Dear Mercatorian,
By growing, Mercator resolutely turns towards users.
Such logical development, which comes also within the
wish of creation of the future operational centre, requires
more than ever to offer quality products which will well
reply to the downstream demand.
Correctly integrating observations in the assimilation
system and qualifying their impact stay one of the key
points to reach this objective.
The stake is double: to maintain/improve the operational
system performance, we need to consolidate the
present by demonstrating the importance of the ocean
data measurements: satellite, Argo floats, moorings and
others in situ measurement instruments, ... Furthermore,
it is necessary to prepare the future by testing new
assimilation methods, by estimating the future
observation systems relevance and by developing
strategy for their integration in the prototypes...
This Newsletter comes within this scope. The first article will describe the mean dynamic topography of the Mediterranean
Sea, as a reference required for altimetric data assimilation. Current and future topographies are described, assessed and
intercomparated in Mersea framework. The second article associates 4D-variational method and Argo drifting floats to
examine the potential we may expect of vertical profiles of temperature and salinity to produce the oceanic state. Finally, the
last article describes the Mercator strategy, developped in the scope of an ESA study, for the future surface salinity
observation system: SMOS.
Among all of this, don't forget Europe, par excellence topically question at these days and which is approached in the News
through the first annual Mersea meeting, held in Toulouse from March 29 to March 31st.
Have a good read and see you for next issue with regional and coastal oceanography topic!
Making a Difference: Understanding the Upcycling and Recycling Difference
Mercator Ocean newsletter 17
1. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 1
GIP Mercator Ocean
Quarterly Newsletter
Editorial
Dear Mercatorian,
By growing, Mercator resolutely turns towards users.
Such logical development, which comes also within the
wish of creation of the future operational centre, requires
more than ever to offer quality products which will well
reply to the downstream demand.
Correctly integrating observations in the assimilation
system and qualifying their impact stay one of the key
points to reach this objective.
The stake is double: to maintain/improve the operational
system performance, we need to consolidate the
present by demonstrating the importance of the ocean
data measurements: satellite, Argo floats, moorings and
others in situ measurement instruments, ... Furthermore,
it is necessary to prepare the future by testing new
assimilation methods, by estimating the future
observation systems relevance and by developing
strategy for their integration in the prototypes...
This Newsletter comes within this scope. The first article will describe the mean dynamic topography of the Mediterranean
Sea, as a reference required for altimetric data assimilation. Current and future topographies are described, assessed and
intercomparated in Mersea framework. The second article associates 4D-variational method and Argo drifting floats to
examine the potential we may expect of vertical profiles of temperature and salinity to produce the oceanic state. Finally, the
last article describes the Mercator strategy, developped in the scope of an ESA study, for the future surface salinity
observation system: SMOS.
Among all of this, don't forget Europe, par excellence topically question at these days and which is approached in the News
through the first annual Mersea meeting, held in Toulouse from March 29 to March 31
st
.
Have a good read and see you for next issue with regional and coastal oceanography topic!
2. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 2
GIP Mercator Ocean
Contents
News : MERSEA : en route to European operational oceanography
By Pierre-Yves Le Traon Page 3
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the
Mediterranean Sea
By Fabrice Hernandez, Marie-Hélène Rio and Laurence Crosnier Page 4
Potential of the ARGO network to produce an oceanic synthesis of the hydrology and the
circulation in the North Atlantic using a 4D-variational method
By Gaël Forget, Bruno Ferron and Herlé Mercier Page 13
An observing system simulation experiment for SMOS: presentation of the study
By Florence Birol, Pierre Brasseur, Lionel Renault, Charles-Emmanuel Testut and Benoît
Tranchant Page 17
3. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 3
News : MERSEA, en route to European operational oceanography
News : MERSEA, en route to European operational oceanography
By Pierre-Yves Le Traon
The European project MERSEA aims at developing the ocean component of the GMES system (Global Monitoring for
Environment and Security). It should lead to the setting up of a European Center for Ocean Monitoring and Forecasting
(ECOMF) that will provide an integrated service for monitoring and forecasting the global and regional ocean. Many institutional
or private applications are foreseen: environment monitoring, maritime transport and security, pollution monitoring and
forecasting, sustainable management of ocean resources, offshore applications... MERSEA will allow us to address the
requirements from policy makers and European and international treaties on the marine environment. The integrated description
of the ocean state that MERSEA will provide will be, in addition, extremely beneficial to ocean, ecosystem and climate research.
MERSEA gathers about fifty European partners and the principal operational oceanography actors in Europe. MERSEA
activities are separated into three main themes:
• In situ and satellite observing systems and provision of data directly useable by models.
• The design, implementation and evaluation of a co-ordinated set of monitoring and forecast systems covering the
global ocean and the oceans and seas surrounding Europe.
• The development and demonstration of information products, applications and services.
MERSEA started in April 2004. The first annual meeting was held at the conference centre of Météo France from March 29 to
March 31
st
. A good opportunity for partners to meet, analyze work progress and work plan for the next years. The guiding line in
MERSEA is the definition of the final system (V3) that MERSEA will deliver in 2008 and the intermediate systems (V1, V2):
models, assimilation methodologies, in situ, satellite and forcing data, interfaces with applications and services. These
definitions are important milestones. They should allow the MERSEA steering committee to better target research activities.
First R&D activities are already showing promising results: setting up of a model global configuration, new global and regional
satellite products, improvements in the in situ monitoring systems, development of applications… At the same time, MERSEA
contributes to the development and/or the consolidation of the infrastructure of operational oceanography in Europe: global and
regional modelling/assimilation centres, data centres, data and product harmonization and distribution…
MERSEA steers well. Teams are in place and all partners share the same challenging objective: the setting up of a state-of-the
art operational oceanography in Europe. The stakes for the future of oceanography are large and MERSEA must be a success.
Over the next three years, MERSEA should pay particular attention to:
• Consolidating and sustaining the in situ and satellite observing systems. MERSEA must be proactive and should
make recommendations and propositions to national and European agencies responsible for these systems. Impact
studies should be carried out to demonstrate the impact of in situ/satellite data for ocean monitoring and forecasting
(the very subject of this Newsletter). Sustaining a European contribution to Argo and developing operational
oceanography satellite missions (high resolution altimetry, sea surface temperature, ocean colour) are among the
most urgent priorities (e.g. ESA Blue Sentinel).
• Consolidating the architecture of the integrated system: in situ and satellite data centers, global and regional
modeling and assimilation centers, transverse activities on data and product harmonization and distribution.
• Developing interfaces with applications from the institutional and private sectors. MERSEA will provide the
backbone information (core services) on the ocean state (hindcast, nowcast and forecast) required by applications
(downstream services). Some applications are developed directly in MERSEA to develop and test interfaces while
others are developed through the ESA GSE (GMES Service Element) projects or national projects.
• Answering requirements from European agencies (e.g. EEA, - European Environment Agency, EMSA - European
Maritime Safety Agency-) and policies/treaties on environment monitoring and maritime security (e.g. IPCC -
Intergovernmental Panel on Climate Change -, OSPAR - Commission for the Protection of the Marine Environment
of the North-East Atlantic -, HELCOM - Baltic Marine Environment Protection Commission-, ICES - International
Council for the exploration of the Sea-, Water Framework Directive). MERSEA should, in particular, develop and
monitor a series of relevant indicators on the ocean state.
4. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 4
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the Mediterranean Sea
The Mean Dynamic Topography used as a reference for altimetric
data assimilation in the Mediterranean Sea
By Fabrice Hernandez, Marie-Hélène Rio and Laurence Crosnier
Introduction
The Mean Dynamic Topography (henceforth MDT) needs to be defined accurately for operational oceanography as it provides
the appropriate mean height to add to altimetric Sea Level Anomalies (SLA) prior any assimilation of the absolute sea level. This
paper will discuss three different topics around the MDT in the Mediterranean Sea. The first section written by F. Hernandez
describes the MDT presently used in the MERCATOR PSY2v1 operational system. The second section, written by M.H. Rio,
presents in detail a new estimation by (Rio et al., 2005) of the MDT thanks to in situ measurements, altimetric data and a
general circulation model. This synthetic MDT estimation, hereafter called MED-RIO05, is likely to be used by MERCATOR in a
near future. The third part, written by L. Crosnier, deals with the differences between MDT within the MERSEA project and the
MED-RIO05 MDT. It also presents the impact of the MDT amplitude in the Gulf of Lion on winter convection.
The operational MERCATOR PSY2v1 Mean Dynamic Topography
By Fabrice Hernandez, with the contribution of Eric Greiner
The MERCATOR PSY2v1 prototype over the Mediterranean Sea was first using a three years averaged MDT of the PAM-5
simulation (free run forced with a perpetual year for atmospheric fluxes). This first MDT is presented in the annex of Newsletter8
(January 2003). It was not realistic and causing strong defaults in the system circulation. Others MDT issued from
more recent model simulations all had a different circulation pattern, as shown in Figure 1. However, some patterns were also
realistic.
MED-16 - ERA-40 ERA-40 forcing 1990 à 2000 (free slip) MED-1998-2002 ECMWF forcing (no-slip)
PAM22 – one year averaged in 1999 PAM22 – one year averaged in 2001
Figure 1
Mean Dynamic Topographies from four model simulations. MED-16 based on OPA model (from K. Béranger). PAM-22 is the
PAM simulation forced with ECMWF atmospheric fluxes (from R. Bourdalle-Badie and Y. Drillet)
5. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 5
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the Mediterranean Sea
Thus we tested the circulation patterns of the four surfaces by comparison to hydrological data. Each MDT plus the concurrent
SLA is compared to the corresponding hydrographic absolute dynamic heights (same time and position than the SLA). This
technique was developed by (Rio and Hernandez, 2004):
)t,x(h)t,x(SLA)x(TDM dyni ↔+
The residual differences: ( )SLATDMh +− are direct values of the MDT relative precision. At a given hydrographic data
location, the comparison of the four residual corresponding to the four MDT give the relative accuracy among these four
surfaces, whatever are the SLA and hydrographic data errors. Note that we assume here no correlation between the MDT and
SLA or hydrographic data error budgets.
In conclusion, we have determined a composite MDT (Figure 5, upper left panel) based on the four model MDT using the
relative differences as weights in the linear combination. Note also that taking into account the residuals allow to reduce
interannual differences between the expected mean value (the 1993-99 period over which the altimetric data have been
averaged) and the mean represented in each of the four surfaces, corresponding to averaging over shorter or shifted periods.
After presenting in this section the MDT presently used in the MERCATOR system, we will present in the next section the best
MDT candidate to be used in the near future in the MERCATOR system.
A synthetic Mean Dynamic Topography of the Mediterranean Sea estimated
from the combination of altimetric data, in situ measurements and a general
circulation model, by M.H. Rio
By Marie-Hélène Rio
Introduction
The Mediterranean MDT (Figure 1) described in this section was computed for the period 1993-1999 using a method described
and applied for the global ocean in (Rio and Hernandez, 2004). It is based on a synthetic technique which consists in
subtracting the oceanic variability as measured by altimetry to in situ measurements of the absolute oceanographic signal in
order to compute local “synthetic” estimates of the Mediterranean MDT. These synthetic estimates and the associated errors
are then used to correct a first guess of the mean field and map the MDT (and the corresponding mean geostrophic circulation)
on a 1/8
th
degree regular grid of the whole Mediterranean basin using a multivariate objective analysis (Bretherton et al., 1976;
Le Traon and Hernandez, 1992). Methods and results are thoroughly described in (Rio et al., 2005). In this note, we briefly
describe the different datasets used and mainly focus on a quantitative validation of the obtained Mediterranean Synthetic Mean
Dynamic Topography (SMDT), already named MED-RIO05, using independent in situ and altimetric data.
Figure 2
Synthetic Mean Dynamic Topography MED-RIO05 computed in this paper. Main features of the mean circulation are
superimposed schematically on the SMDT plot (see text for abbreviation definition)
6. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 6
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the Mediterranean Sea
Data
Three different datasets are needed for the estimation: altimetric data, in situ measurements, and an existing ‘first guess’ of the
Mediterranean MDT. The altimetric data available for this study are 1/8
th
degree weekly maps of (SLA) relative to a seven year
(1993-1999) mean profile computed at CLS and distributed by AVISO. Corresponding maps of geostrophic velocity anomalies
are computed by simple differentiation between adjacent grid points.
The in situ dataset consists in more than 120000 velocity measurements from satellite-tracked surface drifters available from 1
January 1993 to 11 November 1999 (Mauerhan, 2000; Poulain et al., 2004). The only geostrophic component of the in situ
velocities is extracted first applying a 36 hours low pass filter and further removing the Ekman component estimated using a
model by (Mauri and Poulain, 2004).
The first guess used in this study is the average over the period 1993-1999 of outputs from a MFSTEP (Mediterranean
Forecasting System: Toward Environmental Predictions) model run with no data assimilation (Demirov et al., 2003).
Validation using independent in situ data
Hydrological profiles
A total of 16 hydrological profiles were measured during the NORBAL-2 campaign in the Bonifacio gyre area, between the 7th
December 2001 and 11th December 2001. Figure 3 shows the altimetric SLA obtained averaging the maps available for the 5th
and 12th December 2001 (Figure 3a), as well as the corresponding absolute altimetric sea level maps computed using the first
guess (Figure 3b) or the MED-RIO05 SMDT (Figure 3c). The dynamic heights computed in the area relative to 500, 1000 and
1500 m are superimposed on the three plots respectively as squares, circles and triangles (for each reference depth, the mean
of the dynamic heights is readjusted to the SMDT mean). In all three cases, the Bonifacio gyre is clearly visible. However, when
considering only the altimetric SLA (Figure 3a) the gyre intensity is underestimated. The use of the first guess allows to increase
the gyre intensity but the core of the structure is still 5-6 cm higher than obtained by in situ measurements. The use of the
SMDT clearly allows to better reproduce the position and intensity of the gyre.
Figure 3
Absolute altimetric height in the Bonifacio gyre area, for the period 5th to 12th December 2001, computed using to reference the
altimetric anomalies: a) a zero mean field, b) the first guess, and c) the MED-RIO05 SMDT. In situ dynamic heights measured in
the area are superimposed as squares if computed relative to 500 m, as circles relative to 1000m,
and as triangles relative to 1500m.
Sea Surface Temperature satellite data
We display on Figure 4 the maps of SLA and SMDT+SLA obtained in the Alboran Sea on November, 27th, 2002 and January,
8th, 2003 as well as the corresponding maps of Sea Surface Temperature (SST) as obtained from NOAA-16 and -17 AVHRR
7. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 7
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the Mediterranean Sea
data. On both days the anomaly maps feature a cyclonic structure on the western side of the Alboran Sea, coupled with an
anticyclonic structure on the eastern side. The use of the SMDT to compute absolute altimetric sea level allows to highlight a
more complex situation: on November 27th , the absolute altimetric map clearly features a large meandering current all through
the Alboran Sea, with the WAG (Western Alboran Gyre) and EAG (Eastern Alboran Gyre) anticyclonic gyres, centered at 4.4°E
and 2.2°E, in very good agreement with SST data. On January 8th , the WAG has moved further east (centered at 3°E), the
EAG is still visible on 2°E and a new meander has formed at the western extremity of the area (centered at 4.8°E).
Consequently, on that day, three simultaneous anticyclonic gyres are present in the Alboran Sea. This is once more in very
good agreement with what can be see on the SST image obtained on the same day.
Figure 4
Maps of altimetric anomalies (top), absolute altimetric signal (middle) and Sea Surface Temperature (bottom) in the Alboran sea
on November 27th 2002 (left) and January 8th 2003 (right)
Drifting buoy velocities
Several surface drifting buoys have been deployed in the Adriatic from September 2002 to June 2003 as part of the
DOLCEVITA program (Lee, 2004) providing nearly 6400 velocity measurements in the Adriatic and Ionian Seas. Absolute
altimetric velocities at the time and position of the drifting buoy measurements are computed using various MDTs to reference
altimetric anomalies and RMS differences to the in situ absolute velocities are then computed. Results are displayed in Table 1.
The use of the MED-RIO05 SMDT allows to reduce the RMS differences and increase the regression slope between altimetric
and in situ velocities with respect to the use of the first guess or a zero-mean field.
Moreover, comparison was done to evaluate the impact of present GRACE (Gravity recovery and Climate Experiment) and
future GOCE (Gravity Field and Steady-State Ocean Circulation Explorer) geoid models to retrieve the Mediterranean MDT. A
large scale MDT was estimated from GRACE data subtracting the EIGEN-GRACE02S geoid from the altimetric Mean Sea
Surface CLS01 (Hernandez and Schaeffer, 2001) at scales larger than 333km. Then, because the resolution of future GOCE
data are expected to be of 100 km, we simulated a “GOCE” MDT filtering the SMDT scales shorter than 100 km. As presented
before, these two MDT are used for comparing absolute velocities, obtained by adding derived altimetric velocities, to buoy’s
velocities. The results, displayed in Table 1, allow to highlight two major points: first, the present resolution of GRACE geoids is
still too coarse to correctly estimate the MDT of the Mediterranean Sea. Better comparison results to observations are obtained
using no mean at all (0-mean column of Table 1). Second, even a high resolution geoid as GOCE will not allow to retrieve the
shortest scales of the Mediterranean MDT. Better comparison results are obtained with the SMDT, meaning that the scales
shorter than 100 km contained in the SMDT are significant.
8. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 8
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the Mediterranean Sea
SMDT First guess 0-mean MDT“GOCE” MDT GRACE
RMSU (cm/s) 9.6 10.4 10.9 10.4 13.6
RMSV (cm/s) 11.5 12.6 13.5 12.1 13.8
Regression slope 0.45 0.34 0.15 0.30 0.19
Table I
Comparison results: Root Mean Square differences (RMS) in cm/s and Regression slope (Rs) between in situ drifting buoy
velocities and the corresponding absolute altimetric velocities obtained adding to the altimetric anomalies different MDT
solutions.
Conclusion
An estimation of the Synthetic Mean Dynamic Topography MED-RIO05 (and the corresponding mean geostrophic circulation)
has been obtained for the 1993-1999 period, by correcting the mean deduced from the MFSTEP model with drifting buoy
velocity and altimetric data combined using the synthetic method. In addition to be a crucial issue for assimilating absolute
altimetric data into operational forecasting systems like MERCATOR or MFSTEP, the accurate estimate of the Mediterranean
MDT from in situ measurements and altimetric data as described here offers a unique opportunity for the validation of future
GOCE data. Furthermore, it will be complementary to the MDT directly deduced from GOCE geoid whose limits will be precisely
be reach in areas like the Mediterranean Sea, where mean spatial scales are expected to be less than 100 km.
The MDT in the Mediterranean Sea in the MERSEA project
By Laurence Crosnier
MDT differences between MFS, FOAM, MERCATOR and MED-RIO05
Within the MERSEA project, three ocean forecasting systems are releasing analysis and forecast of the ocean state in the
Mediterranean Sea. They are the Italian MFS system, the English FOAM system and the French MERCATOR system. Figure 5
shows the various MDT used as a reference for sea surface height anomalies data assimilation in the MERCATOR, MFS and
FOAM systems as well as the MED-RIO05 SMDT described in section 3. The basin average has been subtracted to all the MDT
in order for them to be comparable. Globally, the MDT from MFS, MERCATOR and MED-RIO05 show a realistic circulation,
with nevertheless a stronger north-south gradient in MFS and MED-RIO05 than in MERCATOR. The FOAM does not show a
realistic circulation in the Mediterranean Sea and will not be discussed further. The main differences between the MERCATOR,
MFS and MED-RIO05 MDTs are located in the following areas:
• In the Alboran Sea, the Anticyclonic Western and Eastern Alboran Gyre (WAG and EAG), as well as the Almeria
Oran Jet (AOJ) are not exactly located at the same place in the MERCATOR, MFS and MED-RIO05. MED-
RIO05 SMDT shows, in-between the WAG and EAG, a stronger cyclonic structure centered at 3˚E 36˚N, which
could be a realistic signature of a cyclonic eddy formed in this area.
• The Algerian Current (AC) is located further away from the coast in MFS than in MERCATOR and MED-RIO05.
Moreover, the AC is meandering in MFS whereas it has a more linear trajectory in MERCATOR. The AC in
MED-RIO05 is trapped along the coast as in MERCATOR, but is stronger and meanders.
• In the Ligurian-Provencal Sea, the cyclonic Gulf of Lion Gyre (GLG) is 8cm stronger in MFS and MED-RIO05
(which uses MFS as a first guess) than MERCATOR.
• In the Tyrrhenian Sea, the MED-RIO05 SMDT shows the Bonifacio Gyre (BG), which is as well in MERCATOR
but with a weaker intensity. The BG is not in MFS.
• MED-RIO05 SMDT shows a strong cyclonic structure in the area to the South and South-East of Sardinia, which
you do not find in MFS and MERCATOR.
• In the South of Sicily, the gradient associated with the AIS (Atlantic Ionian Stream) is stronger in MED-RIO05
than in MERCATOR and MFS, with more pronounced structures of the ABV (Adventure Bank Vortex), MCC
(Maltese Channel Crest), ISV (Ionian Shelfbreak Vortex) and MRV (Messina Rise Vortex).
• In the middle of the Ionian Sea, we notice a 10cm difference located at 17.5ْ E-34.5ْ N between MFS and
MERCATOR, due to an anticyclonic eddy in MFS, not appearing in MERCATOR and MED-RIO05. In this area,
MED-RIO05 shows a large anticyclonic structure, stronger than in MFS and MERCATOR. To the South along
9. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 9
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the Mediterranean Sea
the Lebanese coast, MERCATOR and MFS show a continuous current trapped along the coast. MED-RIO05
does not show such a current.
• In the Adriatic Sea, we notice three realistic cyclones in MED-RIO05. MFS and MERCATOR show the two most
southern structures, the one to the south being stronger in MFS.
• In the south of Crete, we notice that the Western Cretan Cyclone (WCC) is 8cm stronger in MFS and MED-
RIO05 than in MERCATOR. None of the MERCATOR and MFS MDT shows the Ierapetra Anticyclone (IA), the
Mersa Matruh Anticyclone (MMA) and the Shikmona Anticyclone (SA), which are resolved by MED-RIO05.
Nevertheless, the Rhodes Gyre (RG) is well captured by MERCATOR, MFS and MED-RIO05.
• In the south of the Levantine basin along the Egyptian coast at 28ْ E, MERCATOR shows a weak current
trapped along the coast. MFS has a stronger current which meanders. MED-RIO05 shows an even stronger
current, which also meanders.
• In the south-East of Cyprus and along the Lebanese coast, we find the Asia Minor current, stronger and further
away from the coast in MFS than MERCATOR. MED-RIO05 shows, as in MERCATOR, a current trapped along
the coast, but stronger than in MERCATOR.
Figure 5
MDT (in meters) in MERCATOR (top left), MFS (top right), FOAM (bottom left) and MED-RIO05 (bottom right)
MDT impact on the Gulf of Lion winter convection
In this section, we will look at the influence of the Gulf of Lion Gyre MDT amplitude on winter convection. First, we notice that
there is no convection (Figure 6) in the center of the Gulf of Lion Gyre (GLG) in MERCATOR during the 2004 and 2005 winters
(as well as in 2002, see newsletter numéro 11, article from K. Béranger) (lien a mettre), whereas convection is taking place in
MFS.
A salinity section in February 2004 at 5.5°E (not shown) shows a convection plume from surface to bottom at 42°N-5°W in MFS,
whereas salinity stratification has not been eroded in MERCATOR.
Western Mediterranean Deep Water (WMDW) formation in the GLG has been studied extensively by the (Medoc Group, 1970).
WMDW formation has been characterized by 3 different phases:.
10. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 10
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the Mediterranean Sea
Figure 6
Hovmuller diagram showing the maximum of the mixed layer depth (in meters) in the Gulf of Lion as a function of time (from
June 2003 till January 2005) and depth
1. A preconditioning phase, during which the presence of a cyclonic gyre in the center of the Gulf of Lion reduces the
stability of the surface layer (a distinct doming of isopycnals is present near 42ْN - 5ْW).
2. A violent mixing phase, triggered by the onset of the Mistral (dry continental north-westerly winds blowing from the
Rhone river valley). The convection takes place in the center of the gyre. The dense surface water mixes with the
saltier but warmer subsurface water.
3. Finally, a sinking/spreading phase. The newly formed WMDM leaves the formation area, while water advected by
the Ligurian-Provencal Current replaces the surface and intermediate waters and again, a three layer stratification is
re-formed.
We noticed earlier that the GLG amplitude is weaker in the MERCATOR MDT than in the MFS one. Furthermore, we notice on
Figure 7 (middle panel) that the annual mean salinity across the 5.5°E vertical section shows isohalines in a doming shape at
the surface at 42.2°N-5.5°W, as in the MEDATLAS climatology (Figure 7, right panel). The same salinity section in MERCATOR
(Figure 7, left panel) shows a smaller and deeper doming, with isohalines that do not outcrop. The more pronounced doming in
MFS than MERCATOR is most likely related to stronger cyclonic circulation in the MDT from MFS than MERCATOR..
Figure 7
Annual mean (June 2003- June 2004) salinity vertical section as a function of latitude and log10(depth) at 5.5 E in
MERCATOR (left), MFS (middle) and MEDATLAS (right).
The total heat loss (the heat flux term includes the heat flux through the relaxation terms) during winter 2004 is larger by
80W/m2 in MFS than MERCATOR in the GLG area (Figure 8). Again, this is related to the appearance of convection in MFS
and not in MERCATOR. The wind stress amplitude in the GLG during winter 2004 is the same in MERCATOR and MFS (not
shown).
11. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 11
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the Mediterranean Sea
Figure 8
Total Net Heat flux (including relaxation terms) in W/m2 during winter 2004 (January to March) in MFS (left)
and MERCATOR (right)
In conclusion, the MDT differences in the GLG area between MFS and MERCATOR can explain partly the absence of
convection in MERCATOR during the 2004 and 2005 winters. First, the stronger MFS MDT GLG gathers the isotherms and
isohalines in a doming shape at the center of the gyre. The stratification brought closer to the surface is thus more likely to get
eroded by the Mistral wind and convection is more likely to happen in MFS than in MERCATOR. Second, the surface heat flux
loss during the 2004 winter is larger in MFS in the GLG than in MERCATOR. This is going along with the convection event in
MFS and with the fact that MFS uses a Bulk formulae heat flux formulation (MERCATOR does not use such a formulation).
When convection starts, a mixing of the subsurface warmer waters with the cold surface waters occurs. The warmer surface
waters release their heat to the atmosphere. Such a heat loss goes on as long as the ocean is warmer than the atmosphere
thanks to the Bulk Formulae flux formulation in MFS. In other words, in MFS, the cold Mistral wind blowing over the GLG is
eroding the surface ocean stratification from the doming shape, triggering convection in the middle of the gyre as well as a heat
flux loss (through latent and sensible heat fluxes) from the ocean to the atmosphere. Finally, the temperature from an observed
XBT section across the Gulf of Lion on February 22 2005 (Figure 9) shows a rather homogeneous water column from the
surface to 900 meters depth at 42°N, whereas the same section in the MERCATOR model shows a too stratified water column.
The same section in the MFS model (not shown) shows a homogeneous water column similar to the observations, suggesting
that the convection process in MFS is more realistic than in MERCATOR.
Figure 9
Left Panel : Observed In situ temperature (CORIOLIS) along a XBT track across the Gulf of Lion on February 22 2005. Right
Panel : Potential temperature in the MERCATOR model across the same section at the same date.
Conclusion
Three different consecutive studies perform about the MDT in the Mediterranean Sea were presented here. First, due to the
necessity of a mean field for assimilating altimetric SLA in the MERCATOR PSY2v1 prototype, a MDT from model simulations
was derived by E. Greiner and F. Hernandez. More recently the Synthetic MED-RIO05 MDT has been estimated by M.-H. Rio.
And then, the comparison study between Mediterranean forecasting systems (in the framework of the E.U. MERSEA project),
showed that the MERCATOR MDT seems to have a too weak cyclonic gyre in the Gulf of Lion, preventing winter convection
from happening. MFS and MED-RIO05 (which uses MFS MDT as a first guess) present a stronger cyclonic gyre. Model
comparison to observations (XBT section) suggested that the convection process in MFS is more realistic than in MERCATOR.
12. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 12
The Mean Dynamic Topography used as a reference for altimetric data assimilation in the Mediterranean Sea
Moreover, the MERCATOR team might switch and use in the near future the MED-RIO05 SMDT in the Mediterranean basin.
The question remains open whether all the non permanent structures as the the Ierapetra Anticyclone (IA), the Mersa Matruh
Anticyclone (MMA) and the Shikmona Anticyclone (SA) which are present in the 1993-1999 MED-RIO05 SMDT, would
perturbate the model physics.
References
Bretherton, F.P., R.E. Davis, et C.B. Fandry, 1976: A technique for objective analysis and design of oceanographic experiments
applied to MODE-73, Deep-Sea Research, 23, 559-582.
Demirov, E., N. Pinardi, C. Fratianni, M. Tonani, L. Giacomelli, et P. De Mey, 2003: Assimilation scheme of the Mediterranean
Forecasting System: operational implementation, Ann. Geophysicae, 21, 189-204.
Hernandez, F., et P. Schaeffer, 2001: The CLS01 Mean Sea Surface: A validation with the GSFC00.1 surface. Rapport n°, édité
par CLS, Ramonville St Agne. pp. 14.
Le Traon, P.-Y., et F. Hernandez, 1992: Mapping of the oceanic mesoscale circulation: validation of satellite altimetry using
surface drifters, Journal of Atmospheric and Oceanic Technology, 9, 687-698.
Lee, C., 2004: Multi-disciplinary perspectives on a wintertime bora wind event-Intensive studies of the Northern Adriatic, EOS, in
preparation.
Mauerhan, T.A., 2000: Drifter Observations of the Mediterranean Sea Surface Circulation, pp. 111. Thèse de: Master of Science
in Physical Oceanography. Naval Postgraduate school, Monterey, CA.
Mauri, E., et P.-M. Poulain, 2004: Wind-driven currents in Mediterranean drifter data. Rapport n° OGS Tech. Report 1/2004 -
OGA-1, édité par OGS, Trieste, Italy. pp. 25.
Medoc group, 1970: Observation of formation of deep water in the Mediterranean Sea, Nature, 227, 1037-1040.
Poulain, P.-M., R. Barbanti, R. Cecco, C. Fayos, E. Mauri, L. Ursella, et P. Zanasca, 2004: Mediterranean surface drifter
database: 2 June 1986 to 11 November 1999, dans CD-ROM, édité, OGS,Trieste, Italy.
Rio, M.-H., et F. Hernandez, 2004: A Mean Dynamic Topography computed over the world ocean from altimetry, in situ
measurements and a geoid model, Journal of Geophysical Research, 109 (C12), C12032 1-19 - doi10.1029/2003JC002226.
Rio, M.-H., P.-M. Poulain, A. Pascual, E. Mauri, G. Larnicol, et R. Santoleri, 2005: A Mean Dynamic Topography of the
Mediterranean Sea computed from altimetric data and in situ measurements., Journal of Marine Systems (accepted).
13. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 13
ARGO network for an oceanic synthesis of circulation in the North Atlantic using a 4D-variational method
Potential of the ARGO network to produce an oceanic synthesis of
the hydrology and the circulation in the North Atlantic using a 4D-
variational method
By Gaël Forget, Bruno Ferron and Herlé Mercier
Introduction
The ARGO network is composed of drifting floats that provide temperature and salinity profiles from 2000 m up to the surface
every 10 days. With a final objective of 3° average horizontal resolution over all the oceans (more than half are already
deployed), this network will provide a quantity of information on the heat and salt content variability never reached before.
Floats also provide valuable information on water mass motions at their drifting depth.
The objective of this work is to examine the potential we may expect from an assimilation of vertical profiles of temperature and
salinity of ARGO floats to produce analyses of the oceanic hydrology and circulation. Given the Lagrangian character of the
observations and their inhomogeneous space distributions, the 4D-variational (4D-var) assimilation is a suitable method to
produce a data synthesis. Indeed, this method propagates the information contained in the observations following the model
dynamics (advection and waves) and it takes into account the observations at their exact date and position. The method and the
observations used here will be used by Mercator in the future.
Direct and adjoint models
We use a general circulation model (Marshall et al.1997ab) of low resolution (1° on the horizontal and 23 vertical levels). The
exercise is carried out on a North Atlantic configuration that extends from 20°S to 70°N. The model uses a no-slip lateral
boundary condition, a convective adjustment in case of hydrostatic instabilities, a vertical turbulent viscosity (resp. horizontal) of
10-3m2s-1 (resp. 2x104m
2
s
-1
), a vertical turbulent diffusivity (resp. horizontal) of 10
-5
m
2
s
-1
(resp. 103m
2
s
-1
).
The model is forced by the ERA15 ECMWF reanalyses. Temperature and salinity at the sea surface and along the artificially
closed boundaries are restored towards the Reynaud et al. (1998)'s climatology. The model is spun-up from rest over the
period 1979 to 1991. Tangent-linear and adjoint models are produced by the Tangent linear and Adjoint Model Compiler
(TAMC; Giering and Kaminski, 1998). They are used by the assimilation algorithm. The adjoint code of the model thus produced
is checked each time a modification is made in a subroutine of the direct model. We used the non incremental 4Dvar
assimilation technique. The assimilation window is one year length.
Method
In order to objectively quantify the constraint that the hydrology of the ARGO network produces for the reconstruction of the
oceanic state of the ocean using a 4D-var assimilation, a bunch of twin experiments were carried out. These experiments
consisted in:
• choosing arbitrarily a one year period (here summer 1987 to summer 1988) of the model spin-up as the oceanic
state of reference; the summer 1987 represents the reference initial conditions in temperature Tr and salinity Sr.
• simulating a network of Lagrangian floats that produce synthetic profiles of temperature and salinity every 10
days in the oceanic state of reference.
• Taking as a first guess temperature Tb and salinity Sb the summer 1989 fields. Tb and Sb correspond to our a
priori knowledge of the summer 1987 hydrology; Although the forcings remain the same as those used to
produce the oceanic state of reference, the one-year model integration from Tb and Sb produces a really
different oceanic state from the reference one (See the differences (Tb-Tr) and (Sb-Sr) in Fig 2a and 3a).
• adding to the synthetic profiles a Gaussian noise which amplitude is as large as the differences (Tb-Tr) and (Sb-
Sr); this noise is uncorrelated in space and time and simulates the instrumental error and the representativity
error that real observations would have.
• assimilating the synthetic profiles over the one year period in order to find optimal summer 1987 initial
conditions in temperature and salinity (control variables) that produce the best estimate of the reference oceanic
state. Tb and Sb are the first guess of the optimal initial conditions.
14. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 14
ARGO network for an oceanic synthesis of circulation in the North Atlantic using a 4D-variational method
The assimilation proceeds iteratively. For a given iteration i, the integration of the full direct model starting from the initial state
(Ti, Si) provides a cost function Jo(i) that sums the quadratic misfits of the model to the observations weighted by a diagonal
error covariance matrix R. The Jo(i) minimisation constrains the model trajectory. A cost function Jb(i) is added to Jo(i) that
sums the quadratic deviations of the initial state (Ti, Si) to the background (Tb, Sb) weighted by a diagonal error covariance
matrix B. The Jb(i) minimisation contrains the initial condition to stay close to the first guess. The reverse integration with the
adjoint model and the use of a descent algorithm provide new initial conditions in temperature and salinity used as starting point
for the iteration i+1 such that: Jo(i+1)+Jb(i+1)<Jo(i)+Jb(i). When the difference in cost function between two consecutive
iterations becomes sufficiently small, the iterative process is stopped and the last iteration represents the optimal initial
conditions. The matrix R is calculated for each model vertical level from the median value of the standard deviations of
Reynaud et al. (1998)'s climatology. The matrix B is equal to the inverse of the variances associated to (Tb-Tr) and (Sb-Sr) at
each vertical level.
When the intensity of the velocity field at the drift depth is low, a float samples several times the same water column before
being advected to the next grid point (almost 12 profiles available for a model grid of 100km length when the drift velocity is 1
cm/s). In this case, a simple average of the repeated profiles would filter the random noise. When the dynamics is faster, a float
will be able to cross a grid point without making any measurement. In this case, the 4D-variational method shows all its
importance to filter the noise. Indeed, it transmits the information contained in the measurements following the physics of the
model, which allows correlations between profiles that obey to the same dynamics (advection and waves). The longer the
assimilation window length, the further the information contained in measurements is carried away from its origin, and the larger
the scales of the initial state changes. In order to take advantage of these properties, data are assimilated over a one-year long
window length. On one hand, a longer window length requires taking the model error (e.g. forcings) into account. On the other
hand, a shorter window length decreases both the propagation of the information through the adjoint but also the number of
observations. A shorter window length has a cost function more quadratic what facilitates the search of the optimal initial state.
Results
We now illustrate some results obtained when using a network of 300 floats that corresponds to the ARGO spatial density on
the North Atlantic and a drifting depth of 800m (fig. 1). The first guess is composed of summer 1989 temperature and salinity
fields.
Figure 1
Positions and a number (grey scale) of float profiles available per grid point over the period summer 1987-summer 1988.
Month 12 of the model integration starting from the first guess and with the summer 1987-summer 1988 surface forcing fields,
show large scale differences in temperature at 847m with the state of reference. Differences locally exceed 1°C north of 10°N
(Fig. 2a). Further south as well as in the Norwegian Sea, differences are much weaker because of restoring conditions along
the artificially closed boundaries. After profiles of temperature and salinity are assimilated, most of the initial temperature
differences are reduced by a factor of 2 to 5 (fig. 2b). This is also verified for the previous months. However, in some regions,
the initial temperature differences remain high since no information is available to constrain the temperature. The salinity
behaves similarly to the temperature.
15. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 15
ARGO network for an oceanic synthesis of circulation in the North Atlantic using a 4D-variational method
Hydrology analyses are sensitive to the number of floats. A reduction by 3 of the float density (i.e. 100 floats) increases the
number of un-sampled regions and produces residuals often larger than 0.4°C in the open ocean whereas a increase by 3 (900
floats) leads to residuals everywhere lower than 0.2°C except in a few coastal areas (fig. 2cd).
Figure 2
Differences in temperature at 847m for month 12 of model integration between the first guess or an analysis (solution after
assimilation) and the state of reference. The analysis is produced by an assimilation of profiles of temperature and salinity
generated by lagrangian floats in the state of reference. Differences are calculated with: a) the first guess, b) the analysis
produced with 300 floats (typical horizontal resolution of ARGO for the North Atlantic), c) the analysis produced with 100 floats,
d) the analysis produced with 900 floats. The contour interval for the positive values is 0.2°C.
Although the assimilated profiles directly constrain the density field, velocities are also modified during the model integration of
the optimal initial state by adjustment to the density field. Thus, one can legitimately expect from the assimilation a constraint
on the model circulation. At month 12 of model integration, the differences in meridional overturning cell between the first guess
and the state of reference show increased transports in the deep branch (below 1500m) between 10°S and 40°N and a
decrease north of 50°N (fig. 3a). Closer to the surface, only the region from equator to 15°N shows significant differences. After
assimilation, the initial differences in meridional overturning are divided by a factor 2 to 4 (fig. 3b). In these twin experiments, the
meridional overturning cell is constrained by the data collected over one year from the 300 floats. The barotropic circulation is
constrained similarly (not shown).
16. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 16
ARGO network for an oceanic synthesis of circulation in the North Atlantic using a 4D-variational method
Figure 3
Difference in meridional overturning (units 106m3s-1) for month 12 of model integration between the oceanic state of reference
and a) the first guess, b) the analysis produced with 300 floats.
Conclusion
In twin experiments, the true oceanic state is known and errors on observations as well in model (null in this study) are
controlled. Thus, it is possible to diagnose objectively the quality of the analyses produced by the assimilation at each grid point
and time step. This type of diagnostic in the case of real data assimilation can be calculated only where and when independent
observations (i.e. not assimilated) are available.
With a one year assimilation window length, our twin experiments show that temperature and salinity profiles of an ARGO-type
float network constrain the stratification and the circulation of a low resolution model. This constraint acts even if the noise on
observations has an amplitude as large as the difference between the first guess and the true solution. Thus, the assimilation
manages to filter the noise efficiently.
A major difference between our twin experiments and real data assimilation experiments comes from the existence of model
error. Although perfectible (error covariance matrices are diagonal in this study), the application of our assimilation scheme to
real observations still show a strong constraint of the stratification and circulation if we compare the analyses to independent
observations (Forget 2005)
References
Forget, G. 2005 : Profils ARGO et assimilation 4Dvar pour le suivi climatique de l’océan Nord Atlantique, Thèse de l’Université
de Bretagne Occidentale, 140pp.
Giering, R., and T. Kaminski, 1998: Recipes for adjoint code construction. ACM Trans. Math., 24, 437-474.
Marshall, J., C. Hill, L. Perelman, and A. Adcroft, 1997a: Hydrostatic, quasi-hydrostatic, and non-hydrostatic ocean modeling. J.
Geophys. Res., 102, 5733-5752.
Marshall, J., A. Adcroft, C. Hill, L. Perelman, and C. Heisey, 1997b: A finite-volume, incompressible Navier-Stokes model for
studies of the ocean on parallel computers, J. Geophys. Res., 102, 5753-5766.
Reynaud, T., P. Legrand, H. Mercier, and B. Barnier: A new analysis of hydrographic data in the Atlantic and its application to
an inverse modelling study, Intern. WOCE News., 32, 29-31
a b
17. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 17
An observing system simulation experiment for SMOS: presentation of the study
An observing system simulation experiment for SMOS:
presentation of the study
By Florence Birol, Pierre Brasseur, Lionel Renault, Charles-Emmanuel Testut and Benoit
Tranchant.
Introduction
Two parameters determine the ocean density at a given pressure: temperature and salinity. It is well known that temperature
has important effects in determining the ocean circulation and thereby the climate system. Recent studies suggest that salinity is
not only a tracer of the ocean variability but also plays an active role on the ocean dynamics: although salinity has little direct
effect on the atmosphere, its variability affects the circulation through density effects (Cooper, 1988; Roemmich et al., 1994). In
the North Atlantic, sea surface salinity (SSS) modulates the intensity of deep convection and then the thermohaline circulation.
In tropical oceans, salinity impacts vertical stratification, and can thus favour or inhibit vertical mixing. Vertical distribution of
salinity also impacts the surface layer momentum budget, by affecting mixed layer depth and modulating the response to wind
forcing.
It is therefore important for any ocean prediction system to represent this field with a reasonable accuracy. Unfortunately, a well-
known deficiency of ocean models is their tendency to produce spurious drifts in salinity fields. This is mainly due to errors in
freshwater and heat fluxes. Through mixing, errors in the properties acquired at the ocean surface are then transferred to
subsurface and deep layers (Paiva and Chassignet, 2001). Note also that inaccurate initial conditions and model deficiencies
(advection, mixing, or entrainment terms, but also a too low resolution) contribute also largely to the salinity error budget. As a
result, the vertical salinity error structure is complex.
Context of the study
The MERSEA intercomparison exercise has shown that problems in the surface water masses representation is a common
feature of different ocean forecasting systems: MERCATOR, FOAM, TOPAZ and MFSTEP (L. Crosnier, personal
communication). Assimilation of salinity observations would help operational systems to better represent water masses.
Unfortunately, unlike other ocean parameters (as sea level anomalies or sea surface temperature), salinity has been sparsely
measured at sea, limited mostly to summertime observations or commercial shipping lanes. However, over the last decade, it
has been demonstrated that SSS can be measured from space and two new missions will allow to map sea surface salinity in
the near future: the European Space Agency’s SMOS (Soil Moisture and Ocean Salinity, planned for 2007) mission and the
proposed US/Argentinean AQUARIUS mission (planned for 2008). The main characteristics of these two missions are
summarised in Table 1.
SMOS Aquarius
Scientific objectives Soil moisture and SSS SSS
Measurements goals Accuracy of 0.1 psu for a 10-30 days
average and for an area of 200x200 km
Accuracy of 0.2 psu for a monthly average
and for an area of 100x100 km.
Mission characteristics Global coverage every 3 days and ~45
km resolution.
Global coverage every 8 days and ~60 km
resolution.
Table I
SMOS and Aquarius mission main characteristics
A major difficulty affecting each observing system is the definition of a suitable observational network. In the context of SMOS
and Aquarius, the product resolution will be strongly constrained by technical difficulties: because of the complexity of SSS
retrieval, the measurement accuracy for a single pixel will be around 0.5-1.5 psu. The SMOS products accuracy requirement is
specified as 0.1 psu for a 10 days and 2 degree by 2 degree resolution. It has been shown that this level of accuracy can be
obtained only by spatial and temporal averaging of the measurements. An important question we need to address is then: What
is the best strategy to optimally use the future SMOS and Aquarius data in the context of ocean prediction systems, from the
perspective of monitoring the mesoscale ocean circulation? One efficient way to address this question is by conducting
Observing System Simulation Experiments (OSSE).
18. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 18
An observing system simulation experiment for SMOS: presentation of the study
Objective and methodology
Since 2004, the Mercator assimilation team is involved in the ESA’s study of key concepts concerning the future operational use
of the satellite derived SSS products. The main objective of this new activity is to evaluate the difficulties in, and the benefit
from, the assimilation of SSS data products, as expected from SMOS and AQUARIUS, for ocean prediction systems. Another
motivation is to assess and compare the usefulness of the different data combination strategies (for SMOS/Aquarius).
The SMOS Observing System Simulation Experiment (OSSE) approach will consist of performing realistic numerical
assimilation experiments with the aim of quantifying the contribution of future SSS products in addition to data which are already
included in the Mercator assimilation system (SLA, SST, in situ profiles).
The approach consists of:
1. Simulating realistic pseudo-SSS products
2. Performing a simulation including the pseudo-SSS products in the assimilation system.
3. Assessing the contribution of the pseudo-SSS products through the comparison of different simulated fields with the
fields obtained when pseudo-SSS products are not included in the assimilation system. This contribution will be
analysed for both surface and subsurface layers and on a regional basis (MNATL configuration).
The approach adopted in this study is based on a review of state-of-the-art methods recently developed for SSS data
assimilation. We summarise the main points of this review in the next section, while an extended description of the results of the
SMOS OSSEs will be presented in a forthcoming Mercator Newsletter.
Approaches for sea surface salinity data assimilation
Historically, the need for accurate salinity information to reconstruct the state of the ocean was pointed out by Cooper (1988) in
the context of data assimilation. Many studies have shown the importance of salinity correction in ocean assimilation schemes.
However, because of salinity data scarcity, very few projects assimilating this kind of observations have been undertaken so far.
To our knowledge, the first impact study of observed SSS on a model performance was performed by Reynolds et al. (1998)
based on a simple Newtonian relaxation method. This method only allows a control on SSS, and assumes physical adjustment.
However, the results showed no adjustment at depth and that the control at the surface was impaired by the retroaction of the
subsurface layers. Durand et al. (2002, 2003) have shown that the use of an advanced multivariate assimilation method is
necessary to correctly control all the state variables (at the surface as well as in subsurface) related to SSS. Three-dimensional
multivariate error covariances allows to extrapolate SSS information into 3D and multi-parameters information (Figure 1).
Ocean
mixed
layer
SSS information
3D and multi-parameters
information
Figure 1
Simple scheme of the multivariate assimilation of SST and SSS data.
The benefit expected from the assimilation of SSS data products was also evaluated in the framework of the European TOPAZ
project (TOPAZ, Scientific final report, 2003). Based on the SEEK filter (Singular Evolutive Extended Kalman filter; Pham et al.,
1998), experiments were performed to examine the impact of the assimilation of SSS data on the North Atlantic Ocean state
estimation. High resolution altimetric data and sea surface temperature (SST) were assimilated in addition to climatological
SSS. The delicate issue with multivariate assimilation is that any misspecification in the multivariate error covariance may lead
to inappropriate corrections of the forecast fields. This study has shown that the joint assimilation of SST and SSS is able to
correct many deficiencies, even with a fairly large observation error variance. However, it requires to properly assimilate the low
resolution SSS data to avoid spurious effect. In particular, it is important to well estimate the observation operator used in the
19. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 19
An observing system simulation experiment for SMOS: presentation of the study
analysis scheme by computing the model equivalent (temporally and spatially) to the SSS data: in that way, the model
mesoscale signal can not be influenced by SSS.
A general conclusion of recent studies is the importance of the assimilation of different data types, each bringing a specific
contribution to the identification of the ocean state. It is demonstrated to be particularly useful in compensating in weakly
specified cross-field error covariances. Typically, assimilation of altimetric sea level anomalies and in situ profiles is relevant to
better control the subsurface layers, whereas SSS and SST data are intimately linked with the mixed layer behaviour. This last
remark suggests to use an assimilation scheme able to take into account a high number of observations..
Conclusion
The OSSE presented here will be conducted using the new generation Mercator assimilation scheme (SAM2). SAM2 is
particularly suitable for SSS assimilation: this scheme is multivariate, potentially allows dynamical error propagation and is able
to take into account a high number of observations. Note that this scheme is based on the SEEK filter which has already been
used to assimilate pseudo-SSS data (Durand, 2003; the TOPAZ project). The first step of this study has consisted to update
SAM2 to assimilate SSS in addition to conventional data and to validate this new system.
The assimilation of SSS data also assumes the introduction of new information at low resolution in the system, requiring a
specific parameterisation of the analysis scheme. In addition, we have also decided to introduce a representativeness error,
estimated from the SSS data set, and which reflects the model capacity to represent the observed signal (Figure 2). The error
shown on Figure 3 is estimated from the difference between the model and the Levitus climatological SSS and from physical
considerations. Large errors occur where the model mesoscale signal is large: in the Gulf Stream and frontal regions.
Figure 2
Mean annual SSS distribution from Levitus (left) and associated representativeness error (right)
Finally, today, the Mercator system (SAM2) already assimilates different data types: altimetry, SST and in situ profiles.
Consequently, SSS will enrich this list. This study is particularly important because an optimal assimilation of SSS observations
should theoretically allow us to make a significant step toward more realism of, not only the simulated SSS field, but also all
simulated mixed layer properties and then impacts many other applications, as ecosystems studies.
References
Durand F., G. Gourdeau, T. Delcroix and J. Verron, 2002. Assimilation of sea surface salinity in a tropical OGCM: a twin
experiment approach. J. Geophys. Res., doi:10.1029/2001JC000849, 2002.
Durand, F., L. Gourdeau, T. Delcroix, and J. Verron, Can we improve the representation of modeled ocean mixed layer by
assimilating surface-only satellite-derived data? A case study for the tropical Pacific during the 1997-1998 El Niño, J. Geophys.
Res., 108(C6), 3200, doi:10.1029/2002JC001603, 2003.
Godfrey J.S. and E.J. Lindstrom : The heat budget of the equatorial western Pacific surface mixed layer. J. Geophys. Res., 94,
8007-8017, 1989.
20. Mercator Ocean Quarterly Newsletter #17 – April 2005 – Page 20
An observing system simulation experiment for SMOS: presentation of the study
Paiva A.M. and E.P. Chassignet: The impact of surface flux parameterisations on the modelling of the North Atlantic Ocean. J.
Phys. Oceanogr., 31, 1860-1879, 2001.
Reynolds, R., M. Ji and A. Leetmaa: Use of salinity to improve ocean modeling, Physics and Chemistry of the Earth, 1998.
Pham D.T., J. Verron and M.C. Roubaud: A singular evolutive extended Kalman filter for data assimilation in oceanography, J.
Marine System, 16, 323-340, 1998.
Roemmich D., M. Morris, W. Young and J.R. Donguy: Fresh equatorial jets. J. Phys. Oceanogr., 24, 540-558, 1994.
The ToPAZ group: TOPAZ final report, 2003.
- Notebook -
Editorial board
Nathalie Verbrugge
Secretary
Sophie Baudel
Articles
News : MERSEA : en route to European operational
oceanography
Pierre-Yves Le Traon
The Mean Dynamic Topography used as a reference for
altimetric data assimilation in the Mediterranean Sea
Fabrice Hernandez, Marie-Hélène Rio, Laurence Crosnier
Potential of the ARGO network to produce an oceanic
synthesis of the hydrology and the circulation in the North
Atlantic using a 4D-variational method
Gaël Forget, Bruno Ferron, Herlé Mercier
An observing system simulation experiment for SMOS:
presentation of the study
Florence Birol, Pierre Brasseur, Lionel Renault, Charles-
Emmanuel Testut, Benoît Tranchant
Contact
Please send us your comments to the following e-mail address: webmaster@mercator-ocean.fr
Next issue : July 2005