Mercator Océan (2012)

Benefit of Argo data on Mercator Océan Forecasting Systems       

N. Ferry, M. Benkiran, M. Drévillon, E. Greiner, J.-M.Lellouche, L.Parent, C.-E..Testut,, B. Tranchant and The Mercator Océan Team (1)

Corresponding author : Nicolas Ferry

(1) : Mercator Océan, Ramonville Saint-Agne, France

"In Mercator Ocean systems, Argo network has contributed to decrease by 70%
the salinity 7-day error forecast in the top 100m of the ocean."

Figure 1 : Salinity innovation (observation minus model forecast) for the global ocean in GLORYS2V1 reanalysis (1992-2009) in different layer (0-100m, 100-300m, 300-800m, 800-2000m). Argo network begins to be implemented in 2000.

Created in 2002 as a Public Interest Group ("GIP"), Mercator Ocean is a non-profit company since 2010, owned by 5 major French agencies - all founding members -: CNRS (National Centre of the Scientific Research), IFREMER (French Research Institute for Sea operations), IRD (Development Research Institute), METEO FRANCE (French National Weather Service) and SHOM (Oceanographic Department of the French Navy)

Mercator Ocean first mission is the development and operation of high resolution modelling/assimilation systems covering the global ocean (with an increased resolution in the European seas) to provide ocean hindcasts, nowcasts and forecasts to users ranging from scientists to seafarers, public authorities to explorers, industries to individuals, universities to the organisers of regattas and software publishers to fish farmers.

Mercator Ocean has been continuously running on a real time basis ocean forecasting systems on the North Atlantic (since the beginning), the Mediterranean Sea (since 2003) and eventually the whole global ocean (since 2005). Mercator Ocean forecasting system delivers weekly ocean analyses and daily forecasts at global (1/4° resolution) and regional (1/12° over the North Atlantic and Mediterranean Sea and 1/36° in the North-Eastern Atlantic) scale constrained by altimetry, SST and in situ Temperature and Salinity profiles (Dombrowsky et al., 2009). Transition to 1/12° for the global ocean system was achieved in January 2011 and provides eddy resolving ocean analyses and forecasts.

Mercator Ocean is also the coordinator of MyOcean project (and its follow-on MyOcean2) which whose objective is to define and to set up a concerted and integrated pan-European capacity for ocean monitoring and forecasting. The activities benefit several specified areas of use: Maritime security, oil spill prevention, marine resources management, climate change, seasonal forecasting, coastal activities, ice sheet surveys, water quality and pollution. MyOcean EU funded project is the pre-operational marine service of the Global Monitoring for Environment and Security programme (

Ocean operational forecasting is possible only because real time observations are available and can be assimilated by ocean forecasting systems. Sea surface temperature and sea level observations are measured by satellite and distributed routinely in real time. For in situ observations, it is only since the set up of Argo network (beginning of the years 2000) that it is possible to have real time in situ measurements of temperature and salinity. Argo is the single most important in-situ observing system for global operational oceanography. It delivers critical data for assimilation in ocean forecasting models, and is a key component of the GMES programme.

Data and Method

The operational forecasting systems developed by Mercator Ocean are based on primitive equations ocean models which assimilate in real time ocean measurements from satellite (SST, sea level anomaly) and in situ networks like Argo.

Mercator Océan relies on existing data assembly centers to collect, process and validate its input real time and delayed mode data. Input data are used either for data assimilation and for analysis / forecast validation and verification (CAL/VAL) and include :

  •  Altimetry (SSALTO /DUACS product from AVISO data centre) : Along-track and inter-calibrated sea level anomalies from Jason-CS & Jason-2, Cryosat2; Mean Dynamic Topography (also called Mean-Sea-Surface-Height) combining gravity (CHAMP, GRACE) measurements, and altimeter and in situ data [Rio et al., 2011, CNES-CLS09].
  •  In-situ temperature and salinity data (from CORIOLIS centre): ARGO profiling floats, XBT, CTD, TAO/PIRATA/TRITON, surface drifters; weekly retrieval ; off line quality control at CORIOLIS and CLS
  • Sea Surface Temperature : Global NCEP/RTG 0,5° SST product for operational assimilation ; Reynolds 0.25° AVHRR-only SST product for reanalyses ; Eumetsat / Météo-France SAF Ocean&Ice Atlantic high resolution SST product (10 km, daily) for routine validation;
  • Forcing data : 3-hour analyses and predictions from ECMWF for operational forcing / ERA-Interim for reanalysis forcing ;
  • Surface velocity, from the Global Drifter Program, collected and processed by the CORIOLIS centre.
  • Sea level from tide gauges: GLOSS, SHOM and ESEOO database, processed and corrected at CLS.
  • Sea Ice concentration and drift from CERSAT (Ifremer) and OSI SAF.

Mercator Océan uses the OPA/NEMO z-coordinate primitive equation ocean code [Madec, 2008]. It is coupled with the LIM2_EVP Ice Model (Fichefet and Maqueda, 1997) from Louvain La Neuve with elastic-viscous-plastic rheology. The model system can be coupled with biogeochemical models (PISCES, LOBSTER). All systems include a data assimilation scheme which is a reduced order Kalman filter based on the SEEK formulation [Pham, et al., 1998].

Mercator Océan is operating the following systems:

  • Operational Ocean Forecasting Systems (OOFS), operated on a weekly basis, to provide hindcast, real-time analysis and 2-week forecast of the physical state of the ocean
    • PSY4V1: 1/12° resolution / 50 levels global OOFS, with multivariate assimilation of altimetry, SST and T&S in situ profiles. Weekly analysis and 7-day forecast.
    • PSY2V4: 1/12° resolution / 50 levels North Atlantic and Mediterranean Sea OOFS, with multivariate assimilation of altimetry, SST and T&S in situ profiles. Weekly analysis and 14-day forecasts, updated daily.
    • PSY3V3: ¼° resolution / 50 levels global OOFS, with assimilation of altimetry, SST and T&S in situ profiles. Weekly analysis and 14-day forecasts, updated daily.
    • IBI36: 1/36° resolution / 50 levels regional OOFS, with initial condition inherited from North Atlantic and Mediterranean Sea OOFS. Weekly initialization with PSY2V4 and 5-day forecasts, updated daily.
    • BIOMER1: 1° resolution / 50 levels, biogeochemical global OOFS, forced by PSY3V3 ocean physics. Analyses are produced with a 2-week delay with respect to the real time.
    • PSY2G2: 2° resolution (with 0,5° refinement near Equator) / 42 levels global OOFS, with assimilation of altimetry, SST and T&S in situ profiles. Weekly analysis.
  • Ocean Reanalysis Systems (ORS), to provide the best estimation and interannual variability of the physical state of the ocean
    • GLORYS2V1: ¼° horizontal resolution / 75 levels global ocean reanalysis with data assimilation of altimetry, SST and T&S in situ profiles (“altimetric” period - 1993-2009).

The production of OOFS, ORS is monitored and the product quality is assessed in routine. The Mercator Océan system provides a full 3D depiction of the ocean dynamics and thermohaline circulation (T, S, currents, sea surface height, mixed layer depth …), with a priority given to high resolution (eddy resolving) scales. Information is available on a near-real-time and routine basis, by providing weekly Near-Real-Time Analysis and 2-week Forecasts; and on a Reanalysis mode (2 weeks behind real time), with data assimilation. PSY2V4 and PSY3V3 systems are also producing 7-day forecasts updated every day (i.e. using updated analysed and forecast surface atmospheric fields). IBI36 system performs 5-day forecasts updated on a daily basis.





NEMO ocean model coupled to LIM2_EVP sea ice model [Madec, G., 2008]


  1. PSY4V1 : Global (77°S - 90°N)
  2. PSY2V4 : North Atlantic + Mediterranean Sea (20°S-80°N)
  3. PSY3V3 : Global 577°S-90°N)
  4. IBI36 : Iberian Biscay Irish Seas 1/36° (19°W-5°E,26°N-56°N)
  5. BIOMERV1 (77°S-90°N)
  6. PSY2G2 (77°S-90°N)
  7. GLORYS2V1 : Global ocean reanalysis (77°S-90°N)

Horizontal Resolution

  1. PSY4V1 : 1/12°
  2. PSY2V4 : 1/12°
  3. PSY3V3 : 1/4°
  4. IBI36 : 1/36°
  5. BIOMERV1 : 1°
  6. PSY2G2 : 2° (0.5° near Equator)
  7. GLORYS2V1 : 1/4°

Vertical Sampling

  1. PSY4V1 : 50 Z levels, 1m near surface
  2. PSY2V3 : 50 Z levels, 1m near surface
  3. PSY3V3 : 50 Z levels, 1m near surface
  4. IBI36 : 50 levels, 1m near surface
  5. BIOMERV1 : 50 Z levels, 1m near surface
  6. PSY2G2 : 31 Z levels, 10m near surface
  7. GLORYS2V1 : 75 Z levels, 1m near surface

Atmospheric Forcing

ECMWF analyses, 3H forcing, bulk formulation (CORE)


Assimilation Scheme

"Mercator data Assimilation System" version 2 (SAM-2) reduced order

Kalman filter based on the SEEK formulation with 3D-VAR slowly evolving

large-scale bias correction for temperature and salinity


Reynolds 0.25° SST (reanalysis)



Mean Dynamic Topography : CNES-CLS09 [Rio et al., 2011]]


In situ temperature and salinity provided by CORIOLIS GDAC


Forecast range

Forecasts up to 14 days, except for IBI36 where forecast are up to 5 days

and PSY4 where forecasts are up to 7 days

Update Frequency

weekly analyses

daily updates of the forecasts for PSY3V3 (7-day forecasts),

PSY2V4 (7-day forecasts), IBI36 (5-days forecasts)

Hindcast length

14-day hindcast


General Information

Technical Description

Viewing Service

Table 1: Mercator Ocean System information overview


In this section we show the different application of Argo observations in Mercator operational ocean forecasting centre. Temperature and salinity observations provided by Argo floats are presently used for several purposes :

  1. Input for data assimilation the operational systems: the Argo observation help to constrain the model state in order to build the most probable ocean state at a particular time, and hence to perform the best possible forecast.
  2. Input data for the qualification and validation of Mercator ocean analyses and forecasts.
  3. A by-product of (1) is that Mercator forecasts help to quality control the Argo observations. As a consequence, grey list of suspicious observations are established and sent to in situ data centres in order them to improve delayed time observation data bases.

We illustrate these three points using some examples.

2.1) - Data assimilation of Argo data
  •   Impact of Argo Data in operational systems

Three experiments have been performed during the year 2004 using the operational system over the North Atlantic.The REFERENCE run assimilates all observations (SLA, SST,T/S), the NO_TS run assimilates all observations with the exception of the in-situ T/S profiles and the NO_SLA run assimilates all observations with the altimeter SLA. Assimilation diagnostics for the temperature and salinity fields performed between the assimilated data sets and the model 7-days forecast (figure 2) show the instantaneous development of large drift when T/S profiles are not assimilated. Biases are of the order of 0.3°C for the T field in the 300-1000 m layer and of the order of à 0.05 psu for the S field in the 300-600 m layer. At greater depths (between 1000 and 1500m), the model also drifts from the climatology and over time, large biases also develops. Concerning SLA, a large drift of the model appears when SLA is not assimilated, with a bias of more than 4 cm in May. Those three experiments emphasize the complementary roles of in-situ T and S profiles and satellite SLA since T/S observations control the drifts of the model and reduce the biases in the analyses, and SLA observations control the structures and the energy of the model.

Figure 2 : RMS of the differences between the in-situ profiles and the model 7-days forecast as a function of depth and time and for the three experiments. (top) : for the temperature field (°C) and (bottom) for the salinity field (psu).

  •   Impact of Argo Data in a global ocean reanalysis

Another way to see the impact of Argo data on the analysis and forecast skill is shown in Figure 1 (see top of the document). In GLORYS2V1 global ocean reanalysis (Ferry et al., 2012) assimilating quality controlled SST, SLA and in situ observations, very few in situ temperature and salinity observations are available before the 2000s decade. From 2000 and onwards, Argo network develops and provides more and more in situ observations to constrain the ocean interior. The consequence of this in situ observation network change in terms of forecast skill is revealed by Figure 1. Before 2000, the RMS of the innovation (observation minus model forecast) in different layers of the ocean is much larger than afterwards. In the 100-300m layer, the error drops from ~0.2 to 0.12 PSU when Argo network arrives, which represents an error decrease of ~65%.

2.2) - Use of Argo data for quality assessment of operational analyses and forecasts

The quality assessment of operational ocean analysis and forecasting systems is a central task of ocean forecasting centres. Mercator Océan has a long experience in ocean forecast validation as it is routinely done every week since 2001. In 2011, Mercator Océan began to produce his Quarterly Ocean Validation Display (QuO Va Dis?) published four times a year ( This bulletin gives an estimate of the accuracy of Mercator Océan’s analyses and forecast for the passed season. It also provides a summary of useful information on the context of the production for this period. Diagnostics are displayed for the global 1/12° (PSY4), global ¼° (PSY3) and the Atlantic and Mediterranean zoom at 1/12° (PSY2) monitoring and forecasting systems currently producing daily 3D temperature salinity and current products. We present hereafter some of the diagnostics done in order to evaluate the quality of the operational systems.

Figure 3. (a) RMS difference of the observation minus findcast for PSY3V3R1 (global ¼°) temperature in the 0-500m layer in JFM 2012. The squares colour indicates the RMS differences and their size refers to the number of observations used. (b) CLASS4 skill score w.r.t climatology in the 0-500m layer for PSY3V3R1 3-day temperature forecast in JFM 2012. A positive skill score indicates hat the model forecast is closer to the observations than climatology. (c) RMS difference of the observation minus PSY3V3R1 (global ¼°) temperature in the 0-500m layer as a function of time (from January 2011 to June 2012). RMS difference with observations is computed with PSY3V3R1 hindcast (black), nowcast (blue), persistence (cyan), 3-day forecast (green), 6-day forecast (red). RMS difference is also computed w.r.t ARIVO (orange) and WOA 2005 (brown) climatology.

Figure 3 presents three metrics used to monitor the quality of PSY3 temperature analyses and forecasts. All these metrics use Argo in situ observations (temperature, salinity – not shown) as verification data.

Figure 3a maps the RMS difference of PSY3V3R1 temperature delayed time analysis (i.e. 7 to 14 days behind real time) with Argo observations in the layer 0-500m during JFM 2012. The data density is summarized by the square size and the RMS is given by the colour scale. On average, the RMS is of the order of 0.5°C except in the regions of strong meso scale signals like the Gulf Stream and Kuroshio regions, the Aghulas current area and the Antarctic Circumpolar Current where the misfit is larger (1~2°C).

Figure 3b shows the ability of the model 3-day forecast to have a better skill than climatology. This skill score is computed as one minus the ratio of the mean square difference between the model 3-day forecast minus the observation to the mean square difference between the climatology minus the observation. A positive skill score indicates that the model forecast performs better than the climatology. In most areas, the skill score is positive indicating that the model 3-day forecast is more realistic than the climatology. However, some blue pixels are visible on the map. These locally low forecast skill suggest some forecast errors related to mislocated eddies. Part of these blue pixels are also due gross errors in the observations which lead to low forecast skill. In that specific case, the model forecast is right, but the observation contains some errors. This point is illustrated in section 3.

The bottom panel of Fig. 3(c) allows checking the quality of the analyses and forecasts as a function of time in the upper part of the ocean (0-500m). The hindcast, which is a delayed time analysis performed between days 14 to 7 14 behind real time exhibits the best skill. The real time analysis is close to the hindcast but has a slightly larger RMS. The 3 and 6-day forecasts are stable during the time, around 0.9°C RMS and outperform the persistence. When compared to the analyses, one can see that the available climatologies are about 0.5°C RMS larger than the analysis.

Quality control of Argo observations

An important issue for the users of in situ observations provided by Argo is the quality control of these data. Mercator Océan has developed a background quality control for in situ observations (Lellouche et al., 2012, Cabanes et al., 2012) based on model innovation statistics.

A basic hypothesis of data assimilation systems is that observation minus model forecast (i.e. innovation) are normally distributed in each point of the ocean (Gaussian distribution of background errors). Observations for which the innovation is in the tail of the distribution will thus be considered as questionable. Taking advantage of the very large number of temperature and salinity innovations collected in Mercator Océan GLORYS2V1 reanalysis (1993-2009, Ferry et al., 2012), it has been possible to reliably estimate their seasonally and spatially variable statistics (mean, standard deviation). These parameters have then been used to define a space and season dependent threshold value for temperature and salinity innovations (Cabanes et al., 2012). So it is possible to quality control Argo data using the model forecast. This has been done with GLORYS2V1 reanalysis and is currently being implemented in real time operational systems.

Figure 4 illustrates the kind of results that can be obtained with the background quality control. Figure 4a displays the spatial distribution of suspicious temperature and salinity profiles in 2009. It is expected the profile to be randomly distributed in space (a defect on an Argo float is a priori uncorrelated from its geographical position), which is almost the case. We can note at some place accumulation points, in the central Tropical Pacific or East to the Philippines. This corresponds to a moving Argo float with defective sensors being following the mean current. Figure 4c and 4d represent questionable temperature and salinity profile identified with the automated quality control. Both profiles present a large misfit with the model background especially at depth. For the salinity profile, the misfit suggests a sensor bias problem, probably due to a drift of the salinity sensor.

Thanks to this routine monitoring of ocean analysis and forecasts against Argo observations, it is possible to deliver to the user calibrated and validated products. Grey lists of questionable questionable temperature and salinity profiles are also reported to CORIOLIS data centre to improve its CORA delayed time in situ data base (Cabanes et al., 2012). The quality control performed in operational ocean forecasting centres such as Mercator Océan helps to improve the quality of delayed time Argo data.

 Figure 4 : (a) spatial distribution of questionable temperature profiles identified with the background quality control. Example of questionable in situ (b)-temperature and (c)-salinity profile detected with the background quality control. For (b) and (c): let panel: innovation (blue line) and threshold envelope (red). Right panel: Climatological profile (green), model background (red) and Argo float measurements (black). The bold circles indicates the questionable observations.


Argo observations are a key component of the present ocean observing system. It is the single and a unique way to observe in near real time the ocean interior. It provides to operational ocean forecasting centres such as Mercator Océan highly valuable information about the ocean state.

Argo data is used in different ways at Mecator Océan. First, Argo observations are used as input for data assimilation in the operational analysis and forecasting systems as well as in ocean reanalyse. Secondly, Argo observations are used for the analysis / forecast validation. This is an essential task of ocean forecasting centres. A synthesis of the validation routinely performed at Mercator Océan can be found in the quarterly QuoVaDis validation bulletin ( Lastly, Mercator Océan is able to perform quality control on Argo data and to provide grey list of questionable in situ profiles to data centres.

The information provided by Argo observations finally benefits to Mercator Océan products which are available to the users through MyOcean web site, the implementation of GMES marine service.

  • Cabanes C., A. Grouazel, K. von Schuckmann, M. Hamon, V. Turpin, C. Coatanoan, S. Guinehut, N. Ferry, G. Reverdin, S. Pouliquen, P.-Y. Le Traon : The CORA database: Validation and Diagnostics of global ocean in situ measurements, Ocean Science Discussion, in revision, 2012
  • Dombrowsky, E., L. Bertino, G. B. Brassington, E. P. Chassignet, F. Davidson, H. E. Hurlburt, M. Kamachi, T. Lee, M. J. Martin, S. M. : GODAE Systems in Operation. Oceanography, 22, 80-95, 2009
  • N. Ferry, L. Parent, G. Garric, C. Bricaud, C-E Testut, O. Le Galloudec, J-M Lellouche, M. Drevillon, E. Greiner, B. Barnier, J-M Molines, N. C. Jourdain, S. Guinehut, C. Cabanes, L. Zawadzki : GLORYS2V1 global ocean reanalysis of the altimetric era (1992-2009) at meso scale. Mercator Quarterly Newsletter 44, January 2012, 29-39. available at, 2012
  • Larnicol, G., S. Guinehut, M.-H. Rio, M. Drevillon, Y. Faugere and G. Nicolas : The Global Observed Ocean Products of the French Mercator project, Proceedings of 15 Years of progress in Radar Altimetry Symposium, ESA Special Publication, SP-614, 2006.
  • Lellouche, J.-M., M. Drévillon and S. Baudel : How ocean forecasting benefits from d’Argo, Argonautics number 7, June 2006.
  • Lellouche J.-M., 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, C. De Nicola : Evaluation of global monitoring and forecasting systems at Mercator Océan. Ocean Science Discussion, in revision, 2012
  • Levitus, S., T. Boyer, M. Conkright, T. O'Brian, J. Antonov, C. Stephens, S. L, J. D, and R. Gelfeld : World Ocean Data Base. NOAA Atlas, NESDID18, 1998.
  • Oke Peter R., Balmaseda Magdalena A., Benkiran Mounir, Cummings James A., Dombrowsky Eric, Fujii Yosuke, Guinehut Stéphanie, Larnicol Gilles, Le Traon Pierre-Yves, Martin Matthew J. : Observing system evaluations using GODAE systems. Oceanography, 22(3), 144-153, 2009
  • Rio M.-H., S. Guinehut, G. Larnicol : The New CNES-CLS09 global Mean Dynamic Topography computed from the combination of GRACE data, altimetry and in-situ measurements, 2011, accepted for publication in J. Geophys. Res.