up two
weeks back in time.
Daily update of
atmospheric
forcings for daily 5-
day forecast
IBI36QV4

Mix = New parameterization of vertical mixing
Colour = Taking into account ocean colour monthly climatology for depth of light
extinction
Current in wind = taking 50 % of surface current for the computation of wind
stress with bulk CORE
EMP = Adding seasonal cycle for surface mass budget
SMEMP = spatial mean EMP correction
Iceberg = Adding runoff for iceberg melting
Flux corr = Large scale correction to the downward radiative and precipitation
fluxes
Coast error = Observation error s higher near the coast (SST and SLA)
Shelf error = Observation error s higher on continental shelves (SLA)
New MDT = MDT CNES/CLS09 adjusted with model solutions (bias corrected)
Radii = New correlation radii (minimum =130km)
New QC = additional QC on T/S vertical profiles computed from the innovations
SST bulk corr = Procedure to avoid the damping of SST increments via the bulk
forcing function
OBC = Open Boundary Conditions
1° phy= physical forcings are “degraded” from ¼° horizontal resolution to 1°
horizontal resolution, and weekly averaged.
CUTRR= Catch Up to Real Time Run, or calibration hindcast run (before 2013)
Init WOA09= initialisation with World Ocean Atlas climatology

10
II.2. Incidents in the course of OND 2014
No special event to report for this quarter.
III Summary of the availability and quality control of the input data
III.1. Observations available for data assimilation
III.1.1. In situ observations of T/S profiles
System PSY3V3R3 PSY4V2R2 PSY2V4R4
Min/max number of T
profiles per DA cycle
2600/3200 2600/3500 500/800
Min/max number of S
profiles per DA cycle
2000/2500 2000/2500 400/600
Table 2: minimum and maximum number of observations (orders of magnitude of vertical profiles) of subsurface temperature
and salinity assimilated weekly in OND 2014 by the Mercator Ocean monitoring and forecasting systems.
The maximum number of in situ observations is displayed in Table 2. It is unchanged with respect to the
previous quarter.
III.1.2. Sea Surface Temperature
System PSY3V3R3 PSY4V2R2 PSY2V4R42
Min/max number (in 103
)
of SST observations
135/144 137/146 26/27
Table 3: minimum and maximum number (orders of magnitude in thousands) of SST observations (from Reynolds AVHRR ¼°)
assimilated weekly in OND 2014 by the Mercator Ocean monitoring and forecasting systems.
In the Atlantic (PSY2) the number of assimilated SST observations decreases with the increase of Arctic sea ice
cover from January to March.
III.1.3. Sea level anomalies along track
As shown in Table 4 the data assimilated this OND 2014 season come from AltiKa, Jason 2, Cryosat 2 and HY-
2A.
system PSY3V3R3 PSY4V2R2 PSY2V4R4
Min/max number (in 103) of
Jason 2 SLA observations
161/174 161/175 29/31
Min/max number (in 103) of HY-
2A SLA observations
45/154 45/155 8/29
AltiKa SLA observations
112/161 113/162 24/31
Cryosat 2 SLA observations
141/163 142/165 27/32
Table 4: minimum and maximum number (orders of magnitude in thousands) of SLA observations from AltiKa, Jason 2, Cryosat 2
and HY-2A assimilated weekly in OND 2014 by the Mercator Ocean monitoring and forecasting systems.

11
III.2. Observations available for validation
Both observational data and statistical combinations of observations are used for the real time validation of
the products:
• T/S profiles from MyOcean
• OSTIA SST from MyOcean
• Arctic sea ice concentration and drift from CERSAT
• SURCOUF surface currents from CLS
• ARMOR-3D 3D temperature and salinity fields from CLS
• Drifters velocities from Météo-France reprocessed by CLS
• Tide gauges from CLS
• Chlorophyll-a concentrations daily L3 and monthly L4 maps from MyOcean
All products were available in real time during the OND 2014 season, except tropical moorings from 29/11 to
09/12 and from 23/12 to 07/01/2015. All profiles have been retrieved then and the delays had no significant
impact on the Mercator Ocean production and validation procedure.
Grodsky et al (GRL, May 2011) show that drifters’ velocities overestimate current velocities in regions and
periods of strong winds due to undetected undrogued drifters. This information is taken into account for
comparisons with Mercator Ocean currents.
IV Information on the large scale climatic conditions
Mercator Ocean participates in the monthly seasonal forecast expertise at Météo France. This chapter
summarizes the state of the ocean and atmosphere during the OND 2014 season, as discussed in the “Bulletin
Climatique Global” of Météo France.
a)

12
b)
Figure 2: Seasonal OND 2014 temperature anomalies with respect to GLORYS2V3 climatology (1993-2011). (a): SST anomaly (°C)
at the global scale from the 1/4° ocean monitoring and forecasting system PSY3V3R3. (b): heat content anomaly (ρ0Cp∆T, with
constant ρ0=1020 kg/m3 ) from the surface to 300m.
In the tropical Pacific, the ENSO conditions were neutral. The equatorial pacific is slightly warmer than the
GLORYS2V3 climatology, as well as most of the Tropical Indian Ocean and the Indonesian throughflow. As can
be seen in Figure 3, a downwelling Kelvin wave has propagated in subsurface, and temperature anomalies up
to 3°C can be diagnosed on average over OND 2014 near 150m in the central Tropical pacific.
Figure 3: Seasonal OND 2014 temperature anomaly (°C) with respect to GLORYS2V3 climatology (1993-2011), vertical section 2°S-
2°N mean, Pacific Ocean, PSY3V3R3.

13
The cold SST anomaly in the North Pacific surrounded by warm anomalies along the west coast of the US is
linked with the persistent positive phase of the Pacific Decadal Oscillation3
in 2014. In the Norway Sea the
persisting Scandinavian blocking as given rise to warm SST anomalies. In December the positive phase of the
North Atlantic Oscillation dominates the atmospheric circulation, strengthening the warm anomaly in the
subtropical gyre, and the cold anomaly in the centre of the basin. Note a cold SST anomaly in the outflow of
the Mozambique Channel that can be due to the late start of the South African rainy season in December.
Figure 4 : Arctic sea ice extent from the NSIDC, available at http://nsidc.org/arcticseaicenews
As can be seen in Figure 4, the Arctic sea ice extent in OND 2014 is within the envelope of the last 30 years
records of sea ice extent.
V Accuracy of the products
V.1. Data assimilation performance
V.1.1. Sea surface height
V.1.1.1. Basin scale statistics (GODAE metrics)
3
http://www.climate.gov/news-features/understanding-climate/2013-state-climate-sea-surface-temperature

14
a) b)
c)
Figure 5 : Comparison between SLA data assimilation scores (a: average misfit in cm, b: RMS misfit in cm, c: number of
observations) for all available Mercator Ocean systems in OND 2014. The scores are averaged for all available satellite along track
data (Saral/altiKa, Jason 2, Cryosat 2). For each region the bars refer to PSY2V4R4 (yellow), PSY3V3R3 (red), PSY4V2R2 (blue).
SLA assimilation scores for PSY4, PSY3, and PSY2 in OND 2014 are displayed in Figure 5. The different systems
reach similar levels of performance on average. The basin scale SLA bias does not exceed 2 cm except in the
Arctic in PSY4 where it reaches 3 cm (Figure 5 a), and in the Mediterranean Sea where PSY4 is slightly greater
than 2cm and PSY3 with PSY2 reach 5cm . The Arctic SSH undergoes large uncertainties (MSSH, measurement
errors, sea ice limit). In the Mediterranean Sea this quarter, the high resolution 1/12° global system (PSY4) is
less biased than the others. PSY4 is less biased than the other systems in the Atlantic. The three systems have
consistent RMS errors. The RMS error (Figure 5 b) is of the order of 9 cm in the Mediterranean Sea and in the
Arctic, 8cm in the Atlantic and in the Southern Oceans. It is of the order of 5-7 cm, the Indian and Pacific
oceans, and it is close to 4-5 cm in the Tropical Atlantic and Tropical Pacific oceans. The error amplitude is
proportional to the region’s internal variability.
V.1.1.2. North Atlantic Ocean and Mediterranean Sea in all systems

15
The SLA assimilation scores for PSY4, PSY3, and PSY2 in OND 2014 are displayed for sub-regions of the Tropical
and North Atlantic Ocean in Figure 6. The different systems reach identical levels of performance on average:
except for three regions (Iceland, North Madeira and Yoyo Pomme) the biases are less than 2 cm during this
season. The largest biases (4cm) appear in PSY4 and PSY2, in small regions assimilating few data. Note that
consistently with Figure 5 results. Part of the biases can be attributed to local errors in the current mean
dynamical topography (MDT).
a) b)
c)
Figure 6: Comparison between SLA data assimilation scores (a: average misfit in cm, b: RMS misfit in cm, c: number of
observations) for all available Mercator Ocean systems in OND 2014 in sub-regions of the Tropical Atlantic (TAT) and North
Atlantic (NAT). The scores are averaged for all available satellite along track data (Jason 1 G, Jason 2, Cryosat 2). For each region
the bars refer respectively to PSY2V4R4 (yellow), PSY3V3R3 (red), and PSY4V2R2 (blue).
The RMS errors are almost identical in all systems, and stay below 10 cm in most regions, except regions of
high mesoscale variability (Gulf Stream, Gulf of Mexico). This OND 2014 season, the RMS error in the Gulf
Stream 1 region is larger in PSY2 and lower in PSY4. PSY4 biases are small but positive in many regions,
meaning that PSY4 is slightly lower than the observations over these regions. PSY2 exhibits positive biases on
average where PSY3 seems to be more balanced than the previous quarter (consistently with Figure 5).

16
In the eastern Mediterranean Sea (Figure 7), significant positive biases (the model is lower than the
observations) appear this autumn. PSY2 displays biases of 4 to 11 cm, while the global systems PSY3 and PSY4
present smaller biases of 1 to 8 cm. A bias of 11 cm is present in PSY2 in the Aegean Sea (it was nearly 10 cm
last quarter). The Mediterranean sub-regions are circled by coasts, and consequently few observations are
assimilated. Figure 7 c gives the number of observations taken into account by the system, but the
observation error that is applied reduces strongly the influence of observations within a 50 km distance of the
coast.
a) b)
c)
observations) for all available Mercator Ocean systems in OND 2014 in the Mediterranean Sea MED sub-regions: PSY2V4R4
(yellow), PSY3V3R3 (red) and PSY4V2R2 (blue). The scores are averaged for all available satellite along track data (Jason 1 G,
Jason 2, Cryosat 2). See annex B for geographical location of regions.
One can also note that the variability in the Aegean Sea is difficult to model because of the presence of many
islands. PSY4 may perform better in this region due to the presence of the Black sea in the global
configuration PSY4, which is not present in PSY2.
Except in the Aegean Sea, and in the Alboran Sea where very few observations are available, the RMS of the
innovation (misfit) is generally less than 7 cm for all systems in OND 2014.

17
V.1.1.3. Performance at global scale in PSY3 (1/4°) and PSY4 (1/12°)
a) b)
c)
observations) for all available global Mercator Ocean systems in OND 2014 in sub-regions of the SAT, IND, NPA, SPA, TPA, and
ACC basins: PSY3V3R3 (red) and PSY4V2R2 (blue). The scores are averaged for all available along track satellite data (Jason 1 G,
Jason 2, Cryosat 2). The geographical location of regions is displayed in annex B.
As can be seen in Figure 8 the performance of intermediate resolution global PSY3 and the performance of
high resolution global PSY4 in terms of SLA assimilation are of the same order of magnitude. A small positive
bias (less than 3 cm) is present in many regions in PSY4. A strong bias of around 10 cm persists in the “Nino 5”
box centred on the Banda Sea in Indonesia, which corresponds to a MDT problem. Except the Nino 5 region,
the RMS error reaches its highest values in the Agulhas current where the variability is high.
Other regions with high variability, especially in the ACC region but also in eastern Australia, consistently
display RMS errors of more than 8 cm.

18
V.1.2. Sea surface temperature
V.1.2.1. Basin scale statistics (GODAE metrics)
a) b)
c)
Figure 9: Comparison between Reynolds ¼°AVHRR-data assimilation scores (a: average misfit in °C, b: RMS misfit in °C, c: number
of observations) for all available Mercator Ocean systems in OND 2014: PSY4V2R2 (blue), PSY3V3R3 (red), and PSY2V4R4
(yellow).
This OND season, a warm bias of 0.1 to 0.2 °C persists in all systems on average over the Tropical Atlantic and
Pacific (Figure 9). On the contrary, the Arctic Ocean is colder (0.2 °C) than the observations. It is known that
not enough Atlantic (warm) waters enter the Arctic in both PSY3 and PSY4 and during the Summer season
there is too much sea ice melting in the centre of the Arctic (see QUOVADIS 18).
The Mediterranean is warmer than the observations on average over the basin (0.1°C). The RMS error is lower
than 0.7 °C except in the Arctic Ocean due to the strong biases in marginal zones.

19
V.1.2.2. North and Tropical Atlantic Ocean and Mediterranean Sea
a) b)
c)
Figure 10: Comparison between Reynolds ¼°AVHRR SST data assimilation scores (a: average misfit in °C, b: RMS misfit in °C, c:
number of observations) for all available Mercator Ocean systems in OND 2014 in sub-regions of the Tropical Atlantic (TAT) and
of the North Atlantic (NAT): PSY4V2R2 (blue), PSY3V3R3 (red) and PSY2V4R4 (yellow). The geographical location of regions is
displayed in annex B.
In the Atlantic the three systems display consistent regional behaviours in terms of SST bias as illustrated in
Figure 10. In most regions the bias stays below 0.3 °C. A warm bias of around 0.3 to 0.5°C is diagnosed in the
Tropical Atlantic in the Dakar region due to an underestimation of the upwelling by all systems. PSY2 display
strong warm biases in Newfoundland and Gulf Stream region. Note that as for SLA, prescribed SST errors are
higher within 50km off the coast. The RMS error reaches more than 1.5°C in the Gulf Stream regions, and in
the Newfoundland Iceland region where it reaches 1.2°C this season.

20
a) b)
c)
Figure 11: Comparison between Reynolds ¼° AVHRR SST data assimilation scores (a: average misfit in °C, b: RMS misfit in °C, c:
number of observations) for all available Mercator Ocean systems in OND 2014 in the Mediterranean Sea: PSY4V2R2 (blue),
PSY3V3R3 (red) and PSY2V4R4 (yellow). The geographical location of regions is displayed in annex B.
This season, the Mediterranean regions are more strongly biased in Mersa Matruh and Sicily regions (up to
0.4°C Figure 11). In the Aegan and Ionean Sea PSY3 and PSY4 display opposite bias (the Black Sea is present in
PSY4 and not in PSY3 which can explain this result). The RMS error is generally of the order of 0.5-0.8°. As in
SLA, the performance of the systems is lower in the Aegean Sea. Errors are also higher in the Alboran Sea,
Aegean and Siciliy regions but one can note that not many observations are available in this small region. (see
Figure 11). The low rms error in the Ionian region is associated with a high number of observations.

21
V.1.2.3. Performance at global scale in PSY3 (1/4°) and PSY4 (1/12°)
a)
b)
c)
Figure 12: Comparison between Reynolds ¼° AVHRR SST data assimilation scores (a: average misfit in °C, b: RMS misfit in °C, c:
number of observations) for all available global Mercator Ocean systems in OND 2014 in the SAT, NPA, TPA, SPA, IND and ACC
basins: PSY3V3R3 (red) and PSY4V2R2 (blue). See annex B for geographical location of regions.
The global products exhibit a warm surface bias in most regions this OND 2014 season (Figure 12), of
magnitude 0.1°C to 0.4°C. This warm bias is present in (small) coastal regions: in the cold waters of the
California current, in the Peru upwelling or in the Angola upwelling but also in the Pacific Ocean especially in
the Tropical Pacific Ocean (Niño boxes 5,6, 1-2). The strongest warm bias is diagnosed in the Nino 1-2 box
probably with very few data near the coast. A cold bias (0.1°C) is diagnosed in the North Pacific region and in
the Eastern Australia.
PSY3 and PSY4 exhibit the same levels of performance but the warm bias is generally stronger in PSY4 than in
PSY3 by 0.05°C, and so is the RMS error, meaning that the variability is at least as well represented in PSY4
than in PSY3. The RMS error stands between 0.8 °C and 1°C in the tropical Pacific and in the Agulhas current
and the Angola region. It stands between 0.4°C and 0.7°C in the other regions.

22
V.1.3. Temperature and salinity profiles
V.1.3.1. Methodology
All systems innovation (observation – model first guess) profiles from the hindcast or “best analysis” week (J-
14 to J-7) are systematically inter-compared in all regions given in annex B. In the following, intercomparison
results are shown on the main regions of interest for Mercator Ocean users in OND 2014. Some more regions
are shown when interesting differences take place, or when the regional statistics illustrate the large scale
behaviour of the systems.
V.1.3.1.1. Basin scale statistics (GODAE metrics)
a) b)

23
c) d)
e) f)
Figure 13: Profiles of OND 2014 innovations of temperature (°C, left column) and salinity (psu, right column), mean (solid line)
and RMS (dashed line) for PSY3V3R3 (red), PSY2V4R4 (yellow) and PSY4V1R3 (blue) in the North Atlantic NAT (a and b),
Mediterranean Sea MED (c and d) and Arctic Ocean ARC (e an fl, the region starts of north of 67°N). Basin masks are applied to
keep only the main basin of interest (no Mediterranean in the Altantic NAT, no Black Sea in the Mediterranean MED, etc…).

24
On average over the North Atlantic, PSY4 and PSY2 display a warm bias from 50 m to 400 m that is not
present in PSY3 (Figure 13). In the meantime, a slight salty bias of around 0.02 psu is diagnosed in PSY2 and
PSY4 near 100m.
On average over the Mediterranean Sea, the three systems display a warm bias around 100m, and a slight
cold bias between 500 and 800m, they appear too fresh between 50 and 200m. PSY2 displays the best
performance in salinity and PSY4 in temperature.
In the Arctic, a cold bias is diagnosed in both PSY3 and PSY4 near the surface up to 100m (0.2°C), and then we
have a warm bias between 150 and 400m (0.1°C) and again a cold bias between 500 and 900m (0.2°C). A salty
bias of around 0.2 psu can be found at the surface and decreases with depth until 200m. Theses error profiles
show a bad stratification of the systems in this region.
On average over the Tropical Atlantic basin PSY4 and PSY2 exhibit warm (0.3°C) and salty (around 0.05 psu)
biases in the 50-300m layer (Figure 14). No significant temperature or salinity biases appear underneath the
surface layer. PSY4 (closely followed by PSY3) exhibits the best performance in this region. For PSY3, a small
cold bias appears near the surface (0.1°C).
PSY3 is less biased than PSY4 in the South Atlantic Basin, where PSY4 is too warm in the 50-900 m layer. PSY3
is too salty near the surface and PSY4 is too salty in the subsurface near 200m.
In the Indian Ocean, PSY3 exhibits a cold bias (0.2°C near 100m and then 0.1 °C between 150 and 500m). PSY4
also exhibits just a small cold bias in the surface layer (0-100m). On average the subsurface layer (up to 200m)
is slightly too salty for PSY4 and slightly too salty for PSY3 near the surface (0.02 psu). PSY4 exhibits the best
performance in this region.
a) b)

25
c) d)
e) f)
Figure 14: As Figure 13 in the Tropical Atlantic TAT (a and b), South Atlantic SAT (c and d) and Indian IND (e and f) basins.

26
On average over the Pacific Ocean and the Tropical Pacific (Figure 15), PSY4 is warmer (0.1 to 0.2 °C) than the
observations (and than PSY3) from the surface to 800 m. On the contrary PSY3 is close to the observations and
exhibits a small cold bias of around 0.1°C near 200m.
a) b)
c) d)

27
e) f)
Figure 15: As Figure 13 in the North Pacific NPA (a and b), Tropical Pacific TPA (c and d) and South Pacific SPA (e and f).
For the South Pacific Region, in the 100-500m layer PSY4 exhibits a warm bias and PSY3 a cold bias as the
same order (0.1 to 0.2°C).
For the three regions in Figure 15 these systems have the same salinity bias vertical structure with small
amplitude (less than 0.05 psu, with a maximum fresh bias at the surface). The temperature and salinity RMS
errors of PSY3 and PSY4 are identical while the bias is stronger in PSY4 which means that the variability is
probably better represented in PSY4 than in PSY3.
In the Southern Ocean in Figure 16, a warm temperature bias is observed again in PSY4 (up to 0.2 °C) in the
100-1000m layer. PSY3 exhibits the best performance in this region with just a small warm bias (0.1 °C)
between 100 and 200m. For the two systems the salinity is very close to the observations.

28
a) b)
Figure 16: As Figure 13 in the ACC Southern Ocean basin (from 35°S).
V.1.3.1.2. Atlantic sub-regions
a) b)

29
c) d)
Figure 17: As Figure 13 in the North East Atlantic (IBI) and the North Madeira XBT regions.
As shown in Figure 17, in the North East Atlantic Ocean the RMS error appears very noisy and the high
resolution systems are a little bit too warm (0.2 to 0.1°C) in the 100-500m layer.
For the high resolution systems PSY2 and PSY4 the signature of the Mediterranean waters is clear in the T and
S RMS error (Figure 17 c and d). This quarter is characterized by the warm bias in both high resolution systems
PSY2 and PSY4 in the 50-500 m layer. PSY3 on the contrary, exhibits a smaller cold bias in the same layer. All
systems exhibit a warm bias near 1000m. PSY4 and PSY3 (in the North Madeira region) are more biased than
PSY2 this OND season. In the Mediterranean water outflow the RMS error of PSY4 is higher than that of the
other systems, which may be due to the ring off of Portugal (see description in previous QUOVADIS 17).
In sub-regions of the Tropical Atlantic (Figure 18), and in the Gulf Stream (Figure 19) the vertical structure of
the biases is more complex. It reflects the underestimation of the upwelling in the Dakar region, and the
errors in the vertical structure of the currents. In the Dakar region PSY3 exhibits the best performance I
temperature for this OND quarter.
In the Gulf of Mexico the systems are too cold in the 0-500m layer and too fresh near 200m.
In the Gulf Stream, all systems are too warm (up to 1°C near 200m) and salty in the 0-800m layer.
Underneath PSY4 is too cold and fresh until 1500m, while PSY4 is too cold from 800 to 1500m. PSY2 is the
worst system in this region with a RMS error growing up to 1°C.

30
a) b)
Figure 18: As Figure 13 in the Dakar region.
a) b)

31
c) d)
e) f)
Figure 19: As Figure 13 in the Ascension tide (a and b), the Gulf of Mexico (c and d) and the Gulf Stream 1 regions (e and f)
However, one must keep in mind that the smallest sub-regions contain only a few data per assimilation cycle
and that in this case the statistics may not be representative of the whole period.

32
a) b)
Figure 20 : number of observations of T (a) and S (b) assimilated in PSY4V2R2 in the Gulf Stream 1 region during the OND 2014
quarter.
V.1.3.1.3. Mediterranean Sea sub-regions
a) b)

33
c) d)
e) f)

34
g) h)
Figure 21: As Figure 13 in the Algerian (a and b), the gulf of Lion (c and d), the Ionian (e and f) and the Mersa Matruh (g and h)
regions.
Figure 21 shows the T and S error profiles in sub-regions of the Mediterranean Sea. In the Algerian region the
systems are too warm (0.8°C) and fresh (0.1 PSU). The same characteristics, with smaller amplitude, are
presents in the Gulf of Lion. In the central Ionian region, a surface fresh bias is present between 0 and 200m
(up to 0.2 PSU) and between 500m and 1200m (0.05 PSU) with a cold bias (0.2°C). In Mersa Matruh region
PSY4 seems to be the best system, there is a fresh bias of 0.08 PSU for PSY3 near 200m.
V.1.3.1.4. Indian Ocean sub-regions

35
a) b)
c) d)

36
e) f)
Figure 22: As Figure 13 in the North West Indian (a and b), the South West Indian (c and d) and the East Indian (e and f) regions
The analysis of the sub-regions of the Indian basin (Figure 22) shows that PSY3 displays a cold bias (0.2°C) in
the thermocline in all sub-regions. PSY4 displays a warm bias mainly in the South western part of the basin.
Otherwise the two systems display consistent errors in salinity, they are very close to the observations for that
variable, but PSY4 seems to be the best system in this region.

37
SUMMARY: Starting in April 2013, the PSY2, PSY3 and PSY4 systems converge in terms of model version and
parameterizations. The data assimilation method and tunings are the same, and so do the data sets
assimilated by the three systems. In consequence, the three Mercator Ocean systems display consistent data
assimilation statistics for the quarter OND 2014. The systems are close to the observations, and misfits lie
within the prescribed error (most of the time misfit in T is less than 0.1 °C, misfit in S less then 0.05, misfit in
SLA less than 7 cm). The average and RMS errors are larger in regions of high spatial and/or temporal
variability (thermocline, regions of high mesoscale activity, upwelling, fronts, etc…). PSY4 displays a warm
bias in the 0-500m layer in the Pacific zone that is not present in PSY3. This difference is mainly due to the
initialization of PSY4 which took place in October 2012 (in October 2009 for PSY2 and PSY3) from the WOA09
climatology. The analysis of the GLORYS2V3 reanalysis has shown that the first cycles of SLA data assimilation
induce a large warming of the ocean in the 0-800m layer that the in situ observations have to correct. The
SEEK and the 3D-var bias correction need at least a one year period to correct this bias, which is then
propagated to the deep layers (under 1000m) within 5 to 10 years, if no correction is applied.. PSY4 also
displays low biases in SLA concurrent with warm biases in SST. Coastal regions are weakly constrained in all
systems, and consistent biases appear in SLA and SST (warm and high, or cold and low). There are still many
uncertainties in the MDT, mainly near the coasts and in the Polar Regions. The systems do not use the SLA
misfit information in coastal regions, but problems can appear in open ocean regions where there are still
errors in the MDT. Seasonal biases have been reduced with respect to previous versions of the systems,
thanks to the use of a procedure to avoid the damping of SST increments via the bulk forcing function. A warm
bias is diagnosed in SST this quarter. Globally, there is a warm bias in a mixing layer especially between 50
and 150m. This bias is a consequence of a too diffuse thermocline. The bias correction is not as efficient on
reducing seasonal biases as it is on reducing long term systematic biases. No precipitation correction is applied
in the systems, and uncertainties (overestimation) of the convective precipitations induce large fresh biases in
the tropical oceans (Pacific and Atlantic).

38
V.2. Accuracy of the daily average products with respect to observations
V.2.1. T/S profiles observations
V.2.1.1. Global statistics for OND 2014
a) b)
c) d)
e) f)
Figure 23: Average difference MODEL-OBSERVATION of temperature (a, c, and e, in °C) and salinity (b, d, and f, in psu) in OND
2014 between the daily average hindcast PSY4V2R2 (a and b), PSY3V3R3 (c and d) and PSY2V4R4 (e and f) hindcast products and
all available T/S observations from the Coriolis database. The Mercator hindcast products are taken at the time and location of
the observations. Averages are performed in the 0-50m layer. The size of the pixel is proportional to the number of observations
used to compute the RMS in 2°x2° boxes.

39
a) b)
c) d)
Figure 24: RMS temperature (°C) difference (model-observation) in OND 2014 between all available T/S observations from the
Coriolis database and the daily average hindcast PSY3V3R3 products on the left and hindcast PSY4V2R2 on the right column
colocalised with the observations. Averages are performed in the 0-50m layer (a and b) and in the 0-500m layer (c and d). The size
of the pixel is proportional to the number of observations used to compute the RMS in 2°x2° boxes.
Consistently with the data assimilation statistics the systems PSY4 and PSY2 are slightly warmer (~0.1°C) than
the observations at the global scale in the surface layer (0-50m) as can be seen in Figure 23, especially in
Tropical Pacific and in the Gulf Stream and Kuroshio region. PSY3 is slightly colder in the same layer. Especially
for high resolution systems PSY2 and PSY4, fresh biases also appear in the tropics, linked with overestimated
convective precipitations in the ECMWF analyses, and also to the climatological runoffs that are prescribed.
This quarter the bias is significantly stronger in PSY4 than in PSY2 or PSY3.
As can be seen in Figure 24, in both PSY3 and PSY4 temperature errors in the 0-500m layer have similar
amplitude and spatial patterns, and stand between 0.5 and 1°C in most regions of the globe. RMS errors in the
tropical thermocline can reach 1°C. Regions of high mesoscale activity (Kuroshio, Gulf Stream, Agulhas
current) and regions of upwelling in the tropical Atlantic and Tropical Pacific display higher errors (up to 3°C).
The salinity RMS errors (Figure 25) are usually less than 0.2 but can reach higher values in regions of high
runoff (Amazon, Sea Ice limit) or precipitations (ITCZ, SPCZ, Gulf of Bengal), and in regions of high mesoscale
variability. RMS errors of around 0.2 in the Tropics in the 0-50m layer correspond to the fresh bias linked with
overestimated precipitation. High RMS error in the Beaufort Sea correspond to a salty and cold bias in the
Summer Pacific Water (also called summer Pacific halocline water) for both PSY3 and PSY4 see Focus on
Beaufort Sea for a longer description.

40
a) b)
c) d)
Figure 25: RMS salinity (psu) difference (model-observation) in OND 2014 between all available T/S observations from the
Coriolis database and the daily average hindcast PSY3V3R3 products (a and c) and hindcast PSY4V2R2 (b and d), colocalised with
the observations. Averages are performed in the 0-50m layer (a and b) and in the 0-500m layer (c and d). The size of the pixel is
proportional to the number of observations used to compute the RMS in 2°x2° boxes.
As can be seen in Figure 26, the intermediate resolution model (PSY3) and the high resolution model (PSY4)
display comparable temperature and salinity RMS differences.
For the surface layers: PSY4 displays a global average fresh bias (0.02 psu) at the surface layer, while PSY3
exhibits a salty bias between 0 and 300m.
For the subsurface layers: PSY3 is slightly colder than the observations, mainly between 5 and 800m (0.05°C)
while PSY4 appears to warm in the water column (below 0.1°C).
Both systems are more accurate than the WOA09 climatology, for temperature and salinity at all depths
except for the mean error in salinity in the 0-100m layer.

41
a) b)
c) d)
Figure 26 : OND 2014 global statistics of temperature (°C, a and c) and salinity (psu, b and d) averaged in 6 consecutive layers
from 0 to 5000m. RMS difference (a and b) and mean difference (observation-model, c and d) between all available T/S
observations from the CORIOLIS database and the daily average hindcast products PSY3V3R3 (red), PSY4V2R2 (blue) and WOA09
climatology (grey) colocalised with the observations. NB: average on model levels is performed as an intermediate step which
reduces the artefacts of inhomogeneous density of observations on the vertical.

42
a) b)
c) d)
Figure 27: RMS difference (model-observation) of temperature (a and b, °C) and salinity (c and d, psu) in OND 2014 between all
available T/S observations from the Coriolis database and the daily average PSY2V4R4 hindcast products colocalised with the
observations in the 0-50m layer (a and c) and 0-500m layer (b and d)
The general performance of PSY2 (departures from observations in the 0-500m layer) is less than 0.3°C and
0.05 psu in many regions of the Atlantic and Mediterranean (Figure 27). The strongest departures from
temperature and salinity observations are mainly located in the Gulf Stream, the tropical Atlantic and in the
Baltic Sea. Near surface salinity biases appear in the Alboran Sea, the Gulf of Guinea, the Caribbean Sea, the
Labrador Sea, the Gulf of Mexico (see also Figure 23) , the Baltic Sea and the Gibraltar Strait. In the tropical
Atlantic biases concentrate in the 0-50m layer (cold and fresh bias). This is consistent with the bias correction
not being at work in the mixed layer.

43
V.2.1.2. Water masses diagnostics
a) b) c)
d)
e)
cccc
f)

44
g) h) i)
j) k) l)
Figure 28: Water masses (Theta, S) diagrams in OND 2014 in the Bay of Biscay (a to c), Gulf of Lion (d to f), Irminger Sea (g to i)
and Gulf of Cadiz (j to l), comparison between PSY3V3R3 (a, d, g, j), PSY4V2R2 (b, e, h, k) and PSY2V4R4 (c, f, I, and l) in AMJ 2013.
PSY2, PSY3 and PSY4: yellow dots; Levitus WOA09 climatology: red dots; in situ observations: blue dots.
In Figure 28 the daily products (analyses) are collocated with the T/S profiles in order to draw “T, S” diagrams.
The water masses characteristics of the three systems are nearly similar, PSY2 being locally slightly better than
PSY3 and PSY4 (Mediterranean waters)
In the Bay of Biscay the Eastern North Atlantic Central Water, Mediterranean and Labrador Sea Water can be
identified on the diagram.
- Between 15°C and 19°C, 35 and 36.5 psu, warm and relatively salty Eastern North Atlantic Central
Water gets mixed with the shelf water masses. The saltiest ENACW waters are well captured by the
three systems during this OND 2014 season but PSY2 seems to be the best
- The Mediterranean Waters are characterized by high salinities (Salinities near 36 psu) and relatively
high temperatures (Temperatures near 12°C). PSY2 better captures the highest salinities than PSY3 and
PSY3 for the main profiles. A few profiles in PSY4 appear much more saltier in the Mercator systems
than the in situ data.
- The Labrador Sea waters appear between 4°C and 7°C, 35.0 and 35.5 psu and are well represented by
the three systems.
In the Gulf of Lion:

45
- The Levantine Intermediate Water (salinity maximum near 38.6 psu and 13.5°C) is too cold and little
fresh and compact in all systems this OND 2014 season.
- The water mass detected near 38.8 psu and 13-13.3°C may correspond to measurements on the
shelves made by research vessels.
In the Irminger Sea:
- The Irminger Sea Water (≈ 3°C and 35 psu) and the North Atlantic water (at the surface) are well
represented by the systems.
- The Labrador Sea Waters are slightly too salty in all systems.
- Waters colder than 3°C and ≈ 34.9 psu (Iceland Scotland Overflow waters) are presents in all systems
but seem too fresh.
In the Gulf of Cadiz:
- The Mediterranean outflow waters (T around 10°C) are well represented but appears too fresh in
PSY3. The high resolution systems (PSY2 and PSY4) better capture the spread of temperature and
salinity at depth as well as at the surface (Atlantic Surface waters).
In the western tropical Atlantic, all the systems capture the salinity minimum of the Antarctic Intermediate
Waters (AIW). The spread of temperature and salinity at surface and sub-surface is well represented too.
In the eastern tropical Atlantic the three systems represent well the water masses. For the high resolution
systems, the spread of fresh water near the surface is too extended in comparison with observations.
a) b) c)
d) e) f)

46
Figure 29 : Water masses (T, S) diagrams in the Western Tropical Atlantic (a to c) and in the Eastern Tropical Atlantic (d to f) in
OND 2014: for PSY3V3R3 (a, d); PSY4V2R2 (b, e); and PSY2V4R4 (c, f). PSY2, PSY3 and PSY4: yellow dots; Levitus WOA09
climatology; red dots, in situ observations: blue dots.
In the Agulhas current (Figure 30) PSY3 and PSY4 give a realistic description of water masses. The central and
surface waters characteristics are slightly more realistic in PSY4 than in PSY3. In the Kuroshio region the North
Pacific Intermediate waters (NPIW) are too salty for the two global systems. At the surface in the highly
energetic regions of the Agulhas and of the Gulf Stream, the water masses characteristics display a wider
spread in the high resolution 1/12° than in the ¼°, which is more consistent with T and S observations. In the
Gulf Stream region, PSY4 is closer to the observations than PSY3 especially in the North Atlantic Central
Waters (NACW) where PSY3 is too fresh. The saltiest waters near the surface (North Atlantic Subtropical
Waters, NASW) are only captured by PSY4. The two systems are very close in this region.
a) b)

47
c) d)
e) f)
Figure 30: Water masses (T, S) diagrams in South Africa (a and b), Kuroshio (c and d), and Gulf Stream region (e and f): for
PSY3V3R1 (a, c and e); PSY4V1R3 (b, d and f) in OND 2014. PSY3 and PSY4: yellow dots; Levitus WOA09 climatology: red dots; in
situ observations: blue dots.

48
V.2.1.2.1. Focus on Beaufort Sea
a) b) c)
d) e) f)
Figure 31 : Water masses (T, S) diagrams in Beaufort Sea for PSY3 (a) and PSY4 (d), salinity profiles in this region for PSY3 b) and
PSY4 e), and the same for the temperature profiles : PSY3 (c) and PSY4(f). Symbology : yellow dots; Levitus WOA09 climatology:
red dots; in situ observations: blue dots.
The T/S diagrams of in situ profiles are showing three temperature maximums: two bigs and a small. The first
big maximum situated around 35 psu and near 1°C corresponds to the Atlantic Water (AW). The second
maximum (around 31 and 0.5°C) corresponds to the Pacific Summer Water (PSW) (see ref 3-4) and the third
small maximum correspond to the Near Sea Surface Temperature Maximum (NSTM around 28 psu and -1°C
near 30m). Two temperature minimums are also presents: the Pacific Winter Water (PWW) and the remnant
of the previous winter’s surface mixed layer (between the NSTM and the PSW)e. Associated with theses
minimums and maximums we have three haloclines: the lower halocline between the PWW and AW
maximum, the winter halocline around the PSW maximum and the summer halocline near the NSTM water
masse.
The both systems well represent the Atlantic Water but are too salty for the summer Pacific Water but PSY3
sticks better to the observations for this water masse. The NSTM waters are not represented by the two
systems. When we look at the profiles, we see a strong cold and salty bias in the two systems in the whole
water column. This strong bias could be link to a lake of freshwater in the runoffs or too much melting during
the summer. We can notice that the bias correction is not as efficient at this latitude and none of these data
are assimilated, so we are close to a free-run mode in this region.

49
V.2.2. SST Comparisons
Quarterly average SST differences with OSTIA analyses show that the systems’ SST is very close to
OSTIA, with difference values staying below the observation error of 0.5 °C on average. High RMS difference
values (Figure 32) are encountered in high spatial and temporal variability regions such as the Gulf Stream or
the Kuroshio, Zapiola eddy, Agulhas. The error is also high (more than 2°C locally) in marginal seas of the
Antarctic and Arctic Ocean where the sea ice limit increases the SST errors and the SST contrasts. Small
features in the RMS difference appear in PSY3 that are not found in the higher resolution systems PSY4 and
PSY2 (for instance this quarter in the Labrador Sea). High RMS differences also appear in the upper part of the
Circumpolar current, the main features are located in the South of Indian and Pacific Ocean. A dipole of
discrepancies appears in the Tropical Pacific probably link to the positive phase of the North Atlantic
Oscillation (see Information on the large scale climatic conditions).
a) b)
c)
Figure 32 : RMS temperature (°C) differences between OSTIA daily analyses and PSY3V3R3 daily analyses (a); between OSTIA and
PSY4V2R2 (b), between OSTIA and PSY2V4R4 (c) in OND 2014. The Mercator Océan analyses are colocalised with the satellite
observations analyses.
The systems’ tropical oceans are warmer than OSTIA (Figure 33), as do the high latitudes (especially in the ACC
this austral winter). The Pacific subtropical and the Indian Ocean are colder than OSTIA on average like the
South Atlantic Ocean, and some closed or semi-enclosed seas: the Gulf of Mexico, Chinea Sea, Ariabian Sea.

50
a) b)
c)
Figure 33: Mean SST (°C) daily differences in OND 2014 between OSTIA daily analyses and PSY3V3R3 daily analyses (a), between
OSTIA and PSY4V2R2 (b) and between OSTIA and PSY2V4R4 daily analyses (c).
V.2.3. Drifting buoys velocity measurements
Recent studies (Law Chune, 20124
, Drévillon et al, 20125
) - in the context of Search-And-Rescue and drift
applications – focus on the need for accurate surface currents in ocean forecasting systems. In situ currents
are not yet assimilated in the Mercator Ocean operational systems, as this innovation requires a better
characterization of the surface currents biases.
The comparison of Mercator analyses and forecast with AOML network drifters velocities combines two
methods based on Eulerian and Lagrangian approaches.
V.2.3.1. Eulerian Quality control
4
Law Chune, 2012 : Apport de l’océanographie opérationnelle à l’amélioration de la prévision de la dérive océanique
dans le cadre d’opérations de recherche et de sauvetage en mer et de lutte contre les pollutions marines
5
Drévillon et al, 2012 : A Strategy for producing refined currents in the Equatorial Atlantic in the context of the search of
the AF447 wreckage (Ocean Dynamics, Nov. 2012)

51
a) b)
c) d)
e) f)
Figure 34: Near surface (15m) zonal current (m/s) comparison between the Mercator systems analyses and observed velocities from drifters. In the left column:
velocities collocated with drifter positions in OND 2014 for PSY3V3R3 (a), PSY4V2R2 (c) and PSY2V4R4 (e). In the right column, zonal current from drifters in OND
2014 (b) at global scale, AOML drifter climatology for OND with new drogue correction from Lumpkin & al, in preparation (d) and observed zonal current in OND
2014 (f) over PSY2’s domain.

52
a) b)
c) d)
e) f)
Figure 35 : In OND 2014, comparison of the mean relative velocity error between in situ AOML drifters and model data (a, c and
e) and mean zonal velocity bias between in situ AOML drifters with Mercator Ocean correction (see text) and model data (b, d
and f). a and b: PSY3V3R3, c and d: PSY4V2R3, e and f: PSY2V4R2. NB: zoom at 500% in order to see the arrows.
The fact that velocities estimated by the drifters happen to be biased towards high velocities is taken into
account, applying slippage and windage corrections (cf QuO Va Dis? #5 and Annex C). Once this so called

53
“Mercator Ocean” correction is applied to the drifter observations, the zonal velocity of the model (Figure 34
and Figure 35) at 15 m depth and the meridional velocity (not shown) is more consistent with the
observations for the OND 2014 period.
On average in this OND quarter, the usual behaviour compared to drifters’ velocities is that the high
resolution system PSY4 overestimates the surface velocity in the mid-latitudes and the low resolution PSY3
underestimates the currents. PSY4 overestimates the Equatorial Pacific currents (South Equatorial Current and
North Equatorial Counter Current), the Kuroshio region, the Gulf of Alaska and the Southern Oceans. PSY3
underestimates the Pacific South Equatorial Current, the Western subarctic gyre and the South Indian Ocean
and overestimates the Southern oceans and the Gulf Stream region.
For all systems the largest direction errors are local (not shown) and generally correspond to ill positioned
strong current structures in high variability regions (Gulf Stream, Kurioshio, North Brazil Current, Zapiola eddy,
Agulhas current, Florida current, East African Coast current, Equatorial Pacific Countercurrent).
The differences between the systems mainly appear in the Atlantic and Pacific gyres, in the Western subartic
gyre and in the South Indian basin see Figure 35. 50% of the ocean current feedbacks on the wind stress in
PSY2 and PSY3, which slows the surface currents with respect to PSY4.
V.2.3.2. Lagrangian Quality control
The aim of the Lagrangian approach is to compare the observed buoy trajectory with virtual trajectories
obtained with hindcast velocities, starting from the same observed initial location. The virtual trajectories are
computed with the ARIANE software6
(see annex Erreur ! Source du renvoi introuvable.). The metric shown
here (Figure 36) is the distance between the trajectories after 1, 3 and 5 days, displayed at each trajectory
initial point.
a) b)
6
http://stockage.univ-brest.fr/~grima/Ariane/

54
c) d)
e) f)
Figure 36: In OND 2014, comparison of the mean distance error in 1°x1° boxes after a 1-day drift (a and b), after a 3-days drift (c
and d), and after a 5-days drift (e and f).between AOML drifters’ trajectories and PSY3V3R3 trajectories (a, c, and e) and between
AOML drifters’ trajectories and PSY4V2R2 (b,d, and f).
Few differences appear between the systems and most of the high velocity biases that are diagnosed in Figure
35 imply a large distance error (120 to 180km) after a few days drift in Figure 36. For instance in the Pacific
ocean, both PSY3 and PSY4 produce an error larger than 100km after only a 1-day drift near the North Coast
of Papua New Guinea.
In the subtropical gyres and in the North Pacific, which are less turbulent regions, the errors rarely exceed 30
km after 5 days.

55
Figure 37:Cumulative Distribution Functions of the distance error (km, on the left) and the direction error (degrees, on the right
panel) after 1 day (blue), 3 days (green) and 5 days (red), between PSY4V1R3 forecast trajectories and actual drifters trajectories.
Over the whole domain on a long period, cumulative distribution functions (Figure 37) show that in 80% of
cases, PSY3 and PSY4 modelled drifters move away from the real drifters less than : 30km after 1 day, 70km
after 3 days, and 100km after 5 days. As explained before, the remaining 20% generally correspond to ill
positioned strong current structures in high variability regions. PSY4 currents seem to induce slightly more
errors than PSY3 currents essentially for errors below 10-15 km. This is probably related to the high mesoscale
activity of the global high resolution model. This may not be systematic at the regional scale (under
investigation).

56
V.2.4. Sea ice concentration
a) b) c)
d) e) f)
Figure 38: Arctic (a to c) and Antarctic (d to f) OND 2014 Sea ice cover fraction in PSY3V3R3 (a and d), in CERSAT observations (b
and e) and difference PSY3 – CERSAT (c and f).
On average over the OND 2014 period, sea ice concentration of the two systems is close to the observations.
For both PSY3 and PSY4, underestimations of modelled sea ice concentration with observed sea ice
concentration appear near the marginal zones (Greenland Sea, North of Barent Sea, Kara Sea) and in the
Canadian basin (Figure 38 and Figure 39). We can notice that the sea ice observations are not yet assimilated
by the systems. Even if PSY3 and PSY4 are melting too much ice in the summer, in the beginning of winter the
sea ice cover seems to be close to the data. In the Arctic, differences with observations are in the same order
for PSY3 and PSY4. Few discrepancies can be distinguished inside the sea ice pack but it will not be considered
as significant as the sea ice concentration observations over 95% are not reliable.
Model studies show that the overestimation in the Canadian Archipelago is first due to badly resolved sea ice
circulation.
In the Antarctic the summer sea ice concentration is underestimated on average by both PSY3 and PSY4, On
the contrary it is overestimated along the sea ice edge in the Bellingshausen Sea and in the South of Kerguelen
Island.

57
a)
b)
c)
d) e) f)
Figure 39: Arctic (a to c) and Antarctic (d to f) OND 2014 Sea ice cover fraction in PSY4V2R2 (a and d), in CERSAT observations (b
and e) and difference PSY4 – CERSAT (c and f).
Figure 40 illustrates the fact that in Arctic, the sea ice cover in OND 2014 is less than the past years
climatology, especially in the Bering Strait, Greenland Sea, Kara and Barents Sea (see also section IV).
For the Antarctic, the ice cover is greater than the climatology except in the Bellingshausen Sea and in the
Southern Indian Ocean. This decrease is due in large part to persistent northerly winds.

58
a) b)
c)
d)
Figure 40: OND 2014 Arctic sea ice extent in PSY4V2R2 with overimposed climatological OND 1992-2010 sea ice fraction (magenta
line, > 15% ice concentration) (a) and NSIDC map of the sea ice extent in the Arctic for December 2014 in comparison with a 1979-
2000 median extend (b), the same for the Antarctic region for the PSY4V2R2 system c) and the NSIDC estimation d)
(ftp://sidads.colorado.edu/DATASETS/NOAA/G02135/Dec/).

59
V.2.5. Closer to the coast with the IBI36V2 system: multiple comparisons
V.2.5.1. Comparisons with SST from CMS
The bias, RMS error and correlation calculated from comparisons with SST measured by satellite in
OND 2014 (Météo-France CMS high resolution SST at 0.02°) are displayed in Figure 41. The cold bias in the
North Sea present in JAS 2014 has almost vanished. A cold bias is still present in the Irish Sea and in the
Alboran Sea. A warm bias is present along the south Moroccan coast and Canary Islands, and in a large area
west of the Iberian Peninsula between 35°N and 50°N. This last warm bias was present in OND 2013 but not
so extended (not shown). In the Mediterranean Sea a warm bias can be noticed in the south-west corner of
Spain between the coast and the Baleares Islands. Low correlations can be noticed (<0.4) in the northern part
of the domain, associated with a warm bias. PSY2 and PSY4 systems show similar results but the biases are
generally smaller (especially the cold biases in the Irish Sea and in the Alboran Sea).
Note that the number of observations in the Mediterranean Sea is very good for OND 2014 compared to
previous years.
a) Mean Bias IB36V2 b) RMS error IBI36V2
c) Correlation IBI36V2
j) Number of Obs
d) Mean Bias PSY2V4R4 e) RMS error PSY2V4R4 f) Correlation PSY2V4R4

60
g) Mean Bias PSY4V2R2
h) RMS error PSY4V2R2 i) Correlation PSY4V2R2
Figure 41 : Mean bias (observation-model) (a), RMS error (b), correlation (c) between IBI36V2 and analysed L3 SST from Météo-
France CMS for the OND 2014 quarter. Same diagnostics for PSY2V4R4 (d,e,f) and PSY4V2R2 (g,h,i). Number of Météo-France CMS
observations for the OND 2014 quarter (j).
V.2.5.2. Comparisons with in situ data from EN3/ENSEMBLE for OND 2014
Averaged temperature profiles (Figure 42) show that the strongest mean bias and RMS error (more than 2°C)
are observed between surface and 100 m depth, and around 500 m depth, as in OND and JAS 2014. Below the
thermocline, the mean bias is almost zero. In the Bay of Biscay, the bias and RMS error are also maximum
near the surface and at the Mediterranean water level. As shown by the mean temperature profiles, IBI36 and
PSY2 are very close.
a) Temperature, 0-200 m b) Temperature, 0-2000 m

61
c) Temperature, 0-200 m d) Temperature, 0-2000 m
Figure 42 : Over the whole IBI36V4 domain (a and b) and over the Bay of Biscay (c and d): mean “IBI36V4 - observation” salinity
(psu) bias (red) and RMS error (blue) in OND 2014 (a and c), and mean profile from IBI36V4 (black), PSY2V4R4 (green) and from the
observations (red) in OND 2014 (b and d). In the lower right corner of each plot: position of the individual profiles.
The maximum salinity bias and RMS error (Figure 43) occur near the surface. The model is too fresh between
the surface and 120 m depth, with PSY2 closer to the observations than IBI36. The RMS error is slightly
stronger at the Mediterranean Sea Water level. In the Bay of Biscay the Mediterranean waters are too salty.
Note: averaged profiles are discontinuous because the number of observations varies with depth.
a) Salinity, 0-200 m b) Salinity, 0-2000 m
c) Salinity, 0-200 m d) Salinity, 0-2000 m

62
Figure 43: Over the whole IBI36V4 domain (a and b) and over the Bay of Biscay (c and d): mean “IBI36V4 - observation” salinity (psu)
bias (red) and RMS error (blue) in OND 2014 (a and c), and mean profile from IBI36V4 (black), PSY2V4R4 (green) and from the
observations (red) in OND 2014 (b and d). In the lower right corner of each plot: position of the individual profiles.
V.2.5.3. MLD Comparisons with in situ data
Figure 44 shows that the distribution of modeled mixed layer depths among the available profiles is
comparable to the observed distribution for both PSY2 and IBI over the whole IBI domain. Values of the mixed
layer depth between 50 m and 70 m occur too often in the systems compared with the observations
(especially for PSY2). In the Bay of Biscay, IBI36 over-estimates the values between 15 m and 35 m, and under-
estimates the values between 35 m and 60 m.
a)
b)

63
c) d)
Figure 44 : Mixed Layer Depth distribution in OND 2014 in IBI36V4 (a and b) and PSY2V4R4 (c and d), calculated from profiles with
the temperature criteria (difference of 0.2°C with the surface); the models are in grey, the observations in red. a and c: classification
for the whole IBI domain, b and d: classification for the Bay of Biscay
V.2.5.4. Comparisons with moorings and tide gauges
a) b)
Figure 45 : For IBI36V4: RMS error (a, in cm) and correlation (b) for the non-tidal Sea Surface Elevation at tide gauges in OND 2014,
for different regions and frequencies.
The RMS error of the residual (non tidal) elevation of sea surface (Figure 45) computed with harmonic
decomposition (Foreman 1977) and Loess low-pass filtering is between 5 and 16 cm. It is close to 5 cm in the
Canary Islands, Mediterranean Sea and Iberian Coast. The RMS decreases for some frequency bands, and the

64
smallest values occur in the 1-10-day band. The correlation is significant at all frequencies, and reaches high
values for periods lower than 30 days (at high frequencies).
a)
b)
c)
d) e) f)
g)
h)
i)

65
Figure 46 : For IBI36V4: Bias (observation-model) (a, d and g), RMS error (°C) (b, e and h) and correlation (c, f and i) of the Sea Surface Temperature between IBI36
and moorings measurements in October (a to c), November (d to f) and December 2014 (g to i).
In Figure 47 we can see that the SST correlations between the coastal moorings and the IBI model are
generally good for the three months. In December the model is colder than the observations in the Irish Sea,
Celtic Sea and English Channel. Along the coast of the Gulf of Lions, the model is colder than the observations
for the three months.
Estaca de bares
Biscay Les pierres noires
Sandettie lightship Gulf of Lion
Figure 47 : Surface temperature (°C) time series at a selection of moorings which locations are shown on the upper left panel (no
observation for Gulf of Lion).

66
V.2.6. Biogeochemistry validation: ocean colour maps
a) BIOMER 4 October b) Globcolor October
c) BIOMER4 November d) Globcolor November
e) BIOMER4 December f) Globcolor December
Figure 48 : Chlorophyll-a concentration (mg/m
3
) in October(a and b), November (c and d) and December 2014 (e and f) for the
Mercator system BIOMER4 (a, c, and e) and Chlorophyll-a concentration from Globcolour (b, d and f).
As can be seen on Figure 48, the surface chlorophyll-a concentration by the new version of BIOMER (BIOMER4
see Quovadis #18 for a complete description) is in a good agreement with observations on average over the
globe. Discrepancies remain anyway: along the Argentina’s coast, near Bering Strait, or in the subpolar gyres.
For the distribution of Chlorophyll-a concentration errors in Figure 49 , the shape of BIOMER4’s distribution is
Gaussian but the mean bias remains still important.

67
a) b) c)
Figure 49 : Probability Density Function (PDF) of Chl-a bias in log scale (log10(obs)-log10(model)) in the North Atlantic (30-70N;
80W:20E) for October (a), November (b) and December (c) 2014.
The discrepancies at the global scale appear in the RMS differences for the mean OND 2014 season (Figure 50)
especially in the Zapiola region, near the Peru-Chili coast (upwelling region), in the South of Indian Basin near
Kerguelen Islands.
Figure 50 : RMS difference between BIOMER and Globcolour Chl-a concentrations (mg/m
3
) in OND 2014.

68
VI Forecast error statistics
VI.1. General considerations
The daily forecasts (with updated atmospheric forcings) are validated in collaboration with SHOM/CFUD. This
collaboration has been leading us to observe the degradation of the forecast quality depending on the
forecast range. When the forecast range increases the quality of the ocean forecast decreases as the
initialization errors propagate and the quality of the atmospheric forcing decreases. Additionally the
atmospheric forcing frequency also changes (see Figure 51). The 5-day forecast quality is optimal; starting
from the 6th
day a drop in quality can be observed which is linked with the use of 6-hourly atmospheric fields
instead of 3-hourly; and starting from the 10th
day the quality is strongly degraded due to the use of persisting
atmospheric forcings (but not constant from the 10th
to the 14th
day as they are relaxed towards a 10-day
running mean).
Figure 51: Schematic of the change in atmospheric forcings applied along the 14-day ocean forecast.
VI.2. Forecast accuracy: comparisons with T and S observations when and where
available
VI.2.1. North Atlantic region
As can be seen in Figure 52 the PSY2 temperature products have a better accuracy than the climatology in the
North Atlantic region in OND 2014 (note that in the 2000-5000m layer, the statistics are performed on a very
small sample of observations, and thus are not really representative of the region or layer. In general the
analysis is more accurate than the 3-day and 6-day forecast for both temperature and salinity. The RMS error
thus increases with the forecast range (shown for NAT region Figure 52 and MED region Figure 53). The biases
in temperature are generally small (of the order of 0.2 °C) compared to the climatology’s biases (of the order
of 0.8 °C) while the salinity biases are significant (0.12 psu at the surface while the climatology bias is only 0.05
psu). The whole water column is too salty and warm (0.1°C and 0.1 PSU at the surface).

69
a) b)
c) d)
Figure 52: Accuracy intercomparison in the North Atlantic region for PSY2V4R4 in temperature (a and c) and salinity (b and dl)
between hindcast (green), nowcast (yellow), 3-day (orange) and 6-day (red) forecast and WO09 climatology (grey). Accuracy is
measured by a mean difference (a and b) and by a rms difference (c and d) of temperature and salinity with respect to all
available observations from the CORIOLIS database averaged in 6 consecutive layers from 0 to 5000m. All statistics are performed
for the OND 2014 period. NB: average on model levels is performed as an intermediate step which reduces the artefacts of
inhomogeneous density of observations on the vertical.

70
VI.2.2. Mediterranean Sea
In the Mediterranean Sea in OND 2014 (Figure 53) the PSY2 products are more accurate than the climatology
on average. PSY2 is too salty at the surface (0.03 psu) and too warm and fresh between 5-100m (up to 0.04
psu and 0.2°C).
a) b)
c) d)
Figure 53: Accuracy intercomparison in the Mediterranean Sea region for PSY2V4R4 in temperature (°C, a and c) and salinity (psu, b and d) between hindcast,
nowcast, 3-day and 6-day forecast and WO09 climatology. Accuracy is measured by a rms difference (c and d) and by a mean difference (a and b) with respect to
all available observations from the CORIOLIS database averaged in 6 consecutive layers from 0 to 5000m. All statistics are performed for the OND 2014 period.
NB: average on model levels is performed as an intermediate step which reduces the artefacts of inhomogeneous density of observations on the vertical.

71
VI.2.3. Tropical Oceans and global
Results already described in previous sections are confirmed by the profiles shown in Figure 54. PSY4 and
PSY2 are usually warmer (less cold) than PSY3 and the observations in common regions especially in the 0-
800m layer. PSY2 and PSY4 suffer from a fresh bias in the Tropical oceans, especially in the Tropical Atlantic
(up to 0.15 psu in the surface layer). In the other tropical basins PSY3 is less biased than PSY4 in terms of
salinity but suffer from a slightly salty bias in the 0-300m layer.
a) b)
c) d)

72
e) f)
Figure 54: (Obs-model) mean differences in temperature (°C, a, c, and e) and salinity (psu, b, d, and f) for PSY3V3R3 (red),
PSY2V4R4 (yellow) and PSY4V2R2 (blue) systems in the Tropical Atlantic (a and b), the Tropical Pacific (c and d) and the Indian
Ocean (e and f) in OND 2014.
At the surface, PSY4 is too fresh for the Tropical Pacific and PSY3 is too salty in the Indian Ocean; PSY3 beats
PSY4 in Tropical Atlantic and Pacific but not in the Indian Ocean where PSY3 exhibits a strong salty bias
between 0-100m (up to 0.2).
PSY3, PSY4 and PSY2 display similar accuracy levels in terms of temperature RMS error in the Tropical Oceans.
At the surface, PSY3’s salinity slightly outperforms PSY4 on average over the globe. RMS error increases with
forecast range (shown here for PSY3), as could be expected, and that the 5-day forecast always beats the
climatology (except for global mean salinity). Eventually this OND 2014 quarter, we would suggest choosing
PSY3 for Atlantic and Pacific Tropical Ocean and PSY4 for Indian Ocean.
a) b)

73
c) d)
e) f)
g) h)

74
i) j)
Figure 55: same as Figure 52 but for RMS statistics and for temperature (°C), PSY3V3R3 and PSY4V2R2 systems and the Tropical
Atlantic (a and b), the Tropical Pacific (c and d) and the Indian Ocean (e and f). The global statistics are also shown for
temperature (°C, g and h) and salinity (psu, I and j). The right column compares the analysis of the global ¼° PSY3V3R3 (red) with
the analysis of the global 1/12° PSY4V2R2 (blue).
VI.3. Forecast accuracy: skill scores for T and S
a) b)
c) d)

75
e) f)
Figure 56: Temperature (left) and salinity (right) skill scores in 4°x4° bins and in the 0-500m layer in OND 2014 in PSY3V3R3 (a
and b) , PSY4V2R2 (c and d) and PSY2V4R4 (e and f). The skill illustrates the ability of the 3-days forecast to be closer to in situ
observations than the persistence of the analysis, see Equation 1. Yellow to red values indicate that the forecast is more accurate
than the persistence.
The Murphy Skill Score (see Equation 1) is described by Wilks, Statistical Methods in the Atmospheric Sciences,
Academic Press, 2006. This score is close to 0 if the forecast is equivalent to the reference. It is positive and
aims towards 1 if the forecast is more accurate than the reference.
( )
( )∑ ∑
∑ ∑
= =
= =






−






−
−= n
k
M
m
mm
n
k
M
m
mm
ObsRef
M
ObsForecast
M
SS
1 1
2
1 1
2
1
1
1
Equation 1

76
Figure 57 : As Figure 56 but the reference is the WOA 2009 climatology. Temperature (a and c) and salinity (b and d) skill scores
are displayed for OND 2014, for PSY3V3R3 (a and b) and for PSY2V4R4 (c and d). The skill illustrates the ability of the 3-days
forecast to be closer to in situ observations than the climatology, see Equation 1. Yellow to red values indicate that the forecast is
more accurate than the climatology.
The Skill Scores displayed on Figure 56 and Figure 57 show the added value of the PSY3 forecast with respect
to the persistence of the analysis (Figure 56) and to climatology (Figure 57). All Mercator Ocean systems have
a very good level of performance with respect to the climatology (see previous section). The same diagnostics
with climatology for PSY4 are not available for OND 2014. When the reference is the persistence of the last
analysis, the result is noisier and the systems 3-day forecast seems to have skill in some regions in particular:
North East Atlantic, eastern pacific, Northern Indian basin. In some regions of high variability (for instance in
the Antarctic, Gulf Stream, Agulhas Current, Zapiola) the persistence of the previous analysis is locally more
accurate than the forecast.
VI.4. Forecast verification: comparison with analysis everywhere
Not available this quarter.

77
VII Monitoring of ocean and sea ice physics
VII.1. Global mean SST and SSS
a)
b)
c)
Figure 58: daily SST (°C) spatial mean for a one year period ending in AMJ 2013 for PSY2V4R4 (a), PSY3V3R3 (b), and PSY4V2R1
until 31/10/2012 then PSY4V2R2 (c). Mercator Ocean systems are in black and Reynolds AVHRR ¼° analysis in red.
The spatial mean of SST is computed for each day of the year, for PSY2, PSY3 and PSY4. The mean SST is
compared to the mean of Reynolds AVHRR SST on the same domain (Figure 58).
The main feature is the good agreement of PSY2 and PSY3 with the observations all year long and on global
average. A warm bias can be diagnosed in PSY3 and PSY4 with respect to the observations.

78
VII.2. Surface EKE
Regions of high mesoscale activity are diagnosed in Figure 59: Kuroshio, Gulf Stream, Niño 3 region in the
eastern Equatorial pacific, Zapiola eddy, Agulhas current. PSY3 at ¼° and PSY4 at 1/12° are in very good
agreement. EKE is generally higher in the high resolution PSY4 system, for instance this quarter at the Equator
near 120°W.
a)
b)
Figure 59: eddy kinetic energy EKE (m²/s²) at 15m for PSY3V3R3 (a) and PSY4V2R2 (b) for OND 2014.

79
VII.3. Mediterranean outflow
The Mediterranean outflow displayed in Figure 60 is more diffuse in all systems than in the climatology of the
Gulf of Cadiz. All systems now display similar vertical structure in the region.
a) b) c)
d) e) f)
Figure 60: Comparisons between OND 2014 mean temperature (°C, a to c) and salinity (psu, d to f) profiles in Mercator systems
(black): PSY2V4R4 (a and d), PSY3V3R3 (b and e) and PSY4V2R2 (c and f), and in the Levitus WOA05 (green) and ARIVO (red)
monthly climatologies.
VII.4. Sea Ice extent and area
The time series of monthly means of sea ice area and sea ice extent (area of ocean with at least 15% sea ice)
are displayed in Figure 61 and compared to SSM/I microwave observations. Both ice extent and area include

80
the area near the pole not imaged by the sensor. NSIDC web site specifies that it is assumed to be entirely ice
covered with at least 15% concentration. This area is 0.31 million square kilometres for SSM/I.
a) b)
Figure 61: Sea ice area (a, 10
6
km2) and extent (b, 10
6
km2) in PSY3V3R3 (blue line), PSY4V1R3 until October then PSY4V2R2
(black line) and SSM/I observations (red line) for a one year period ending in OND 2014, in the Arctic (upper panel) and Antarctic
(lower panel).
These time series indicate that sea ice products from PSY4 and PSY3 are now very similar on average over the
polar domains. In the Arctic both systems perform very well in terms of average quantities. In the Antarctic
the seasonal cycle of the sea ice area is overestimated, with an overestimated sea ice area this austral Spring
in both systems (while the extent is better captured), and too much melting in summer for both systems.

81
I Annex A
I.1.Table of figures
Figure 1: schematic of the operational forecast scenario for IBI36Q (green) and PSY2Q (blue). Solid lines are the PSY2 weekly
hindcast and nowcast experiments, and the IBI36 spin up. Dotted lines are the weekly 14-day forecast, dashed lines are
daily updates of the ocean forecast forced with the latest ECMWF atmospheric analysis and forecast. The operational
scenario of PSY3, PSY4 and PSY3Q, PSY4Q is similar to PSY2’s scenario. In the case of PSY4, only weekly hindcast,
nowcast and 7-day forecast are performed................................................................................................. 7
Figure 2: Seasonal OND 2014 temperature anomalies with respect to GLORYS2V3 climatology (1993-2011). (a): SST anomaly
(°C) at the global scale from the 1/4° ocean monitoring and forecasting system PSY3V3R3. (b): heat content anomaly
(ρ0Cp∆T, with constant ρ0=1020 kg/m3 ) from the surface to 300m. ......................................................... 12
Figure 3: Seasonal OND 2014 temperature anomaly (°C) with respect to GLORYS2V3 climatology (1993-2011), vertical section
2°S-2°N mean, Pacific Ocean, PSY3V3R3.................................................................................................... 12
Figure 4 : Arctic sea ice extent from the NSIDC, available at http://nsidc.org/arcticseaicenews........................ 13
Figure 5 : Comparison between SLA data assimilation scores (a: average misfit in cm, b: RMS misfit in cm, c: number of
observations) for all available Mercator Ocean systems in OND 2014. The scores are averaged for all available satellite
along track data (Saral/altiKa, Jason 2, Cryosat 2). For each region the bars refer to PSY2V4R4 (yellow), PSY3V3R3 (red),
PSY4V2R2 (blue). ........................................................................................................................................ 14
observations) for all available Mercator Ocean systems in OND 2014 in sub-regions of the Tropical Atlantic (TAT) and
North Atlantic (NAT). The scores are averaged for all available satellite along track data (Jason 1 G, Jason 2, Cryosat 2).
For each region the bars refer respectively to PSY2V4R4 (yellow), PSY3V3R3 (red), and PSY4V2R2 (blue).15
observations) for all available Mercator Ocean systems in OND 2014 in the Mediterranean Sea MED sub-regions:
PSY2V4R4 (yellow), PSY3V3R3 (red) and PSY4V2R2 (blue). The scores are averaged for all available satellite along track
data (Jason 1 G, Jason 2, Cryosat 2). See annex B for geographical location of regions............................ 16
observations) for all available global Mercator Ocean systems in OND 2014 in sub-regions of the SAT, IND, NPA, SPA,
TPA, and ACC basins: PSY3V3R3 (red) and PSY4V2R2 (blue). The scores are averaged for all available along track satellite
data (Jason 1 G, Jason 2, Cryosat 2). The geographical location of regions is displayed in annex B.......... 17
Figure 9: Comparison between Reynolds ¼°AVHRR-data assimilation scores (a: average misfit in °C, b: RMS misfit in °C, c:
number of observations) for all available Mercator Ocean systems in OND 2014: PSY4V2R2 (blue), PSY3V3R3 (red), and
PSY2V4R4 (yellow)...................................................................................................................................... 18
Figure 10: Comparison between Reynolds ¼°AVHRR SST data assimilation scores (a: average misfit in °C, b: RMS misfit in °C, c:
number of observations) for all available Mercator Ocean systems in OND 2014 in sub-regions of the Tropical Atlantic
(TAT) and of the North Atlantic (NAT): PSY4V2R2 (blue), PSY3V3R3 (red) and PSY2V4R4 (yellow). The geographical
location of regions is displayed in annex B................................................................................................. 19
Figure 11: Comparison between Reynolds ¼° AVHRR SST data assimilation scores (a: average misfit in °C, b: RMS misfit in °C,
c: number of observations) for all available Mercator Ocean systems in OND 2014 in the Mediterranean Sea: PSY4V2R2
(blue), PSY3V3R3 (red) and PSY2V4R4 (yellow). The geographical location of regions is displayed in annex B.20
Figure 12: Comparison between Reynolds ¼° AVHRR SST data assimilation scores (a: average misfit in °C, b: RMS misfit in °C,
c: number of observations) for all available global Mercator Ocean systems in OND 2014 in the SAT, NPA, TPA, SPA, IND
and ACC basins: PSY3V3R3 (red) and PSY4V2R2 (blue). See annex B for geographical location of regions.21
Figure 13: Profiles of OND 2014 innovations of temperature (°C, left column) and salinity (psu, right column), mean (solid line)
and RMS (dashed line) for PSY3V3R3 (red), PSY2V4R4 (yellow) and PSY4V1R3 (blue) in the North Atlantic NAT (a and b),
Mediterranean Sea MED (c and d) and Arctic Ocean ARC (e an fl, the region starts of north of 67°N). Basin masks are
applied to keep only the main basin of interest (no Mediterranean in the Altantic NAT, no Black Sea in the
Mediterranean MED, etc…)........................................................................................................................ 23
Figure 14: As Figure 13 in the Tropical Atlantic TAT (a and b), South Atlantic SAT (c and d) and Indian IND (e and f) basins. 25
Figure 15: As Figure 13 in the North Pacific NPA (a and b), Tropical Pacific TPA (c and d) and South Pacific SPA (e and f). 27
Figure 16: As Figure 13 in the ACC Southern Ocean basin (from 35°S). .............................................................. 28
Figure 17: As Figure 13 in the North East Atlantic (IBI) and the North Madeira XBT regions.............................. 29

82
Figure 18: As Figure 13 in the Dakar region......................................................................................................... 30
Figure 19: As Figure 13 in the Ascension tide (a and b), the Gulf of Mexico (c and d) and the Gulf Stream 1 regions (e and f)31
Figure 20 : number of observations of T (a) and S (b) assimilated in PSY4V2R2 in the Gulf Stream 1 region during the OND
2014 quarter............................................................................................................................................... 32
Figure 21: As Figure 13 in the Algerian (a and b), the gulf of Lion (c and d), the Ionian (e and f) and the Mersa Matruh (g and h)
regions........................................................................................................................................................ 34
Figure 22: As Figure 13 in the North West Indian (a and b), the South West Indian (c and d) and the East Indian (e and f)
regions........................................................................................................................................................ 36
Figure 23: Average difference MODEL-OBSERVATION of temperature (a, c, and e, in °C) and salinity (b, d, and f, in psu) in OND
2014 between the daily average hindcast PSY4V2R2 (a and b), PSY3V3R3 (c and d) and PSY2V4R4 (e and f) hindcast
products and all available T/S observations from the Coriolis database. The Mercator hindcast products are taken at the
time and location of the observations. Averages are performed in the 0-50m layer. The size of the pixel is proportional
to the number of observations used to compute the RMS in 2°x2° boxes. ............................................... 38
Figure 24: RMS temperature (°C) difference (model-observation) in OND 2014 between all available T/S observations from
the Coriolis database and the daily average hindcast PSY3V3R3 products on the left and hindcast PSY4V2R2 on the right
column colocalised with the observations. Averages are performed in the 0-50m layer (a and b) and in the 0-500m layer
(c and d). The size of the pixel is proportional to the number of observations used to compute the RMS in 2°x2° boxes.
.................................................................................................................................................................... 39
Figure 25: RMS salinity (psu) difference (model-observation) in OND 2014 between all available T/S observations from the
Coriolis database and the daily average hindcast PSY3V3R3 products (a and c) and hindcast PSY4V2R2 (b and d),
colocalised with the observations. Averages are performed in the 0-50m layer (a and b) and in the 0-500m layer (c and
d). The size of the pixel is proportional to the number of observations used to compute the RMS in 2°x2° boxes. 40
Figure 26 : OND 2014 global statistics of temperature (°C, a and c) and salinity (psu, b and d) averaged in 6 consecutive layers
from 0 to 5000m. RMS difference (a and b) and mean difference (observation-model, c and d) between all available T/S
observations from the CORIOLIS database and the daily average hindcast products PSY3V3R3 (red), PSY4V2R2 (blue)
and WOA09 climatology (grey) colocalised with the observations. NB: average on model levels is performed as an
intermediate step which reduces the artefacts of inhomogeneous density of observations on the vertical. 41
Figure 27: RMS difference (model-observation) of temperature (a and b, °C) and salinity (c and d, psu) in OND 2014 between
all available T/S observations from the Coriolis database and the daily average PSY2V4R4 hindcast products colocalised
with the observations in the 0-50m layer (a and c) and 0-500m layer (b and d)....................................... 42
Figure 28: Water masses (Theta, S) diagrams in OND 2014 in the Bay of Biscay (a to c), Gulf of Lion (d to f), Irminger Sea (g to
i) and Gulf of Cadiz (j to l), comparison between PSY3V3R3 (a, d, g, j), PSY4V2R2 (b, e, h, k) and PSY2V4R4 (c, f, I, and l)
in AMJ 2013. PSY2, PSY3 and PSY4: yellow dots; Levitus WOA09 climatology: red dots; in situ observations: blue dots.
.................................................................................................................................................................... 44
Figure 29 : Water masses (T, S) diagrams in the Western Tropical Atlantic (a to c) and in the Eastern Tropical Atlantic (d to f) in
OND 2014: for PSY3V3R3 (a, d); PSY4V2R2 (b, e); and PSY2V4R4 (c, f). PSY2, PSY3 and PSY4: yellow dots; Levitus WOA09
climatology; red dots, in situ observations: blue dots. .............................................................................. 46
Figure 30: Water masses (T, S) diagrams in South Africa (a and b), Kuroshio (c and d), and Gulf Stream region (e and f): for
PSY3V3R1 (a, c and e); PSY4V1R3 (b, d and f) in OND 2014. PSY3 and PSY4: yellow dots; Levitus WOA09 climatology: red
dots; in situ observations: blue dots. ......................................................................................................... 47
Figure 31 : Water masses (T, S) diagrams in Beaufort Sea for PSY3 (a) and PSY4 (d), salinity profiles in this region for PSY3 b)
and PSY4 e), and the same for the temperature profiles : PSY3 (c) and PSY4(f). Symbology : yellow dots; Levitus WOA09
climatology: red dots; in situ observations: blue dots. .............................................................................. 48
Figure 32 : RMS temperature (°C) differences between OSTIA daily analyses and PSY3V3R3 daily analyses (a); between OSTIA
and PSY4V2R2 (b), between OSTIA and PSY2V4R4 (c) in OND 2014. The Mercator Océan analyses are colocalised with
the satellite observations analyses. ........................................................................................................... 49
Figure 33: Mean SST (°C) daily differences in OND 2014 between OSTIA daily analyses and PSY3V3R3 daily analyses (a),
between OSTIA and PSY4V2R2 (b) and between OSTIA and PSY2V4R4 daily analyses (c)......................... 50
Figure 34: Near surface (15m) zonal current (m/s) comparison between the Mercator systems analyses and observed
velocities from drifters. In the left column: velocities collocated with drifter positions in OND 2014 for PSY3V3R3 (a),
PSY4V2R2 (c) and PSY2V4R4 (e). In the right column, zonal current from drifters in OND 2014 (b) at global scale, AOML
drifter climatology for OND with new drogue correction from Lumpkin & al, in preparation (d) and observed zonal
current in OND 2014 (f) over PSY2’s domain. ............................................................................................ 51
Figure 35 : In OND 2014, comparison of the mean relative velocity error between in situ AOML drifters and model data (a, c
and e) and mean zonal velocity bias between in situ AOML drifters with Mercator Ocean correction (see text) and
model data (b, d and f). a and b: PSY3V3R3, c and d: PSY4V2R3, e and f: PSY2V4R2. NB: zoom at 500% in order to see
the arrows. ................................................................................................................................................. 52

83
Figure 36: In OND 2014, comparison of the mean distance error in 1°x1° boxes after a 1-day drift (a and b), after a 3-days drift
(c and d), and after a 5-days drift (e and f).between AOML drifters’ trajectories and PSY3V3R3 trajectories (a, c, and e)
and between AOML drifters’ trajectories and PSY4V2R2 (b,d, and f)........................................................ 54
Figure 37:Cumulative Distribution Functions of the distance error (km, on the left) and the direction error (degrees, on the
right panel) after 1 day (blue), 3 days (green) and 5 days (red), between PSY4V1R3 forecast trajectories and actual
drifters trajectories..................................................................................................................................... 55
Figure 38: Arctic (a to c) and Antarctic (d to f) OND 2014 Sea ice cover fraction in PSY3V3R3 (a and d), in CERSAT observations
(b and e) and difference PSY3 – CERSAT (c and f). ..................................................................................... 56
Figure 39: Arctic (a to c) and Antarctic (d to f) OND 2014 Sea ice cover fraction in PSY4V2R2 (a and d), in CERSAT observations
(b and e) and difference PSY4 – CERSAT (c and f). ..................................................................................... 57
Figure 40: OND 2014 Arctic sea ice extent in PSY4V2R2 with overimposed climatological OND 1992-2010 sea ice fraction
(magenta line, > 15% ice concentration) (a) and NSIDC map of the sea ice extent in the Arctic for December 2014 in
comparison with a 1979-2000 median extend (b), the same for the Antarctic region for the PSY4V2R2 system c) and the
NSIDC estimation d) (ftp://sidads.colorado.edu/DATASETS/NOAA/G02135/Dec/). ................................. 58
Figure 41 : Mean bias (observation-model) (a), RMS error (b), correlation (c) between IBI36V2 and analysed L3 SST from
Météo-France CMS for the OND 2014 quarter. Same diagnostics for PSY2V4R4 (d,e,f) and PSY4V2R2 (g,h,i). Number of
Météo-France CMS observations for the OND 2014 quarter (j). ............................................................... 60
Figure 42 : Over the whole IBI36V4 domain (a and b) and over the Bay of Biscay (c and d): mean “IBI36V4 - observation”
salinity (psu) bias (red) and RMS error (blue) in OND 2014 (a and c), and mean profile from IBI36V4 (black), PSY2V4R4
(green) and from the observations (red) in OND 2014 (b and d). In the lower right corner of each plot: position of the
individual profiles....................................................................................................................................... 61
Figure 43: Over the whole IBI36V4 domain (a and b) and over the Bay of Biscay (c and d): mean “IBI36V4 - observation”
salinity (psu) bias (red) and RMS error (blue) in OND 2014 (a and c), and mean profile from IBI36V4 (black), PSY2V4R4
(green) and from the observations (red) in OND 2014 (b and d). In the lower right corner of each plot: position of the
individual profiles....................................................................................................................................... 62
Figure 44 : Mixed Layer Depth distribution in OND 2014 in IBI36V4 (a and b) and PSY2V4R4 (c and d), calculated from profiles
with the temperature criteria (difference of 0.2°C with the surface); the models are in grey, the observations in red. a
and c: classification for the whole IBI domain, b and d: classification for the Bay of Biscay ..................... 63
Figure 45 : For IBI36V4: RMS error (a, in cm) and correlation (b) for the non-tidal Sea Surface Elevation at tide gauges in OND
2014, for different regions and frequencies............................................................................................... 63
Figure 46 : For IBI36V4: Bias (observation-model) (a, d and g), RMS error (°C) (b, e and h) and correlation (c, f and i) of the Sea
Surface Temperature between IBI36 and moorings measurements in October (a to c), November (d to f) and December
2014 (g to i). ............................................................................................................................................... 65
Figure 47 : Surface temperature (°C) time series at a selection of moorings which locations are shown on the upper left panel
(no observation for Gulf of Lion)................................................................................................................ 65
Figure 48 : Chlorophyll-a concentration (mg/m
3
) in October(a and b), November (c and d) and December 2014 (e and f) for
the Mercator system BIOMER4 (a, c, and e) and Chlorophyll-a concentration from Globcolour (b, d and f). 66
Figure 49 : Probability Density Function (PDF) of Chl-a bias in log scale (log10(obs)-log10(model)) in the North Atlantic (30-
70N; 80W:20E) for October (a), November (b) and December (c) 2014.................................................... 67
Figure 50 : RMS difference between BIOMER and Globcolour Chl-a concentrations (mg/m
3
) in OND 2014...... 67
Figure 51: Schematic of the change in atmospheric forcings applied along the 14-day ocean forecast............. 68
Figure 52: Accuracy intercomparison in the North Atlantic region for PSY2V4R4 in temperature (a and c) and salinity (b and dl)
between hindcast (green), nowcast (yellow), 3-day (orange) and 6-day (red) forecast and WO09 climatology (grey).
Accuracy is measured by a mean difference (a and b) and by a rms difference (c and d) of temperature and salinity with
respect to all available observations from the CORIOLIS database averaged in 6 consecutive layers from 0 to 5000m. All
statistics are performed for the OND 2014 period. NB: average on model levels is performed as an intermediate step
which reduces the artefacts of inhomogeneous density of observations on the vertical. ........................ 69
Figure 53: Accuracy intercomparison in the Mediterranean Sea region for PSY2V4R4 in temperature (°C, a and c) and salinity
(psu, b and d) between hindcast, nowcast, 3-day and 6-day forecast and WO09 climatology. Accuracy is measured by a
rms difference (c and d) and by a mean difference (a and b) with respect to all available observations from the CORIOLIS
database averaged in 6 consecutive layers from 0 to 5000m. All statistics are performed for the OND 2014 period. NB:
average on model levels is performed as an intermediate step which reduces the artefacts of inhomogeneous density of
observations on the vertical....................................................................................................................... 70
Figure 54: (Obs-model) mean differences in temperature (°C, a, c, and e) and salinity (psu, b, d, and f) for PSY3V3R3 (red),
PSY2V4R4 (yellow) and PSY4V2R2 (blue) systems in the Tropical Atlantic (a and b), the Tropical Pacific (c and d) and the
Indian Ocean (e and f) in OND 2014........................................................................................................... 72

84
Figure 55: same as Figure 52 but for RMS statistics and for temperature (°C), PSY3V3R3 and PSY4V2R2 systems and the
Tropical Atlantic (a and b), the Tropical Pacific (c and d) and the Indian Ocean (e and f). The global statistics are also
shown for temperature (°C, g and h) and salinity (psu, I and j). The right column compares the analysis of the global ¼°
PSY3V3R3 (red) with the analysis of the global 1/12° PSY4V2R2 (blue). ................................................... 74
Figure 56: Temperature (left) and salinity (right) skill scores in 4°x4° bins and in the 0-500m layer in OND 2014 in PSY3V3R3 (a
and b) , PSY4V2R2 (c and d) and PSY2V4R4 (e and f). The skill illustrates the ability of the 3-days forecast to be closer to
in situ observations than the persistence of the analysis, see Equation 1. Yellow to red values indicate that the forecast
is more accurate than the persistence....................................................................................................... 75
Figure 57 : As Figure 56 but the reference is the WOA 2009 climatology. Temperature (a and c) and salinity (b and d) skill
scores are displayed for OND 2014, for PSY3V3R3 (a and b) and for PSY2V4R4 (c and d). The skill illustrates the ability of
the 3-days forecast to be closer to in situ observations than the climatology, see Equation 1. Yellow to red values
indicate that the forecast is more accurate than the climatology. ............................................................ 76
Figure 58: daily SST (°C) spatial mean for a one year period ending in AMJ 2013 for PSY2V4R4 (a), PSY3V3R3 (b), and
PSY4V2R1 until 31/10/2012 then PSY4V2R2 (c). Mercator Ocean systems are in black and Reynolds AVHRR ¼° analysis
in red. ......................................................................................................................................................... 77
Figure 59: eddy kinetic energy EKE (m²/s²) at 15m for PSY3V3R3 (a) and PSY4V2R2 (b) for OND 2014. ............ 78
Figure 60: Comparisons between OND 2014 mean temperature (°C, a to c) and salinity (psu, d to f) profiles in Mercator
systems (black): PSY2V4R4 (a and d), PSY3V3R3 (b and e) and PSY4V2R2 (c and f), and in the Levitus WOA05 (green) and
ARIVO (red) monthly climatologies. ........................................................................................................... 79
Figure 61: Sea ice area (a, 10
6
km2) and extent (b, 10
6
km2) in PSY3V3R3 (blue line), PSY4V1R3 until October then PSY4V2R2
(black line) and SSM/I observations (red line) for a one year period ending in OND 2014, in the Arctic (upper panel) and
Antarctic (lower panel)............................................................................................................................... 80
Figure 62 : illustration of QC: Quality test example chosen for windage (eg. 1%) we reject or correct a drift that differs little
from the windage (less than 70% of the drift angle <40 °)......................................................................... 88
Figure 63: Example of the surface currents Lagrangian quality control algorithm on a global map (bottom panel) and zooms
(upper panels). ........................................................................................................................................... 89
Figure 64: Illustration of the surface currents Lagrangian quality control algorithm.......................................... 90

85
II Annex B
II.1. Maps of regions for data assimilation statistics
II.1.1. Tropical and North Atlantic
1 Irminger Sea 12 Puerto Rico XBT
2 Iceland Basin 13 Dakar
3 Newfoundland-Iceland 14 Cape Verde XBT
4 Yoyo Pomme 15 Rio-La Coruna Woce
5 Gulf Stream2 16 Belem XBT
6 Gulf Stream1 XBT 17 Cayenne tide
7 North Medeira XBT 18 Sao Tome tide
8 Charleston tide 19 XBT - central SEC
9 Bermuda tide 20 Pirata
10 Gulf of Mexico 21 Rio-La Coruna
11 Florida Straits XBT 22 Ascension tide

86
I.1.1. Mediterranean Sea
1 Alboran 8 Ionian
2 Algerian 9 Egee
3 Lion 10 Ierepetra
4 Thyrrhenian 11 Rhodes
5 Adriatic 12 MersaMatruh
6 Otranto 13 Asia Minor
7 Sicily

87
I.1.2. Global ocean
1 Antarctic Circumpolar Current 13 Nino3
2 South Atlantic 14 Nino4
3 Falkland current 15 Nino6
4 South Atl. gyre 16 Nino5
5 Angola 17 South tropical Pacific
6 Benguela current 18 South Pacific Gyre
7 Aghulas region 19 Peru coast
8 Pacific Region 20 Chile coast
9 North Pacific gyre 21 Eastern Australia
10 California current 22 Indian Ocean
11 North Tropical Pacific 23 Tropical indian ocean
12 Nino1+2 24 South indian ocean

88
III Annex C
III.1. Quality control algorithm for the Mercator Ocean drifter data correction
(Eric Greiner)
Before estimating the bias, it is essential to conduct a quality control. We must consider an individual
monitoring of buoys, and a comparison with the geostrophy and windage. In real time, this is not possible,
and I propose below a simple test developed by position (date by date) which involves only the mean wind (2
days) and the buoy drift. Basically, we found drifters where drift is close to argue between 0.2 and 3% of the
wind (almost the same direction with a drag corresponding to a loss of drogue). For these buoys, if the
contamination is real, then the error due to the wind is important with respect to current real at 15m depth.
We test different values of windage (wind effect for a fraction of a given wind between 0.2% and 3%). If a
questionable observation is found for a given windage, we estimate a correction. We apply at the end an
average correction QC (windage among all acceptable). We although increase the error of observation. Note
that in delayed time, we could correct all the data from the buoy, at least in a 10-day window. Note however
that a buoy that has lost its drogue can give a good measure if the wind is low
• No anomaly : slippage correction of 0.07% of the 10m wind speed
• Windage > 0.2% or < 3% correction of 1% of windage
Figure 62 : illustration of QC: Quality test example chosen for windage (eg. 1%) we reject or correct a drift that differs little from
the windage (less than 70% of the drift angle <40 °)
Note that a correction of more than 3% is not normally possible (construction of the buoy). This may
correspond to breaking waves and swell. Between 2% and 3%, there is ambiguity between Stokes and
windage. In other words, it is likely that beyond 2%, we eliminate all or part of the effect of waves and swell. If
waves and swell are not aligned with the mean wind (swell remote for example), then the correction will be
approximate. Ideally, you should use the Stokes drift from a wave model like Wavewatch3.
When calculating the equivalent models with AOML positions, which were filtered to remove 36h gravity
waves and reduce positioning errors, we must :
• add 0.07% wind averaged over 48h 10m : slippage correction
• windage correction and modify the error

89
III.2. Algorithm of the Lagrangian verification of the Mercator Ocean surface
currents forecast
Algorithm of the Lagrangian verification of the Mercator Océan surface currents forecast. The Mercator Océan
surface currents quality control now combines two methods based on Eulerian and Lagrangian approaches
using the AOML drifters network.
The Lagrangian approach is slightly different from the Eulerian approach. In the Eulerian approach, the
consecutive positions of a drifter are considered as independent buoys recording velocity observations. Then,
these observations are compared to the modeled velocity.
We aim here at studying the trajectory of the buoy along with the trajectory of a modeled buoy which would
drift from the same starting point. The computation of the trajectory of the modeled drifter is made possible
by ARIANE (see ref 1 above)
The algorithm aims at producing the maps on Figure 63. It shows the mean 1-to-5-days distance error –that is
the distance between the modeled trajectory and the observed trajectory- in 1°x1° boxes. The mean D-days
distance error is computed by averaging the D-days distance errors computed for all the drifters that crossed
the box.
The individual points of a trajectory are not independent, merely because the location of a drifter is to a large
extent determined by its former location.
Figure 63: Example of the surface currents Lagrangian quality control algorithm on a global map (bottom panel) and zooms
(upper panels).
Let us consider the example above (Figure 64). The thin grey line represents the trajectory of the drifter on a
daily frequency. The thick ones represent the system drifting, starting from an observed point. Their colors
show the distance to the corresponding point in the observed trajectory after D-days. Considering only this
drifter, if we compute the D-days distance error starting from the t=0 observed point (01/02/2013 in the

90
example), we may compute it again only from the observed point t=D days. This way we may reasonably
assume the two distance errors are uncorrelated and use most of the data (see ref 2 above).
Figure 64: Illustration of the surface currents Lagrangian quality control algorithm.
References:
1- Ariane : utility developed at LPO (http://wwz.ifremer.fr/lpo_eng/Produits/Logiciels/ARIANE)
2- Scott, R. B., N. Ferry, M. Drevillon, C. N. Barron, N. C. Jourdain, J.-M. Lellouche, E. J. Metzger, M.-H. Rio,
and O. M. Smedstad, Estimates of surface drifter trajectories in the equatorial Atlantic: A multi-model
ensemble approach, Ocean Dynamics, 62, 1091-1109, 2012, doi:10.1007/s10236-012-0548-2.
3- M. Steele, J. Morison, W. Ermold, I. Rigor, and M. Ortmeyer(2004), Circulation of summer Pacific
halocline water in the Arctic Ocean., J. Geophys. Res., 109,C02707,doi:10.1029/2003JC0022009109
4- Jackson, J. M., E. C. Carmack, F. A. McLaughlin, S. E. Allen, and R. G. Ingram (2010), Identification,
characterization, and change of the near-surface temperature maximum in the Canada Basin, 1993–
2008, J. Geophys. Res., 115, C05021, doi:10.1029/2009JC005265.

91
PSY4PSY4V2R2 QUALITY SYNTHESIS, year 2414
Figure 1: Eddy kinetic energy (EKE) averaged over the year 2014,
calculated from the PSY4V2R2 daily 15m currents
The PSY4 global system with a Q/QR ° horizontal
resolution provides a good representation of the
global ocean circulation and the mesoscale features
(Figure Q). This is the Mercator system that exhibits
the best performance in terms of surface currents
(Figure R)
PSY4 also provides the best sea level anomalies
forecast in all regions of the ocean, except in the
Mediterranean where PSYR is the most reliable of
all Mercator Ocean systems.
Data assimilation in PSY4 is at the state-of-the-art
like his contemporary systems PSYR and PSY3.
PSY4 remains close to T and S in situ observations
on average between 0 and R000m. Differences with
in situ observations reach amplitudes greater than
Q ° C and 0.R psu only locally (in the thermocline,
in regions of high variability).
Figure 2: Relative error between the near surface(15 m) current velocities observed by drifting buoys and PSY4V2R2 velocities on average over the year 2014. The
red (blue) indicates that PSY4V2R2 underestimates (overestimates) the current. The arrows indicate the direction of the difference between observations and
PSY4V2R2, and are visible when zooming to 200%.
Globally, PSY4 underestimates the currents inside the gyres particularly the North and South Pacific and
overestimates in the Gulf of Alaska and in the following Tropical currents:
- For the Pacific: the west side of the Equatorial Counter Current and the east part of the South Equatorial
Current.
- For the Atlantic: the South Equatorial Current and the Brazil Current associated with the Equatorial
Counter Current.
The Circumpolar current is also overestimated in the South Pacific Ocean and the south Indian Ocean
In global average the sea surface temperature is too warm (about 0.Q°C), mainly in boreal fall and winter.
While the sea ice melts too much in the Arctic ice pack during summer, the sea ice extent is nevertheless
overestimated in Arctic during boreal summer. For the Antarctic, the extension is underestimated during
Austral summer and overestimated during Austral winter.

92
PSYR, the Atlantic and Mediterranean system with 1/12 ° horizontal resolution is operated since R00R by
Mercator. Designed as a test platform for PSY4, the global high resolution system at Q/QR°, PSYR’s
performance has been gradually improving over the versions (Figure Q). The current PSYRV4R4, for which a 7-
year time series is available, and the global Q/QR ° PSY4VRRR display very similar results. Both PSYR and
PSY4 have more accurate surface fields (U, V, SSH, SST) than PSY3 which has a ¼° horizontal resolution. In
R0Q4, PSYR still provides the initial conditions and the boundary conditions for the IBI system. The global
system PSY4 will progressively replace PSYR for IBI and for all Mercator users in R0Q6.
Figure 1: RMS difference between the SLA forecast of PSY2 systems since 2005 and SLA observations from Jason-1 and Jason-2, in an area near the southern
boundary of the domain in the Tropical Atlantic west between 5 ° S and 10 ° S.
Figure 2 : Mean and RMS deviation of
temperature (left) and salinity (right), for
Mercator systems with respect to CORIOLIS in
situ database in 2014, in the North East Atlantic
(top) and the Mediterranean (bottom).
With almost identical model and data
assimilation settings, the three
Mercator systems PSY3, PSYR and
PSY4 show similar performance with
respect to in situ profiles in R0Q4.
Globally, the high resolution systems
PSYR and PSY4 have slightly more
accurate water masses especially in
regions where finer structures need to
be resolved, for instance in the
Mediterranean Sea (Figure R).
PSYR is preferred to PSY4 in R0Q4,
as PSY4 is still close to his spin up
phase.
.
Discrepancies with in situ
observations reach amplitudes
greater than Q ° C and 0.R psu only
locally (in the thermocline, in regions
of high variability). Some dense
water masses are not represented by
the system, for example some of the
Levantine Intermediate Waters, the
waters of the Labrador Current. The
densest waters of the Bay of Biscay
are well represented. Deficits of salt
in the surface layers are generally
related to errors in atmospheric
forcing, while the denser bottom
waters are under too much mixing.

93
IBI36IBI36V4 R1 QUALITY SYNTHESIS, year 6014
IBI36 (Iberia-Biscay-Ireland) is
embedded into the PSYR
system (in R0Q4: PSYRV4R4)
and has no data assimilation.
IBI36 inherits from the water
masses characteristics of PSYR.
With higher spatial (Q/36 °) and
temporal resolution than PSYR
(hourly surface output), IBI36 is
able to represent the tides with a
mean square error between QQ
and QR cm for the semi-diurnal
component MR (Figure Q).
Higher errors appear locally for
instance on the Belgian coast
and in the German Bight. The
mean square error of residual
height relative to observations
is QQ cm on average (not
shown).
a b
Figure 1: Error on the amplitude of the tidal components of IBI36 calculated with a harmonic analysis
of the 2014 hourly surface elevation fields : (a). Difference in cm (zoom on the northern area ) between
the amplitude of the M2 tidal component of the FES2004 atlas and that of IBI36. The colored dots
represent the difference between the amplitude of the M2 tidal component observed at tide gauges and
that of IBI36 (b) RMS difference between the amplitudes of nine tidal components of tide gauges and
those of IBI36.
The IBI36 model also benefits from more realistic physics than the other Mercator Ocean systems (vertical mixing,
forcing bulk formulae and river inputs). Deviations from satellite observations of high-resolution SST are low on average
over a large part of the domain (Figure R). These discrepancies are greater on the shelf. Along the Moroccan and Iberian
coasts where upwelling occurs (the model tends to be too warm). In the Irish Sea, English Channel and North Sea, the
biases are linked to the mixing induced by the tides (the model tends to be too cold). Discrepancies are also noticeable in
the Mediterranean Sea (in the Alboran Sea). The monthly evolution of errors shows a lower performance of the model
from mid-spring to summer (maximum RMS) and in autumn (maximum bias in November).
Figure 2 : (a) Average differences (°C) between IBI36 and SST satellite observations (SST L3S for the year 2014). (b) Average number of satellite observations by
grid point and by month (top), monthly SST average (middle panel) and RMS (bottom panel) difference between IBI36 and satellite SST.
a b

94
Figure 1: Murphy Skill Score for temperature (left) and salinity (right) on the layer 0-500m for the year 2014.The score is one minus the ratio
between the 3-day forecast accuracy and the climatology accuracy. The accuracy is given by the mean square error between the in situ
observation (ARGO) and the forecast or the World Ocean Atlas 2009 climatology. Therefore a positive score implies good predictive ability of
the system.
The global ¼° PSY3 forecasts are clearly superior to climatology in the three oceans. The system’s accuracy
in terms of temperature (and to a lesser extent in terms of salinity, for which there are fewer observations) is
particularly good in the mid-latitudes. Some regions are an exception, where the climatology is closer to the
observations of salinity including: Arabian Sea, the Gulf of Guinea, the western Tropical Atlantic, the Japan
Sea, the Weddell Sea, the Indonesian Throughflow, Southern Australia. The 3-day forecast is better than the
persistence of the PSY3 analysis in certain areas only (not shown), particularly in the tropics.
a)
b)
Figure 2: a) PSY3V3R3 temperature anomalies between 0 and 500 metres (reference:
Glorys2v3 climatology), for February 2014, compared to b) that is the equivalent for ECMWF
ocean reanalysis.
PSY3 analyses are used to create the initial
conditions for the Météo-France ocean-
atmosphere coupled system that provides
seasonal forecasts. As illustrated on Figure R,
the “climatic” signal of PSY3 (temperature
and heat content) is comparable to ECMWF
reanalysis, particularly in zones of interest
for seasonal forecast skill (convergence
zones, tropics and equator).
PSY3 is very accurate in terms of
concentration and extent of sea ice. Melting
is nevertheless too strong in northern
summer, because of a warm bias in the
atmospheric forcings and relative high
accumulation of sea ice in the Canadian basin
due to a strong convergence which prevents
the melting. In southern hemisphere, the sea
ice extent is slightly overestimated during
austral winter and underestimated during
summer.
PSY3 provides a realistic estimate of the
surface currents (especially in terms of
direction), at the same order of PSY4,
especially in the equatorial countercurrents.
Subsurface equatorial currents are well
represented, mostly west of equatorial basins.

95
BIOMER4V1R1 QUALITY SYNTHESIS, year
2414
Figure 1: Concentration of chlorophyll a (mg Chl.mM3
) on average in 2014 (left) in the BIOMER system and (right) from the GlobColour
observations (ocean colour data)
The global system BIOMER4VQRQ at QO4° delivers weekly an estimate of the main biogeochemical
variables in the ocean using the PISCES model and the physical fields of PSY3V3R3.
As can be seen on Figure Q, BIOMER reproduces the large scale structures of chlorophyll
concentration corresponding to specific biogeographic zones (eg. double gyres, or the Antarctic
Circumpolar Current). However, important biases remain in the tropics especially in the western
tropical basins and in the southern ocean. Nitrate concentrations in the equatorial surface layer are too
high (FigureR).
Figure 2: Nitrate concentrations in the layer 0M10m depth (µmol N LM1
) on average over the year 2014 (left) in the BIOMER system and (right) from
the World Ocean Atlas 2009 climatology.
Figure 3: Median and 80th percentile time series for monthly
chlorophyll concentrations (mg.mM3
) in 2014 in the North Atlantic
Ocean. Blue: BIOMER4V1R1. Red : Globcolour
The seasonal cycle of surface chlorophyll is
relatively well reproduced in the North Atlantic
(Figure 3) with a spring and a fall bloom.
However the onset of the spring bloom is not
yet in line with the observations (one month
delay) and chlorophyll concentrations are too
low in summer.

QUOVADIS_NUM19_OND_2014

More Related Content

Viewers also liked (17)

Similar to QUOVADIS_NUM19_OND_2014 (20)

More from Mercator Ocean International (20)

Recently uploaded (20)

QUOVADIS_NUM19_OND_2014