Mercator Ocean newsletter 13

The Mercator Ocean quarterly letter N°13 – April 2004 – Page 1PSY2G : moving towards global operational oceanography

PSY2G: moving towards global operational oceanography

By Nicolas Ferry

Introduction

In just a few years, Mercator has developed 3 ocean analysis and forecasting systems for the North Atlantic. Today a new stage has been reached with PSY2G, Mercator’s first low resolution (2°), global ocean analysis system.

Each week, the PSY2G system does operational analysis of the global ocean by assimilating all available altimetry observations in an identical way to the work being done with PSY1v1 and PSY2v1. PSY2G thus provides weekly ocean bulletins which are used, in particular, for seasonal forecasting by Météo France. Mercator has thus shown its determination to contribute to GODAE (the Global Ocean Data Assimilation Experiment).

In terms of objectives, PSY2G, due to its horizontal resolution, is of course not designed to assimilate mesoscale processes like PSY1 or PSY2 but rather to assimilate global scale climate signals. This is also a significant technical stage before the migration towards PSY3 (global to ¼°) and the setting up of global mode multivariate assimilation (SAM1v2).

In this newsletter, we shall be describing the PSY2G prototype in detail. The system’s performances are discussed by referring to a global ocean reanalysis for an 11 year period (1993 to 2003).

Description of the PSY2G system

An overall view of the system

The PSY2G prototype, based on the ORCA2 ocean model, assimilates altimetry in a global mode by using the SAM1V1 assimilation technique. This system performs ocean analysis by assimilating along-track, altimetry observations of sea level anomalies (SLA). The PSY2G system is based on 3 components:

The ORCA2 ocean model of LODYC which uses the OPA8.2 code version,

The SAM1V1 assimilation interface, which has already been used in PSY1v1 and PSY2v1, and which is partly based on the SOFA optimal interpellation software (De Mey et Benkiran, 2002), and is inspired by Cooper and Haines (1996) water column lifting and lowering method for correcting the ocean state from a change in sea level,

the PALM software coupler (the PALM team, 2002), which can be used to couple the ORCA2 computer module with that of SAM1V1 without any significant change in the Fortran code.


The ORCA2 ocean model

ORCA2 is the global general ocean circulation model developed by LODYC (Madec et al., 1998, see http://www.lodyc.jussieu.fr/opa/ ). This model is based on the OPA8.2 code which solves the incompressible Navier-Stokes equations with a Boussinesq approximation.

The ORCA2 configuration of OPA used in Mercator is a rigid lid model (using primitive equations). The ocean model includes a closure scheme of about 1.5 for the turbulent kinetic energy to describe the physics of the mixing layer (Blanke and Delecluse, 1993). The Jerlov water type II is used for the whole basin. Laterally, there is 'free slip' while the sea bottom friction is quadratic. The diffusion is harmonic for speed and the trace indices (temperature and salinity). In addition, for the trace indices, diffusion is isopycnal. Density is calculated from the temperature, salinity and depth. The step method is done using Arakawa’s C grid. The realistic topography is based on the ETOPO5, the 5 minutes global atlas. The model’s integration time step is 1 hour 36 minutes, i.e. 15 time steps per day.

Horizontal steps discretisation

The ORCA2 ocean model is characterized by its stretched horizontal grid with its two poles located in the northern hemisphere on the North American and Asian continents (see also http://www.lodyc.jussieu.fr/opa/Docu_Free/ORCA_slide/ORCA_config_PDF/meshmask_grille.pdf).

Figure 1.1 : Horizontal grid of the ORCA2 model. Note the two poles located in the northern

hemisphere as well as the precisin tightening of the grid around the equator.

The grid mesh is smaller in the latitude range from 5°N / 5°S (2° longitude × 0,5° latitude) and

becomes more isotropic (Mercator grid) at medium latitudes (about 2° longitude × 2°cosφ latitude). This grid is also particular in that it gives a more precise resolution in the Mediterranean Sea (1°×1°) and in the Red Sea (≈1°×2°). We can see that the southern part of the Ross Sea (to the south of New-Zealand) is not resolved by the model which considers that this bay is closed south of 78.5° S.

Vertical discretisation

The vertical grid is a ‘Z’ type. The model has 31 levels of which 21 are in the first 1,000 meters

of the ocean. The thickness of the levels varies from 10 meters at the surface (within the first 100 metres ) to 500 meters below the 3,000 meter level. The maximum depth is 5,500 meters .


Figure 1.2 : Vertical discretisation in ORCA2

Surface forcing

The surface forcing used is identical to the other Mercator systems: the model is forced by the

daily surface fluxes (constant for 24 hours) coming from the ECMWF operational analysis. The result of this choice is that the diurnal cycle is not resolved by the model. Reynolds’ analysed SST (Reynolds & Smith, 1994) is used to restore the simulated SST (first layer of the model) towards this field which is assumed to represent the observations with a restoring constant of 40 W.m-2.K-1. The surface salinity is restored towards the monthly climatology of Levitus (1998). The model is moreover forced in an explicit way by the flow rate of rivers based on Baumgartner’s & Reichel’s climatology (1975).

There is no ice model as the ice is ‘diagnosed’ (ice-IF): it is said that there is ice below a fixed freezing temperature (approximately -1.8°C), and when the ice is formed, all of the surface fluxes are set to zero (the ocean is no longer forced). The ice disappears as soon as the surface restoring temperature is greater than the freezing point of sea water.

The initialisation of the model for temperature and salinity is based on the Levitus climatology (1998) with a null initial speed field.

Adaptations of ORCA2 for Mercator requirements

Based on previous work conducted with PSY1v1, deep restoring towards climatology below

1,500 metres was added. The isopycnal lifting and lowering method inspired by Cooper & Haines (1996) is not conservative and creates water masses at the subsurface and especially at the bottom. These water masses are those of the model, but the amounts imposed by the lifting and lowering are likely to be artificial without being tied to the climatology (in temperature and salinity). This restoring starts at 1,500 metres and reaches a maximum below 2,300 meters where it occurs with a 50-day restoring constant. It also decreases closer to the coast or the continental shelves. The idea is thus to not constrain the circulation on the western edge or on the bottom.

Moreover, the global ocean simulation offers the possibility of realistically describing the

variation in mean sea level. Even though the rigid lid model does not conserve the mass, it is possible to calculate the variations of the global ocean mass that can be related to variations in volume


(Greatbatch, 1994) and then to convert this into variations of the mean global level. This variability of sea level on a global scale is slight (an amplitude of a few millimetres per year) and reflects the steric changes due to fluxes of heat and fresh water received by the ocean. The global sea level has been calculated in PSY2G thus distinguishing it from basin models used in Mercator up to now which were based on the Azores region for adjusting mean sea level. PSY2G is thus capable of representing the height of sea level due to climate change, in agreement with altimetry observations.

Global SAM1V1 assimilation

A quick description of the SAM1V1

Very briefly, the SAM1V1 technique, based on the lifting and lowering of water columns (Cooper

et Haines, 1996) consists in calculating an increase in sea level δη which will then be converted into an increase in temperature, salinity and speed.

We start by calculating a change in the sea level based on innovations of sea level along tracks: it is SOFA, the reduced order optimal interpolation tool, which calculates this increment δη.

Based on the statistics for the 3 last months of ocean model simulation, δη is partitioned into a baroclinic contribution (δηbcline) and a barotropic contribution (δηbtrope).

The baroclinic part will be used to build the temperature increment (δT) and the salinity increment (δS) by vertically shifting the isopycnals for each water column of the model so that the dynamic height of the ‘corrected’ water column less that of the previous one is equal to a δηbcline. This movement is done according to a vertical method which indicates for each point along the vertical the vertical amplitude of the shift of isotherms/isohalines (see below). Once this increment of the mass field has been calculated, we can then deduce an increment of the geostrophic speed δUgeo .

Moreover the barotropic contribution, δηbtrope is used to build a barotropic current function increment (δψ) by using the approximation δψ=gH/f×δη, which is valid when the bathymetry is significant (H=ocean depth, f=Coriolis parameter, g=gravity). We then deduce a barotropic speed increment, δUbar .

Finally, a geostrophic adjustment procedure is applied to the increment of the total speed δUgeo+δUbar so that the geostrophic equilibrium is kept within ‘acceptable’ proportions.

The whole of the assimilation scheme is shown below:


The mean reference surface

The assimilation of altimetry data requires knowledge of the mean sea surface height (MSSH) in

order to determine the model equivalent of sea level anomalies observed. The MSSH used in PSY2G is the result of a free mode simulation for the period 1993/1999, corresponding to the period in which the anomalies in sea level are referenced. It shows the usefulness of restoring at depth the model toward the temperature and salinity values of Levitus, thus avoiding too great a drift in the thermohaline content of the ocean water column and consequently of the mean ocean surface. This mean surface is shown below :

Figure 1.3 : Mean surface resulting from free simulation for the period 1993 to 1999 (in m).


Spatial correlation scales

The use of spatial (in longitude and latitude) and temporal correlation scales is important for a

realistic assimilation. We used existing spatial correlation scales calculated from sea level anomaly maps (based on Topex/Poseidon altimetry measurements) calculated every ten days. A minimum of 3 times the size of the model mesh was imposed in order to filter out spatial scales which were too small and incompatible with the model resolution and dynamics. The observations were thus smoothed and decimated along the tracks.

The vertical mode

In the lifting-lowering method for water columns, a vertical shift amplitude for water masses is

applied to a vertical mode. The PSY2G system is designed to describe and simulate large-scale structures of ocean dynamics in which the horizontal gradients remain fairly weak due to the fact that there are practically no mesoscale signals (except in the tropical zone). Consequently, the amplitude of the lifting and lowering of water columns during incremental correction is limited to +15 m.

The vertical mode defines the distribution along the vertical of the lifting and lowering amplitude.

Physically, it is a good idea to represent a vertical structure of the first isopycnal shift mode, thus keeping a strong dependence of the depth of penetration of this mode on the latitude. The vertical mode, which is null in the mixing layer (we did not wish to change this part of the ocean during the assimilation), reached a maximum (equal to 1) under this layer at the level of the main thermocline and penetrated to deeper layers while being attenuated until it became null at a depth zo, which varied as a function of latitude This results very simply in an increase in the vertical extension of the first barocline mode with latitude. Figure 1.4 shows this dependency as a function of latitude with zo which is 300 m at the equator and increases with latitude to reach values of about 3,000 m at 70° of latitude.

Figure 1.4 : Structure of the vertical mode at 30°W on 1st February 1994 at different

latitudes (0°N-blue, 10°N-green, 30°N-red, 50°N-black). Note that the vertical extension differs as a function of latitude


PSY2G reanalysis (1993 to 2003)

PSY2G offers us a light configuration in terms of CPU cost (about 30 minutes of CPU / assimilation cycle on VPP5000). We were thus able to do long assimilation simulations at a reasonable cost, with the advantage of finally being able to evaluate and validate the system over a long period of time.

This is why we set up an 11-year reanalysis (1993 to 2003) with the help of PSY2G, which also served for spin up to real time.

Methodology

The objective was to perform a global reanalysis by simulating all of the available altimetry observations for the period 1993 to 2003. For starting up the model, we were inspired by what had been used for PSY2v1, i.e. a short spin-up before the assimilation in order to keep small the drift of the model water masses.

PSY2G was thus forced (in ‘free’ mode without assimilation) for one year (1992) then from 6 January 1993, the assimilation cycles followed each other every 7 days until the middle of 2003, which was the date to become operational. It should be remembered that the forcing used was based on ECMWF operational analysis.

The altimetry data from the Topex-Poseidon, ERS-2, GFO et JASON-1 satellites was used.

The results of these 10 years of global reanalysis are described below. A first section is devoted

to a statistical analysis of the assimilation and we then look at the thermal structure of the ocean in PSY2G.

Results

Statistical aspects of the data assimilation

After each analysis cycle, a certain number of assimilation diagnostics were done in order to

verify a posteriori the assimilation stage. We shall review some of these diagnostics.

Mean assimilation increment.

At each stage of analysis an increment in sea level, δηa, was applied to the model. With an

unbiased assimilation system, the mean of this increment over the 10 years (1992 to 2002) must be null or very small. Figure 2.1 represents this mean.


Figure 2.1: mean increment in sea level (1993-2002), in cm.

Over a vast part of the ocean, the increment is effectively almost null, a few millimetres less

than the absolute value. Nevertheless, we can see regions in which the system obviously has a bias in sea level while on the whole remaining about one centimetre less: this is the South Atlantic (to the South of 40°S) where the model has a negative bias (the mean increment is positive), at the eastern part of the tropical Pacific (approximately the Nino3 box) with a positive bias, just as for the North Western Atlantic, in particular the subpolar gyre and the Norwegian sea.

We may wonder about these biases, especially as assimilation is done by using a reference mean surface derived from the model. One possible reason for this is that the assimilation method slightly, but significantly modifies the mean state of the model (in particular barotropic circulation) for the 10 years of reanalysis. The assimilation helps to modify circulation and transport of heat/salt and hence the mean surface.

Temporal variability of the increment.

The temporal variability (standard deviation) of the analysis increment is indicative of ocean

regions which are slightly or strongly corrected (figure 2.3). The strongest analysis increments are applied in regions of strong mesoscale variability in sea level: the Western edges of the major ocean gyres as well as the Antarctic Circum-polar Current (ACC) and the Agulhas current.

Figure 2.2 : variability (standard deviation) of the analysis increment for the period 1993-2002 (cm).


Seasonal variability of the correction.

A more detailed analysis shows that the SLA increment has a seasonal cycle. These seasonal

variations are illustrated for two periods of the year, in winter (January to March) and 6 months later, from July to September (figure 2.3). With respect to the mean annual increment (figure 2.1), we can see strong values on the equatorial waveguide and at high latitudes (<40°S and >50°N). But there is a strong contrast between the two seasons with a systematically negative increment in winter (in other words the model’s sea level is too ‘high’) and a rather positive increment in summer, especially in the ACC region.

Figure 2.3 : mean climatological increment for two different seasons :

in winter (JFM) and in summer (JAS).

Let us first look at the Pacific tropical region where, during the winter, the increment is negative.

This results in a ‘warm’ bias in the model surface layers, in other words, the equatorial upwelling is too weak in the model. Previous studies (Hackert et al., 2001) show that the structure and intensity of the surface wind is a key element for correctly representing variability of sea level in the tropical Pacific. In this case, the assimilation thus tries to correct the model bias, which is perhaps due to wind forcing. The negative bias in the tropical Pacific could thus be due to inconsistency between the surface forcing (wind) and assimilated observations. Another possibility would be a bias in the haline content of surface layers which might have a non-negligible impact on the SLA (Maes and al., 2000).

Another remarkable region is the southern seas (<40°S) in which the ACC circulates and where

the seasonal bias is astonishingly spatially homogeneous. During the southern summer (JFM), the


increment is negative and 6 months later, the situation is the reverse, with local biases of 0.5 to 2 cm. In other words, during the summer (southern), the ocean is too ‘warm’ and during the winter, the ocean is too ‘cold’. But if we look more closely at the seasonal cycle of mean sea level in the southern Pacific for the SLAforecast (SLA forecast by the system) and the SLAanalysis (analysed SLA) (figure 2.4), we then see a slight phase offset between the two time series, with the SLAforecast being about 1 month ahead of the SLAanalysis (in other words the SLA observed) and a slight over-estimation of the amplitude of the sea level forecast. This phase offset and difference in amplitude may be due to errors in the heat flux. We nevertheless have to be careful as PSY2G often gives a very poor rendering of ice-ocean interactions which, we know, all play an important role in this area of the ACC.

Figure 2.4 : Mean seasonal cycle of SLAforecast (black curve) and SLAanalysis (red curve) averaged over

the South Pacific (40°-60°S, 210°-70°W) (cm).

In terms of assimilation, the increment bias will be converted mainly into a barotropic

contribution (Hbar) and thus modify the barotropic circulation. The assimilation will thus not try to correct the thermal content of the ocean whereas it is obviously the baroclinic part of the SLA which is responsible for this bias.

Finally, the third region in which the increment bias is strong is the North Atlantic (>50°-60°N), with a systematic overestimation of the SLA by the model in winter (negative increment bias) reaching as much as 5 cm. In this region, the increment is projected almost equally onto the baroclinic and signal barotrope contributions. A more detailed analysis shows that the baroclinic correction (by lifting-lowering) contributes to making the isotherms deeper, between 700 m and 1,500 m, which results in a slow drift (increase) of the dynamic height during reanalysis. This is due in fact to an underestimation by the model of the amplitude of the seasonal cycle of the SLA. The assimilation tries to correct this fault, in particular by means of the lowering-lifting of water columns. In summer, when the ocean is stratified, the correction is efficient and assimilation has the effect of placing the isotherms (in other words by increasing the thermal content of the ocean) and after 2 to 3 years, the model’s SLA in summer is closed to the observed values. On the other hand, in winter, the movement of the isotherms is altogether inefficient, as the ocean is not very stratified and the mixing layer is very deep.

Ocean analysis

Which mean state?


How was the initial state of the ocean modified by SAM1V1 assimilation during the analysis? The simplest diagnosis is to represent the temporal evolution of the mean temperature profile during these 10 years (Figure 2.5a).

Figure 2.5 : Mean profile of ocean temperature (between 60°S and 60°N) as a function of time (°C) for

(a) PSY2G and (b) PSY2G -FREE

To make it easier to interpret these results, we have represented figure 2.6b with exactly the same diagnosis but for the PSY2G-FREE simulation which is the equivalent of the PSY2 reanalysis but without assimilation. For PSY2G, the surface layer (0 to 200 m) is characterized by a slight regular lifting of the isotherms during the 10 years, particularly marked during the first 4 years. The seasonal thermocline (200 to 1 500 m) on the contrary shows a lowering of the isotherms, not very obvious at 200 m, which reaches a maximum around 1,000 m and is again weak beyond 1,600 m. The deep ocean (>1 600 m) whose temperature changes very little or even not at all below 1,800 m, due to the model restoring at depth. On the other hand, in PSY2G-FREE, the mean thermal content of the ocean shows a very slight drift at all depths.

A diagnosis similar to that of figure 2.6 was done for salinity (not shown). PSY2G shows a modification in the distribution of salt content in the water column similar to that observed for temperature with a slight decrease in salinity for the first 200 metres and, on the contrary, an increase in salt content between 200 and 1,500 m in depth. In the end, the impact on the density field is null (no drift) with the changes in salinity being compensated by those in temperature. For PSY2G-LIBRE, there is no modification of salt content and hence no drift in the mass field either. This drift in the characteristics of water masses of PSY2G in the first 1,500 metres is the direct effect of correction by lifting/lowering the columns.

To reveal the geographic distribution of this change of the ocean, figure 2.6 shows the difference PSY2G minus Levitus for the mean annual temperature field at two depths (140 and 500 m). The cooling of the upper layers shown in figure 2.5 is revealed at 140 m by a cold bias in the tropical Pacific (reaching up to -4°C locally). To a lesser extent, we find a similar structure in the Atlantic. Deeper down (at 500 m), we can see the trace of heating noted in figure 2.6 in the three oceans, with a maximum between 10N° to 40°N and 10S° to 40°S, except for the Atlantic where the northern subtropical gyre undergoes cooling.

b) PSY2G-FREE a) PSY2G


Figure 2.6 : Difference in mean temperature between PSY2G (mean over 1993-2002) and the Levitus

atlas (annual mean) at 140 m and 500 m, in °C.

To make it easier to interpret these results, we have represented figure 2.7 with exactly the

same diagnostics but for the difference between PSY2G-FREE and Levitus.


Figure 2.7 : Difference in mean temperature between PSY2G-FREE “libre” (averaged over 1993-2002)

and the Levitus atlas (annual mean) at 140 m and 500 m, in °C.

If we show the results of the two experiments PSY2G and PSY2G-FREE facing each other

(figures 2.6 and 2.7), we can see a few similarities.

At 140 m in depth, the structures in PSY2G are identical to those of PSY2G-FREE but differ in their amplitude, especially between 30°N and 30°S. On the other hand, at higher latitudes (>40°), the two simulations are very close. Note the strong negative bias south of 40°S which can be seen in both simulations and which is likely due the lack of an ice model or to errors in the flux of surface heat. We tried to determine whether the strong negative bias of the Pacific was real or not by comparing the reanalysis with TAO (tropical Atmosphere Ocean) mooring measurements. This strong bias with Levitus was not in fact observed even though the strong El Nino event of 1997/98 contributed a slight negative bias for the period 1993 to 2002. This negative bias is thus an artefact of the assimilation.

Deeper down at 500 m, the two systems behave in a radically different way with significant differences in the latitude range 40°S to 40°N (at greater latitudes the differences with respect to Levitus are similar). While the simulation without assimilation reveals few temperature anomalies at 500 m (figure 2.8), on the other hand PSY2G shows differences with Levitus of about 1 to 2°C. The main part of the signal is shown on the tropical sides of the subtropical gyres for the northern and southern Pacific, and to a lesser extent in the tropical Atlantic and northern Atlantic (South of the Gulf Stream, with a negative anomaly). In keeping with the result suggested by figure 2.5, we can see a global heating at 500 m. A comparison for 2003 of PSY2G and ARMOR analysis (ARMOR is an ocean analysis of temperature and salinity independent of PSY2G see a description in the following section) reveals that this large scale heating in PSY2G was not observed in depth but is more likely an assimilation artefact. A more detailed analysis reveals that the assimilation by lifting lowering tends to erode the main thermocline gradient which ‘relaxes’ and then helps to heat the deep ocean (500 to 1,500 m) whereas the surface layers (at 140 m) cool more (by rising of isotherms).


Inter-annual variability for 2003:

In addition to the climatology aspect that we have examined up to now, it is interesting to compare the inter-annual variability of PSY2G to that of independent ocean observations (for instance T/S profiles) available for the period 1993-2003. Such a comparison raises certain problems to the extent that we have a low resolution global ocean system (of about 2°) whereas the in situ observation such as for T/S profiles reveals both the mesoscale activity of the ocean and the lower frequency climate signals which are difficult to distinguish. However, multi-data combination techniques developed at Mercator (ARMOR, Guinehut et al., 2003) have made it possible to develop global three dimensional weekly ocean analyses (see http://bulletin.mercator-ocean.fr/html/produits/armor/armor_courant_fr.jsp) for 2003. To simplify, the technique used consists in combining various data sets (Levitus atlas, SST, sea level anomalies, T/S profiles) and then to use an EOF 1D combination to construct synthetic profiles for temperature and salinity.

In this section we describe the results of inter-comparison between the weekly PSY2G analysis and those from ARMOR for 2003 (it may be noted that during 2004, a 10 year ARMOR reanalysis will be available and should make it possible to do an inter-comparison for a longer period). More particularly, we analysed the correlations between the weekly temperature inter-annual anomalies in the PSY2G and ARMOR analysis (by subtracting the mean seasonal cycle) and we believe that the ARMOR analyses are realistic.

We have shown significant correlation regions at two depths: on the surface and at 150 m (figure 2.8). More precisely, the coloured areas correspond to regions in which the correlation between PSY2G and ARMOR is significant, and the range of colours indicates the percentage variance in ARMOR anomalies explained by PSY2G (the significant correlation threshold is 0.63 and corresponds to a decorrelation time of about 36 days).

Figure 2.8 : Percentage of variation explained by PSY2G in comparison to ARMOR for surface

temperature anomalies (z=10 m), and at 150. Only significant correlation areas (>0,63) are shown.

The strongest spatial coverage for significant correlation took place on the surface with an

explained percentage of variance greater than 40 % on the whole. We can see a significant lack of symmetry between the two hemispheres with the best scores being obtained between 5°S and 40°N. The temperature anomalies along the Pacific equatorial rail are fairly well simulated as is the subtropical Pacific gyre. In the Atlantic, the latitude range 5°N-20°N shows a fairly good score. At higher latitudes, only the north-eastern part of the Atlantic subtropical gyre is well identified by PSY2G.

At 150 m in depth, we find the large structures which are also present on the surface in particular at the equator in the western tropical Pacific. An interesting aspect is that a significant correlation area is revealed between 5°N and 20°N in the Pacific, at the level of the northern equatorial current. In the North Atlantic, the area with a strong percentage of explained variance on the surface has practically disappeared: only a small area to the East of Africa persists.

This brief comparison (for 2003) between analyses of the PSY2G system and ARMOR show that the model with assimilation is capable of reproducing the surface and subsurface temperature fields limited to latitudes 10°S-40°N. It is worth noting that the correlations are very limited and even practically inexistent in the southern hemisphere. The results also show a certain limit to the SAM1V1 assimilation technique for large scale simulation of the temperature field.


Conclusions and perspectives

We have described the architecture of the Mercator (PSY2G) low resolution, global system as well as the results of the reanalysis for 1993-2003.

After 10 years of assimilation, the system produces a certain number of biases in sea level (of a few millimetres to 1-2 cm in certain areas) which reflect the change in the state of the ocean following assimilation of anomalies in altimeter sea level measurements. These biases generally have a seasonal cycle. The forcing uncertainty also introduces a lack of precision into the system and sometimes conflicts between the assimilated observations and the forcing. Sensitivity studies conducted with other foncings would probably improve some aspects of PSY2G.

It also appears that the assimilation method consisting of projecting the information onto the vertical and correcting by lifting-lowering isopycnals is fairly simplistic and limited. The ocean analysis thus reveals an erosion of the main thermocline which heats (cools) the deep ocean (the surface layers) in an unrealistic way. Moreover, even though the first mode of isopycnal vertical shift is related to a large part of the ocean variability, assimilation of altimetry measurements alone appears to be insufficient for adequately constraining the ocean: assimilation of other observations (temperature and salinity profiles, SST, etc) appears to be necessary.

Finally, the short analysis of inter-annual variability for 2003 using ARMOR reveals that this first version of the global ocean analysis system is able to a certain extent to reproduce the temperature signals observed on the surface and at the subsurface. Hopefully, the introduction of multi-data multivariate assimilation techniques such as those of SAM1V2 (see the other article in this newsletter) will offer interesting possibilities for future evolution of PSY2G. This work has already been planned and should improve the performances of the Mercator global analysis/forecasting systems.

References

Baumgartner A., and E. Reichel, 1975: The World Water Balance., 179pp, Elsevier.

Blanke, B., and P. Delecluse, 1993: Variability of the tropical Atlantic ocean simulated by a general circulation model with two different mixed layer physics. J. Phys. Oceanogr., 23, 1363-1388.

Cooper, M, and K. Haines : 1996: Data assimilation with water property conservation. J. Geophys. Res., 101, 1059-1077.

De Mey, P. & M. Benkiran, 2002: A multivariate reduced-order optimal interpolation method and its application to the Mediterranean basin-scale circulation. In : Ocean Forecasting : Conceptual basis and applications, N. Pinardi and J.D. Woods, Eds, Springer Verlag, Berlin, Heidelberg, New York, 472pp.

Guinehut, S., Le Traon, P.Y., Larnicol, G., Philipps, S., 2003: Combining Argo and remote-sensing data to estimate the ocean three-dimensional temperature fields - A first approach based on simulated observations. Journal of Marine Systems, in press.

Greatbatch, R. J., 1994: A note on the representation of steric sea level in models that conserve volume rather than mass. J. Geophys. Res., 99, C6, 12767-12771.


Hackert, E.C., A.J. Busalacchi, and R. Murtugudde, 2001: A wind comparison study using an ocean general circulation model for the 1997-98 El Niño. J. Geophys. Res., 106, 2345-2362.

Levitus, S., T.P. Boyer, M.E. Conkright, T. O'Brien, J. Antonov, C. Stephens, L. Stathoplos, D. Johnson, R. Gelfeld (1998) NOAA Atlas NESDIS 18, WORLD OCEAN DATABASE 1998: Vol. 1: Introduction, U.S. Gov. Printing Office, Wash., D.C., 346 pp.

Madec G., Delecluse P., Imbard M., Lévy C., 1998 : OPA 8.1 ocean general circulation model reference manual, Notes du pôle de modélisation IPSL, 91 pp

Maes, C., D. Behringer, R. W. Reynolds and M. Ji., 2000: Retrospective analysis of the salinity variability in the western tropical Pacific Ocean using an indirect minimization approach. J. Atmos. Ocean Tech., 17, 512-524

the PALM team, 2002 : PALM and PrePALM Proto Quick Start Guide for version 2.0.0 and following. Cerfacs technical report TR/CMGC/01/xx.

Reynolds, R. W. and T. M. Smith, 1994 : Improved global sea surface temperature analyses using optimum interpolation. J. Climate, 7, 929-948.

Environment

Mercator Ocean newsletter 13