; WP7_Activity_Report_m12
Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out
Your Federal Quarterly Tax Payments are due April 15th Get Help Now >>

WP7_Activity_Report_m12

VIEWS: 4 PAGES: 27

  • pg 1
									                               Activity report
                     WP 7 Modelling and data assimilation
Section 1 – Work Package objectives and major achievements during the reporting period

The general objectives of WP7 are (i) to perform research and development on physical modelling,
ecosystem modelling and data assimilation as required for the operational objectives of MERSEA during
the whole life of the project, (ii) to provide all the necessary tools (physical model codes, biogeochemical
model codes, data assimilation codes) that are required by the MERSEA project to reach its operational
objectives, and (iii) to bring ad hoc scientific innovations into the project and to gather all the relevant
European capabilities in order to make sure that operational systems are maintained at the most advanced
level thus benefit from the progress achieved in the research community, during the project life-time and
beyond.

This work package is divided into three main components, dedicated to physical models, biogeochemical
models and data assimilation methods :

The main activities of the Physical models task (Task 7.1) are:
 Transfer the intermediate high resolution (1/4°) physical model code and insure relevant expertise on
   running the code, analysing and assessing the results at global scale.
 Realize R&D activities for continuous model improvements (numerical schemes, bottom boundary
   layer, tidal mixing, mixed layer,...).
 Contribute to the development of a global ORCA high resolution (1/12°) physical model.
 Develop and transfer the nesting capabilities (software and expertise) to make the one-way and two-
   way connections between the large scale model and the regional models.

This first year of MERSEA activity is characterized by noticeables advances (partial steps, numerical
scheme, …) in the development of the 1/4° resolution global model and the preparation of the first
transfer for implementation.

The main activities of the Biogeochemical models task (Task 7.2) are:
 Develop a biogeochemical model to be coupled to the MERSEA global physical model: nitrogen-
   based model with implicit representation of size classes in phytoplankton and sinking particles,
   mechanistic model of DOM production as a function of plankton size, mechanistic model of DON
   and DOC consumption with a minimum number of variables, inclusion of different limiting nutrients
   and carbon.
 To develop the prototype of a coupled physical/biological assimilative system covering the North
   Atlantic, which will be used to demonstrate the capacity to routinely estimate and forecast
   biogeochemical variables.

This first year of MERSEA activity is characterized by considerable progress in the development of a
new, implicitly size-structured ecosystem model, including first calibration studies at various time-series
sites. In addition, a number of sensitivity studies have been performed with presently available ecosystem
models coupled to OPA. These will form a reference basis for future assessments of improved ecosystem
model components. During the first year of MERSEA, a beta-version of the platform has been developed
using a rigid-lid OPA model of the North Atlantic at eddy-permitting resolution (MNATL), a sequential
assimilation scheme (the SEEK filter) and a first-generation biogeochemical model (N1P1-type).

The main activities of the Data Assimilation task (Task 7.3) are:
 Explore ensemble-based assimilation schemes (EnKF, SEEK, SIR) which will more rigorously
   tackle the statistical nature of the ocean forecasting problem (non gaussian error statistics, biased
   estimations, non-linear processes)
                                                     1
   Improve the capacity of the integrated system to assimilate data from different sources (satellite
    altimetry, SST, SSS, in situ profiles) in the free-surface OPA circulation model

This first year of MERSEA activity is characterized by significant advances with regard to the
implementation of the SEEK filter into the free-surface OPA model and to important steps forward for the
use of the SIRF into a biogeochemical model.




Section 2 – Workpackage progress of the period

Task 7.1. Physical Modelling


Task objectives
The main objective of the physical modelling subtask is to upgrade the numerical ocean general circulation
model (GCM) NEMO1 which is one of the major components of the operational system developed in
MERSEA. The starting point is the OPA8.1 ocean GCM developed at LODyC (Madec et al., 1998). The
ultimate goal is to deliver to the project a global model at eddy-permitting resolution (1/4°), based on the
NEMO code, an up-graded version of this model including a fully prognostic sea-ice model (the LIM model
from the University of Louvain la Neuve, Fichefet et al., 1997), and recent parameterisation of topography
(partial steps), an improved parameterisation of the mixing at the surface and in the interior, and most advanced
numerical scheme. The model should also include grid-refinement capabilities, in relation with the need of
WP10. A first version of the model, referred to as the DRAKKAR2 configuration ORCA-R025, will be
delivered to the project at the end of the first 12 month period. A final version, validated by sensitivity
experiments will be delivered at month 30.
1
NEMO: Nucleus for a European Model of the Ocean.
2
Because inherited from the DRAKKAR project.


Scientific / Technical Progress made in this task


Task 7.1.1 OPA HR model developments
The task in on schedule. Progress made in the various subtasks are summarised below.
Subtask 7.1.1.1. Partial steps and global 1/4° model configuration
This subtask concerns principally the validation of the new numerical treatment of the bottom topography, and
the implementation of the global 1/4° model configuration.
   Representation of the bathymetry by the introduction of partial steps.
In a regular z-coordinate model (also referred to as a level model), the Full-Step (FS) representation of the
bottom topography uses a vertical grid spacing made of pre-defined regular levels. For example, in a 46 level
model as the one used by MERCATOR, levels below the main thermocline have a constant size of 250 m (Fig.
7.1.1.1a). Therefore, the bathymetry is represented by steps of 250 m height, and the total depth at a given point
in the model, which is adjusted to the depth of the closest grid cell, is only an approximation of the observed
depth.
In the Partial Step (PS) representation of the bottom topography, the size of the grid cell closest to the bottom is
adjusted to the observed depth of the ocean (Fig. 7.1.1.1b). Therefore, the size of the bottom grid cell is highly
variable (from 25 m to 250 m in the model developed for MERSEA), which make the calculation of the
pressure gradient more difficult. The great interest is a better representation of the real f/h contours, and a better
communication between bottom cells, as shown in Fig. 7.1.1.1b.

                                                          2
Both FS and PS representation of bottom topography have been made available in NEMO, and implemented in
the global 1/4° model.




                         Full Step                                                                      Partial Step
      Fig. 7.1.1.1. a) Full Step representation of bottom topography: the ocean total depth is adjusted to the depth of the closest grill
                  cell. b) Partial Steps: the depth of the grill cell closest to the bottom is adjusted to the real ocean depth.




   Comparison of the impact of the partial step representation of topography in OPA8 with a model of
    the South Atlantic at eddy permitting resolution (1/3°).

The Argentinian Basin is a region where interactions between bottom topography and mesoscale
turbulence give rise to a very particular large scale circulation pattern, the Zapiola permanent anticyclone,
a now well observed feature that only the SPEM model (a sigma coordinate model) was able to reproduce
(De Miranda et al., 1999). In particular, regular FS z-coordinate models always failed to simulate the
Zapiola Anticyclone. This area is thus very suitable to test new representation of the bottom topography at
eddy permitting resolution in numerical models.
    We carried out at LEGI a series of model simulations with the Clipper 1/3° south Atlantic Model
SATL3 to test the impact of the partial steps available in the modified version of OPA8. We remind here
that this model uses the z-coordinate (42 vertical levels), and has three open boundaries at Drake passage
(68°W), between South Africa and Antarctica (30°E) and at 16°S between Brazil and Angola. It is driven
by daily climatology of air-sea fluxes derived from ECMWF re-analysis for the Clipper project. Two
experiments have been carried out with SATL3, which differ only by the representation of the bottom
topography:
    - Experiment S-AS1: the bottom topography is represented by the usual full cells (Fig. 7.1.1.1a).
    - Experiment S-AS2: the bottom topography is represented by the new partial cells (Fig. 7.1.1.1b).
Both model configurations were integrated over a 10 year period. A detailed analysis and comparison of
these two simulations is presented in a report from LEGI which evaluate the impact of partial steps to this
configuration. Results show an very significant improvement of the simulated circulation in case of
partial steps, in terms of dynamics and thermodynamics. In particular, compared to the full step
simulation, the partial step solution shows:
 A more realistic circulation in the Confluence region, with a Malvinas Current reaching 40°S, and a
    good representation of the separation of the Brazil Current and a realistic split of the subtropical and
    subantarctic fronts.
 A signature of the Zapiola Anticyclone over the whole water column, with a simulated transport of
    the order of 80 to 100 Sv close to the value of 100 Sv deduced from observations. To our knowledge,
    this strong circulation feature was never reproduced by any full-step z-coordinate model.
 A "C-shape" distribution of the eddy kinetic energy (eke) in the western south Atlantic (Fig. 7.1.1.2)
    in better agreement with satellite estimates deduced from T/P and ERS.
 An improved distribution of water masses which now keeps to the large scale water mass distribution
    of revealed by the WOCE sections.
                                                       3
                     Full Step topography                                                   Partial Step topography

Fig. 7.1.1.2: Eddy Kinetic Energy (eke) in the Argentinian Basin. Yellow indicates values greater than 500 m 2s2. The partial steps solution shows
a C-shape distribution which agrees well with satellite observations of T/P and ERS. This improvement is due to a remarkable improvement of
the mean currents and of the effect of topography on mesoscale turbulence.


Our conclusions are thus very favourable to the use of Partial Steps in future applications of NEMO (the
new version of OPA code) in MERSEA. However, the validation and understanding of the Partial Cells
presented here is only preliminary and additional investigations of the impact of this representation of the
bottom topography in eddy resolving models, and in the global 1/4° model in particular, is still necessary.

    A North Atlantic 1/4° prototype
A first prototype of a North Atlantic regional model at an eddy permitting resolution of 1/4° resolution has been
implemented with NEMO (referred to NATL4) to be used to assess the impact of the partial steps representation
of topography in the North Atlantic. NATL4 is in fact a sub-domain of the global configuration ORCA -R025
(Fig. 7.1.1.3), using the same grid, bathymetry and forcing. NEMO includes the new version of OPA9 as ocean
model coupled with the LIM sea-ice model. The development required a great number of tests. We used this
configuration to test the representation of topography by using the Partial Steps against the usual but less
accurate Full Steps. Experiments confirmed the incompatibility between the vorticity advection scheme and the
partial steps topography, with the development of spurious barotropic circulation cells. A new vorticity
advection scheme has thus been implemented, which solves that problem. Preliminary experiments show that
the new advection scheme produces, by itself a significant improvement of the representation of all boundary
currents.




                                     Figure 7.1.1.3: Model domain and bathymetry of NATL4 configuration.



                                                                        4
   New numerical scheme for momentum advection
A new numerical scheme to treat the vorticity term in the momentum equation when the partial step
configuration is used has been implemented and tested in NEMO. This scheme conserves energy and enstrophy.
It results in a better representation of the dynamics, and a considerable reduction of the variance on the vertical
velocity at the bottom.

   The global 1/4° configuration
In this first year, we developed the 1/4 ° global DRAKKAR configuration ORCA-R025 and we carried
out sensitivity experiments to confirm the choices made about parameterisation and numerical scheme
after experiments carried out with South Atlantic (SATL3) and North Atlantic (NATL4) configurations.
This large configuration counts 1442x1021x46 grid points on an ORCA-type 3 polar grid (Fig. 7.1.1.4).
Grid, topography, masking and initial conditions are inherited from the POG global configuration of
Mercator-Ocean.
The implementation on massively parallel machines has been carried out at the IDRIS Centre on the IBM
                                                                      Computer IBM (186 proc.). Taking into account
                                                                      the scientific constraints of the project (in
                                                                      particular the use "partial steps"), and the
                                                                      technical constraints (computing time), it was
                                                                      essential to work with the new version of the
                                                                      code NEMO. Parallelisation required a
                                                                      significant work on the sea-ice model and the
                                                                      implementation of specific procedures of
                                                                      exchange between the processors located in the
                                                                      folding zone of imposed by the two geometrical
                                                                      poles of the grid in the northern hemisphere.
                                                                      Intensive testing of the MPI version of the code,
                                                                      with the partials steps and sea-ice produced
                                                                      exactly reproducible experiments, independently
                                                                      of the way the cutting onto processors is done.
                                                                      Test of the ORCA-R025 configuration have
    Fig. 7.1.1.4: Polar sight of the northern hemisphere of the       been made on 186 processors (Fig.7.1.1.5). In
    computational grid of ORCA-R025 (resolution 1/4°, 1 point out of
    20 is represented). Making the code parallel implied to develop   terms of performance, the time step is 1440 s
    specific procedures of exchange between the processors located in (60 time steps per day), and one year of model
    the folding zone of imposed by the two geometrical poles of the   simulation requires 2200h CPU on 186
    grid.
                                                                      processors, and take about 12 h of elapsed time.




                                                          5
                                 Fig. 7.1.1.5. Decomposition on 186 IBM processors of the
                                 DRAKKAR global 1/4° ocean circulation model. Colours indicate
                                 the ocean depth. Boxes represent the domain account for by a
                                 processors. Crossed boxes are 'land processors' not retained in the
                                 calculation. Numbers in abcissa and ordinate indicate model grid
                                 points.




   Experiments with the global 1/4° ORCA-R025 model
A series of seven 10-years experiments have been run at LEGI. All experiments use the same climatological
forcing . All experiments have in common a laplacian lateral isopycnal diffusion on tracers (300 m2s-1), an
horizontal biharmonic viscosity for momentum (-1.5 1011 m4s-1). In the equatorial wave guide, an extra
laplacian viscosity ( 500 m2s-1) is applied on the first three levels. The analysis of these experiments, as well as
additional experiments, will be carried out in the next 12 month period for validation and tuning of this global
configuration. Early conclusions of the experiments already carried out are listed below.
-   The new vorticity scheme in momentum equation represent the most drastic improvement compared
    to the previous era, in particular on the western boudary current systems.
-   The use of partial steps represents another undeniable improvement with regards to full steps
    representation of the bottom topography. In the Atlantic, this is particularly clear in the Argentine
    basin and the representation of the permanent Zapiola eddy, and in the "North West Corner" off
    Flemish Cap (see Fig. 7.1.1.6)




                                                               6
                        Fig. 7.1.1.6. Simulation G22 with the DRAKKAR configuration ORCA-R025 (with
                        partial step, new vorticity scheme): Time mean (year 8 to 10) of the sea surface
                        height (ssh, in meters). Contour lines at hogh latitudes represent the sea-ice
                        concentration. With regards to full steps representation of the bottom topography,
                        this solution represents a remarkable improvement of the dynamical circulation
                        features. In the Atlantic, this is particularly clear in the Argentine basin (where the
                        Zapiola eddy is clearly marked) and in the "North West Corner".


The present global 1/.4° DRAKKAR configuration ORCA-R025 already represents a significant improvement
compared to other state of the art model at that resolution. Although this model will receive, as planned, a
thorough validation in the next 12 month period, in particular with the addition of the parameterisations
developed within the subtasks 7.112 and 7.113, it is now ready for delivery to MERCATOR on schedule
(D.7.1.1).


Subtask 7.1.1.2. Tidal mixing
A parameterisation of the effect of tides on the vertical mixing has been developed and introduced in the new
version of OPA code, name NEMO. It is based on the idea of Simmons et al. (2004). It uses as input data the
amount of energy dissipated by barotropic tides into internal waves provided by Lyard and Le Provost (2002)
global ocean tidal model. A first test has been made at low resolution in ORCA2-LIM (2° resolution). It shows a
realistic geographical distribution of the vertical diffusivity (Fig. 7.1.1.7). The effects on the global properties of
the circulation of the world ocean are a significant improvement of the simulated thermohaline structure and of
the meridional overturning circulation (Fig. 7.1.1.8).




                                                                  7
                        Fig. 7.1.1.7. Vertical average of the tidally induced vertical eddy diffusivity coefficient.




                     Fig.7.1.1.8. Global effective meridional overturning stream-function at the equilibrium
                       (i.e. after 1,500 years). Upper panel: with a constant vertical diffusion of 0.1 cm/s
                           in and below the thermocline. Lower panel: with a the addition of the tidally
                                                     driven vertical mixing.




Subtask 7.1.1.3. Improved mixed layer
A 300 levels version of ORCA2lim 2° and of ORCA-r05lim 1/2° have been developed to assess the behaviour
of the TKE scheme with a O(1m) vertical resolution in the top 300 meters. The analysis of the 10 yearlong
simulations using the two resolutions is underway.
A 1D vertical model using as much as possible the modules of the NEMO code has been developed and tested.
It provides the framework for all the further development and tests of new vertical physics that will be
undertaken in the following months.


Subtask 7.1.1.4. Contribution to the development of the global 1/4° model
                                                                    8
This task is planned to start at month 18.
However, a first prototype of a global ocean model at an eddy permitting resolution of 1/4° with a fully
prognostic sea ice model is being implemented (referred to ORCA-R025, see section 7.1.1.1) on a massively
parallel computer under MPI. It uses the new version of the OPA code, named NEMO, fully written in Fortran
90. It can be used with full steps or with partial steps. An automatic domain decomposition software has been
especially developed for the ORCA grid.


Subtask 7.1.1.5 Contribution to the development of the 1/12° model

This task is planned to start at month 18.



Task 7.1.2 Nesting methods and Agrif package

The task in on schedule. Progress made in the various subtasks are summarised below.

Subtask 7.1.2.1: Synthesis of the needs and comments expressed by the users of the AGRIF package in WP10.
The AGRIF tool has been (or is currently being) integrated within several regional models used in WP10. These
models are:
    BSHCmod (DMI, model of the Baltic sea)
    OPA9 (DFO, model of the Newfoundland basin)
    Hycom (NERSC, in cooperation with SHOM)
These implementations have already required a number of software developments :
       Simplification of the recursive calls to any part of the code
       optimisation of several Agrif procedures for the treatment of Fortran90 modules
       code transformations for run-time optimisation
       design of a preprocessing tool, which helps producing forcing fields for high resolution grids. This tool
        is available through a Java graphical interface.

Subtask 7.1.2.2 : Development of the AGRIF package
This task is planned to start at month 13. However, a preliminary work has been conducted to answer the needs
and comments expressed by the users of the AGRIF package in WP10.




Milestones reached
M.7.1.1: First implementation and transfer of the global ORCA 1/4° including partial steps (Mo 12)
Deliverables obtained
D.7.1.1: First implementation and transfer of the global ORCA 1/4° including partial steps (Mo 12)
Deviations from work plan (if any)
None




                                                       9
Task 7.2: Biogeochemical modelling

Scientific / Technical Progress made in this task


Task 7.2.1. Biogeochemical model development and assessment
The aim of this task is to develop a prognostic and robust ecosystem model that can successfully describe
the observed temporal and regional variability in phytoplankton biomass and in associated carbon and
nitrogen fluxes. When tested against available data, e.g. via data assimilation procedures, present ecosystem
models generally fail to adequately fit the data. This suggests that models lack important ingredients and are
not complex enough. On the other hand, all data assimilation studies performed so far could not constrain
more than 10-15 parameters or degrees of freedom of the ecosystem models. In fact, even relatively simple
NPZD-type models have already a few degrees of freedom that cannot be constrained by the presently
available data, suggesting that these models might already be too complex. These apparently contradictory
findings about the required level of ecosystem model complexity might indicate that current model
structures are not adequate. Within this task we therefore try to make progress by a new approach fo
following different structural elements for improving marine ecosystem models.
The first structural element to be considered is size. Theoretical studies, laboratory studies, and empirical
evidence from open-ocean systems all indicate that size and physiology are closely related. During the first
year of the project we have concentrated on the role of size for nutrient uptake, phytoplankton growth, and
exudation. In order to keep the number of difficult-to-constrain model parameters as small as possible,
different size classes were not resolved explicitly, but a size-spectral approach was chosen instead. By
combining prognostic equations for the total biomass and the number of cells with the assumption of a
linear log-log spectrum, the slope of the spectrum and hence the entire particle size distribution can be
diagnosed at every model time step from only two prognostic equations which, moreover, share most of the
parameters.
The new implicitly size-structured model attempts to use first-principles whenever possible. One of the
processes implemented so far is nutrient uptake by phytoplankton that, corresponding to diffusion across
cell membranes, depends on the surface-to-volume ratio and hence on the size of the cells (Aksnes & Egge,
1991). The same is assumed to hold for exudation (Bjornsen, 1988). A first consequence of these
assumptions is that small cells are favoured in well-lit oligotrophic regimes, whereas a decrease in light
levels causes an increase in optimum cell size (Figure 7.2.1.1).




                                        A prototype of an NPZD-type model with implicit size spectral
                                        representation in the phytoplankton has been developed and
                                        successfully tested against observations at various JGOFS time
                                        series and process study sites (Figure 7.2.1.2).




                                           Fig. 7.2.1.1.




                                                           10
                        Figure7.2.1.2: Annual cycle of simulated chlorophyll (mg m-3) at six time-series sites.
                         Shading reflects more than 50% of the total chlorophyll in the <5


These initial studies have so far been used one-dimensional model configurations. The same sites will be
used to calibrate the evolving model against observations using data assimilative methods and metrics
currently developed in task 7.2.3 .
A variant, newly developed NPZD+DON model (Huret et al., 2005) has been coupled into an improved
version of the MERCATOR MNATL (1/3°) model. Interannual simulations over the 1996-1999 period
have been analyzed. Model basin-scale structures are being compared with WOCE and AMT (Atlantic
Meridional Transect) in the Western and Eastern Atlantic ocean, respectively (Figures 7.2.1.3 and 7.2.1.4
below). The NPZD+DOM model will also serve as starting point for developing a better mechanistic
description of DOM dynamics. This will be particularly important for simulations of the carbon cycle
because of the generally high carbon-to-nitrogen ratios of the dissolved organic matter. To refine this
nitrogen-based ecosystem model, the work with a cell-quota model including detailed algal physiology
(Salihoglu, 2005) to allow non-Redfield coupling of carbon and nitrogen cycles just got started upon arrival
of Baris Salihoglu in January 2005. It is being tested in a 1D setting at the North Atlantic JGOFS time series
sites.


Task 7.2.2. Integration into OPA

Subtask 7.2.2.1. Prototype of a coupled physical/biological assimilative system

The aim of this sub-task is to assemble the basic elements of the assimilative platform needed to simulate
the primary production in the North Atlantic basin at eddy-permitting resolution. The developments
carried out in WP7 will be continuously incorporated to upgrade the platform in the perspective of the
demonstration experiment in the North Atlantic (WP7.2.2.4).
During the first year of MERSEA, a beta-version of the platform has been developed using a rigid-lid
OPA model of the North Atlantic at eddy-permitting resolution (MNATL), a sequential assimilation
scheme (the SEEK filter) and a first-generation biogeochemical model (N1P1-type). The different
elements of the system are described here below.

Physical model configuration
The physical model is a z-coordinate, rigid lid, primitive-equation model based on the OPA code. The
horizontal resolution is 1/3° in longitude and 1/3° cos(  ) in latitude (i.e. 28 km at 40° N). On the
vertical, the discretization includes 43 levels with a grid spacing that increases from 12 m at the surface to
200 m below 1500 meters. The computational domain covers the North Atlantic from 20° S to 70° N

                                                                    11
latitude, and from 98.3° W to 20.6° E longitude. The Southern boundary at 20°S, the Northern boundary
at 70°N, and the Gibraltar Strait are closed but the model solution is relaxed toward climatology within
buffer zones defined off Portugal, in the Norwegian Sea and along the Southern boundary to simulate the
supply of Mediterranean Water and the exchange with the Arctic and South Atlantic basins. The
atmospheric forcing fields of heat, freshwater and momentum are derived from reanalyses of the ECMWF
6-hour forecasts. The use of prescribed heat fluxes as surface boundary conditions, however, results in
unrealistic sea-surface temperature mainly because of the lack of interactivity between the ocean model
and the atmosphere. To restore some kind of control on the model SST, a « flux correction » formulation
is applied, so that the actual net heat flux is expressed as the sum of two components: a prescribed flux
and a correction term proportional to the difference between the climatological SST and the model
surface temperature.
Biological model
The biogeochemical model adoped in the first version of the platform is the P3ZD biogeochemical model
[Aumont et al., 1998] that was developed at LODYC in order to study the oceanic carbon cycle at global
scale. This model has been derived from the HAMOCC3 (HAmburg Model of Ocean Carbon Cycle)
model which was initially dedicated to the study of the carbon and silica cycle but without explicit
representation of phytoplankton and zooplankton dynamics. P3ZD includes 7 tracers inherited from
HAMOCC3 (phosphate, dissolved inorganic carbon, alkalinity, oxygen, calcite, silicate and particulate
organic carbon), with two additional compartments for phyto- and zooplankton and one compartment for
dissolved organic carbon. These 10 biogeochemical variables are treated as two sub-models. The first
sub-model describes the production-export-remineralization cycle (which will be the focal point for the
demonstration system that is being considered here) and can be understood as a simple NPZD model plus
one compartment for semi-labile dissolved organic material (i.e. N1P1-type model). The second sub-
model is simply driven by the first one, and represents the cycling of carbon and silica.
The first sub-model is schematically represented in figure 7.2.2.1, showing five compartments : phosphate
( PO 4  ), dissolved (DOC) and particulate organic carbon (POC), phytoplankton (PHY) and
      3


zooplankton (ZOO). Although the major limiting nutrient in the ocean is nitrate, phosphate is chosen as
limiting nutient in this model to get rid of nitrification processes, exchange of inorganic nitrogen with the
atmosphere and bacterial degradation.




                 Figure 7.2.2.1. Schematic representation of the production-export-remineralization cycle in P3ZD


Assimilation scheme
In this first prototype, the physical model variables only are governed by the assimilation process. The
assimilation method is based on a Kalman filter with reduced order error covariance matrix, known as the
SEEK filter. A first comparison with the assimilation run obtained using the scheme developed by Testut
et al. [2003] indicates the excessive supply of nutrients in the euphotic zone through spurious mixing and
advection mechanisms. This can be partly attributed to several factors, e.g. the statistical method which is
unable to maintain the model constraint of hydrostatic stability, the discontinuous nature of the sequential
algorithm, or the lack of consistent corrections between the physical and biological components of the
                                                     12
state vector. To overcome this problem, three variants of the assimilation algorithm have been developed
to improve the representation of the model dynamics and its subsequent impact on the biological
variables: (i) a “restratification” approach based on the restoration of monotonic vertical density gradient
after the analysis stage; (ii) an « incremental analysis updating » scheme to ensure temporal continuity of
the assimilation run, and (iii) a method combining statistical corrections of T/S properties in the mixed
layer and a lifting-lowering of isopycnals below.
First assessment of the coupled assimilative platform
In order to roughly check how the assimilation can impact the representation of the biological fields,
comparisons have been made between free runs and simulations with assimilation. In these experiments,
the assimilated observations consist of Sea Surface Temperature (SST), Sea Surface Salinity and Sea
Surface Height (SSH) data. The SST data originate from AVHRR observations gathered and processed
within the NASA Pathfinder project (figure 7.2.2.2). The altimetric data is obtained as a sum of a time-
invariant dynamic topography computed by inverse methods and Sea Level Anomalies from Aviso
project combining Topex/Poseidon and ERS altimeter data. The SSS data are monthy climatological
estimates from Levitus. The observation error have been specified as a sum of a measurement error (5 cm
for SSH and 0.5°C for SST) and an estimated representativeness error.

          AVHRR SST (7 day )                       TP along track data (7 day)           Levitus SSS

                                                                              cov
                                                                              erag
                                                                               e)




                      Figure 7.2.2.2. Model domain and examples of data sets assimilated in the physical model


Simulations performed with the original assimilation scheme show unrealistic input of nutrient in the
euphotic zone because of spurious assimilation-induced mixing and advection. This can be explained by
the fact that purely Gaussian estimation error of T/S properties are inconsistent with the hydrostatic
stability assumption made in the physical model. Annual simulations during 1998 with online coupling
have been performed to assess the benefit of the new strategies by comparison with the free run, using a
number of physical and biogeochemical diagnostics. Figure 7.2.2.3 illustrates the response of the system
(expressed in surface chlorophyll concentration) with and without assimilation of data in the physical
component of the coupled model.




Coupling without assimilation            Coupling with assimilation                     SeaWiFS data




                                                                13
Figure 7.2.2.3. Surface Chl_a concentration (in mg/m3) obtained on April 24th, 1998 in the coupled simulation without assimilation (left), in the
coupled simulation with assimilation (center), and from SeaWiFS data (right)



Future developments
During the second year of the project, the physical model component will be upgraded using a new OPA
free-surface model of the North Atlantic at 1/4° resolution and bulk parameterizations of air-sea fluxes
(NATL4). The biogeochemical component will be made more complex using the second- and third
generation models developed in WP7.2.2.3 (N2P1-type available at month 18, and N4P2-type available at
month 36). The assimilation scheme will be upgraded using the free-surface version of the SEEK filter
(available from WP7.3.2 at month 18) and improved mixed layer control (available from WP7.3.2 at
month 30).


Subtask 7.2.2.2. Assimilation impact on biogeochemical variables.
This task has just started.


Subtask 7.2.2.3. Improved biogeochemical processes
To study the role of the different type of nutrients on patterns of marine primary productivity, we have
used 3 differents biogeochemical model with an increased complexity approach:
- A simple phosphate based model (P3ZD);
- A nitrogen model (LOBSTER) including an explicit regeneration loop using ammonia;
- A multi-plankton type model (PISCES) using both macronutrients (P, Si) and micronutrients (Fe).
To compare these models, we have carry various experiments using same physical analysis made by the
MERCATOR system (MNATL-PSY1V1).

Subtask 7.2.2.4. North Atlantic bloom experiment
Different biogeochemical models have been coupled in an offline mode with MERCATOR system. We have
carry various experiments using same physical analysis made by the MERCATOR system (MNATL-PSY1V1)
for the year 1998.
New simulations using PSY1V2 system, allowing both assimilation of in-situ and satellite physical variables,
are under way with a special attention on plankton burst due to shock at the base of the mixed layer induced by
the assimilation procedure.


Task 7.2.3. Assessment and metrics


Data required for validating (satellite ocean colour and biogeochemical time-series data) and forcing for the
ecosystem models have been collected. As a next step, data assimilation experiments are planned to be carried
out with the biogeochemical models and developed parameter estimation schemes.
Subtask 7.2.3.1. Identification of key observational data and the definition of an internal metrics.
One-dimensional time-series data have been collected for the JGOFS sites BATS (320N, 640W), NABE
(480N, 210W) and EUMELI as well as for the Labrador Sea and Ocean Weather Station (OWS) “India”
(590N, 190W). The data sets contain: chlorophyll and nutrient data; estimates of primary and, at some
stations, bacterial production; particular organic matter data are also available. In addition, SeaWiFS ocean
colour data (1997-2004) and nitrate data (World Ocean Database, 1998) on a global and basin (North
Atlantic) scale have been made available.
At the present stage of the study, BATS chlorophyll “a”, Chlobs, and dissolved inorganic nitrogen, DINobs,
data are the only data used to constrain the ecosystem models developed in WP 7.2.1.

                                                                      14
                C, P = Arg min { ( Chlmod - Chlobs)2               chl_obs
                                                                          -1
                                                                               + ( DINmod - DINobs)2             DIN_obs
                                                                                                                        -1
                                                                                                                             },
here, C and P ecosystem model components { Chlmod , DINmod …} and biological parameters,
respectively;    chl_obs and DIN_obs are the error level of the observations. Investigation of the impact of
different formulations of and weightings in the cost function is needed and is presently under way.
Subtask 7.2.3.2. Intercomparison and assessment of biogeochemical models:
A version of the SIR filter has been applied for assimilating the Bermuda Atlantic Time-Series chlorophyll
and dissolved inorganic nitrogen data into the REcoM and the current version of the size dependent NPZD
ecosystem model (developed at Kiel, WP 7.2.1) in order to estimate poorly known biological parameters.
Tables 7.2.3.1 and 7.2.3.2 possess results of the first parameter estimation experiments at the BATS site
with the RecoM and size dependent NPZD ecosystem model, respectively.


   Symbol               Parameter                                                           Initial               Optimal                 Units
                                                                                            Value                 Value
   lossN                phytoplankton loss of organic nitrogen                              0.05                   0.048                  day-1
   lossC                Phytoplankton loss of organic carbon                                0.40                   0.27                   day-1
                        Initial slope of the P-I curve                                      0.10                  0.12                    D2W1day-1
   Vp*                  Phytoplankton maximum drowth rate constant                          0.70                  0.74                    day-1
   resH                 Respiration by heterotrophs                                         0.01                  0.008                   day-1
   lossH                Heterotrophs mortality                                              0.10                  0.079                   day-1
   Agg1                 Stickiness for PCHO-PCHO                                            0.0075                0.0062                  day-1
   Agg2                 Stickeness for TEP-PCHO                                             0.24                  0.22                    day-1
   degChl               Chlorophyll degradation rate constant                               0.05                  0.04                    day-1



                                              Table 7.2.3.1. RecoM biological parameters for BATS site



           Symbol         Parameter                                                Initial value         Optimal value            units

            P             Maximum phytoplankton growth rate                        2.00                  1.8                      day-1
           Kno3           Half-saturation constant for nutrient uptake             0.05                  0.026                    mmolNm-3

            P             Phytoplankton mortality                                  0.03                  0.016                    day-1

            1             Maximum phytoplankton exudation rate                     0.25                  0.12                     day-1

            Z             maximum zooplankton grazing rate                         2.00                  1.22                     day-1
           kG             Zooplankton ingestion half-saturation constant           0.50                  0.50                     mmolNm-3
           kZ             quadratic zooplankton mortality                          0.20                  0.12                     mmolNm-3

            2             zooplankton excretion rate                               0.03                  0.017                    day-1

            4             detritus remineralization rate                           0.05                  0.032                    day-1

            5             DON remineralization rate                                0.05                  0.023                    day-1


                                         Table 7.2.3.2. Size dependent NPZD model biological parameters.



Figure 7.2.3.1. depicts the monthly means of the REcoM model chlorophyll (right panels) and dissolved
inorganic nitrogen (left panels), simulated with the initial set of the model biological parameters (upper
penels) and obtained after the data assimilation (middle panels), against the BATS chlorophyll and DIN
data.



                                                                           15
                                            Figure 7.2.3.1.



Results of the data assimilation experiment carried out for the size dependent NPZD ecosystem model
developed in WP 7.2.1 are presented on figure 7.2.3.2.




                                                    16
                                                  Figure 7.2.3.2.



As a next step, this will be expanded to the other 1D sites, before the SIR filter will also be applied to the
LOBSTER and PISCES models used in WP 7.2.2 to provide best biological parameter estimates for the
ecosystem model versions.


Milestones reached
none
Deliverables obtained
none
Deviations from work plan
The experience post-doc (Baris Salihoglu) could be recruited for the Toulouse part of subtask 7.2.1 only on
January 1st, 2005, that is 8 months after the start of the project. Still, the development of the new ecosystem
model is well under way,but some shifts in delivering the coupled carbon-nitrogen model may be expected.




                                                       17
Task 7.3: Data assimilation

Scientific / Technical Progress made in this task

Task 7.3.1: Development of advanced sequential filters


Subtask 7.3.1.1. Ensemble Kalman filter


-Starting point and objectives:
The EnKF is an open-source freeware downloadable from http://www.nersc.no/~geir/EnKF. The
assimilation of ice parameters has been studied in Lisæter et al. (2003). It is now implemented in the
TOPAZ system for assimilation of ice concentrations. The forecast bulletins can be consulted on
http://topaz.nersc.no. Our objective is to explore advanced assimilation schemes
for ocean forecasting problems.

-Progress in the task:
Improved sampling procedures (cf. Evensen, Ocean Dynamics 2004) have been applied in the TOPAZ
system in order to improve results for a lower CPU cost. Initital tests of different analysis schemes have
been performed (e.g. perturbations of measurements and square-root schemes) All improved routines are
served and documented on the EnKF web page.
A bias reduction technique has been tested in the context of assimilation of simulated « CRYOSAT-like »
ice thickness into a half-resolution prototype of the TOPAZ system. The tests show measurable
improvements and an article is under preparation (Lisæter 2005).

Subtask 7.3.1.2. SIR filter

The required software for coupling SIRF to ecosystem models has been developed. The guided
resampling has now been implemented and put into the SIRF code in order to reduce the number of
model trajectory ensemble members needed for the data assimilation scheme.Validation of the data
assimilation technique with the RecoM and size dependent NPZD ecosystem models is planned for the
next few months. A version of the SIR is being tested which uses local updating.This means that only
those ensemble members are usedin the ensemble that are close to the observations locally, and members
from different locations are glued together using the relative weights of the ensemble members, since the
latter will be smooth functions of space.

Subtask 7.3.1.3. Adaptive SEEK filter

Task not started.


Task 7.3.2: Global OPA data assimilation system

Subtask 7.3.2.1 Transition to free surface model

During the first year of the project, the coupling of altimetric data assimilation with OPA free surface
global configuration has been implemented, and an algorithm (IAU scheme) to properly initialise the
model forecast (eliminating inconsistent initial conditions) has been developed. This has been evaluated
using twin experiments with the ORCA2 global configuration, with an example of the control of model
errors.

a. The model configuration


                                                    18
The OPA global free surface configuration that will be used for the development of the assimilation
scheme is the ORCA2 configuration. This is a low resolution configuration (about 2°x2°, with a
meridional grid spacing refinement going down to 1/2° in the tropical regions) from the ORCA grid
family (see Figure 7.3.2.1). This is coherent with OPA model developments (task 7.1) and operational
applications (WP9) which are using global configurations from the same ORCA grid family. The low
resolution is justified here (task 7.3.2) in a first stage because the first development of the assimilation
that are scheduled are not very affected by the model resolution. It is then only in a second stage that the
new assimilation developments will be checked with higher resolution regional model. A good candidate
                                                           for this (again from the ORCA grid family) is the
                                                           North Atlantic 1/4° resolution configuration
                                                           (NATL4).

                                                         The model air-sea fluxes are computed from bulk
                                                         formulation, feeded by daily NCEP data and
                                                         ERS/TAO winds. Relaxation to Levitus surface
                                                         salinity climatology is applied.

                                                         b. The assimilation scheme

                                                       The next generation of assimilation schemes that
                                                       is currently being transferred to the MERSEA
                                                       operational system is derived from a reduced
                                                       order Kalman filter (the SEEK filter) developed
                                                       at LEGI for the last few years. Figure 7.3.2.2
                                                       gives a summary of the SEEK filter possibilities.
                                                       The objective of this task (7.3.2) is to provide the
                                                       developments of the scheme that are necessary to
                                                       fit the requirements of operational use, focusing
                 Figure 7.3.2.1: ORCA2 model grid.     on those that are most expected to generate
                                                       immediate, first order and low cost operational
                                                       benefit. Moreover, the developments that are
planned in this project are quite independent of the particular scheme in use, and could potentially be
plugged in most sequential assimilation algorithm.

The SEEK filter was chosen because it is thought to make a significant progress with respect to currently
used optimal interpolation scheme, and because of its immediate availability. It is obvious, however, that
it is affected by well-known difficulties that need to be considered to build the next generation of
assimilation schemes. This is the purpose of task 7.3.1 (see above).

As any other Kalman filter, the SEEK filter requires a parameterisation of the error covariance on the
various sources of information: initial condition (P°), observations (R), and model (Q), that need to be
thought specifically for any kind of experiment (see below). Concerning the assimilation of altimetry into
ORCA2, we always use in this study an estimation space covering all variables of the state vector:
temperature (T), salinity (S), zonal (U) and meridional (V) velocity.

c. The initialisation scheme




                                                    19
In the standard Kalman filter, it is assumed that a correction is computed and applied to the model state
each time a new observation is available. In our implementation, however, as done in most applications,
and practically in all ocean basin-scale operational system currently running, we divide the time in a
number of equal periods or assimilation windows. For each assimilation window, we gather all
innovations collected during that assimilation window, and we make only one analysis and model
correction at the end of each assimilation window. The justification for such suboptimal scheme is that,
having more observations for each statistical analysis, the solution is less sensitive to approximation in
                                                                        the parameterisation of the
                                                                        various error covariance matrices
                                                                        (P°, R, Q). The optimal
                                                                        assimilation window is then a
                                                                        compromise         between     the
                                                                        minimisation of the model drift
                                                                        between two successive analyses,
                                                                        and the minimisation of spurious
                                                                        effects due to inadequate error
                                                                        parameterisation. This problem is
                                                                        of course amplified by the lack of
                                                                        observations, which is a constant
                                                                        preoccupation         in    ocean
                                                                        assimilation systems, most often
                                                                        underconstrained         by    the
                                                                        observations. In our study, the
                                                                        assimilation window has been set
                                                                        to 5 days.

                                                                         The consequence of this situation
                                                                         is that the correction that we
                                                                         apply to the model can often be
                    Figure7.3.2.2: Schematic of the SEEK                 non negligible with respect to the
                            assimilation method.
                                                                         signal that we are controlling.
                                                                         This leads to two kinds of
problems: (i) significant time discontinuity of the solution, and (ii) spurious high frequency oscillations
(like gravity waves in a free surface ocean model), if no adequate initialisation procedure is applied. The
purpose of this subtask (7.3.2.1) is precisely to eliminate these two problems from global OPA free
surface configuration with altimetric data assimilation.

In order to tackle these problems, Bloom et al. (1996) have proposed an algorithm called Incremental
Analysis Update (IAU), consisting in incorporating the analysis increment in a gradual manner. There are
many variants of the IAU algorithm, differing by the position of the time window during which the
increment is incorporated. The variant that we have chosen for this study can be described as follows (see
figure 7.3.2.3): a first guess model forecast is done from time i to i+1 and an analysis is performed at time
i+1. An increment is calculated on the state vector using the Kalman gain as in the usual Kalman filter.
Then the model is run again from time i to i+1, but with the increments applied gradually to all variables
of the state vector. The obtained state at i+1 will then provide the initial conditions for the next model
forecast. It is expected that these new initial condition remain close to the Kalman analysis state, but that
any unbalanced or non-physical increment has been removed. In that sense, this IAU forecast can also be
viewed as a model reinitialisation procedure.

During this IAU forecast, the additional forcing term imposed to the model equations is a fraction of the
analysis increment x:
                                 dx/dt = M (t) x + (t) x with 0tc (t) dt
In the present case, as in most IAU methodologies, this forcing term is the increment divided by the
duration of the assimilation cycle tc.

                                                     20
                                              (t) = 1/tc
Hence, this value is constant during the IAU forecast and the time-integrated quantity of this value over
the IAU forcing time window is the increment.

                                                                                This variant of the IAU scheme
                                                                                leads to an increase in the
                                                                                integration time of the model of
                                                                                100%. This particular variant
                                                                                was chosen because it offers the
                                                                                possibility to compare the
                                                                                forecast, the analyzed and the
                                                                                IAU corrected states at the same
                                                                                time. In the IAU variant from
                                                                                Bloom et al. (1996), half of the
                                                                                increment is applied before the
                                                                                analysis time and half of the
                                                                                increment is applied after the
                                                                                analysis time, only increasing
                                                                                the cpu time by 50%.

                                                                                d. Twin assimilation
                                                                                experiments
                  Figure 7.3.2.3: Schematic of the IAU initialisation scheme.

                                                                            This assimilation system (the
SEEK filter assimilating altimetric data within ORCA2) with the new IAU initialisation scheme has been
tested using twin experiments. (Realistic assimilation experiments with higher resolution model [NATL4]
will be performed before this system is delivered [at month18].) The experiment setup is designed in such
a way that the only source of error in the system is model error. This is indeed an efficient way to produce
a strong drift of the model during one assimilation cycle, leading to large SEEK corrections, and hence,
an effective checking of the IAU protocol.

The reference simulation (the true ocean) is the standard ORCA2 interannual simulation for the year
1993. Synthetic altimetric observations are then sampled from this reference simulation to be assimilated
in a modified simulation (the false ocean) in which the atmospheric forcing has been altered. The initial
condition has been kept the same, so that the only source of error is the atmospheric forcing: the wind
stress has been doubled in the tropical zone (between -30° and +30°) with respect to the reference
simulation. Again such unrealistically large model error has been produced in order to provide an
adequate platform for IAU evaluation.

Since it is not the purpose here to study the way to produce an effective error subspace for the SEEK
filter, we choosed an ideal situation on that respect (which does not affect the validity of the IAU
evaluation). We assume that the true model is known so that we can generate an ensemble of differences
between the true and false model 5-day forecasts from an ensemble of initial conditions (the same for the
2 models). The covariance of this ensemble is then a good estimator of the 5-day model error covariance,
and is adequate to parameterize the forecast error covariance in the SEEK filter (assuming that the initial
error covariance remains small cycle after cycle). As an ensemble of initial conditions, we used model
states from the reference simulation every 5 days (73 members). From each initial condition, we run the
true and false model 5-day forecasts using forcing data synchronized with the initial conditions, this
allowing to sample also the forcing variability over the year 1993. From the ensemble of differences after
5 days, we retain only the first 30 EOFs to parameterize the SEEK forecast error covariance matrix.
Moreover, we used this matrix globally as it is, without any further local parameterization.

e. Evaluation of the scheme

Figure 7.3.2.4 and 7.3.2.5 show the RMS error for the false ocean without assimilation (green curve) and
                                                               21
with assimilation of altimetric data (red curve). The figure also shows the RMS error for the first guess
forecasts (without the IAU increment) using the false model (black curves) and for the SEEK analysis
applied to these forecasts (black circles). This is the analysis error before IAU initialisation. We can see
from the figure that the IAU initialisation only slightly increase the analysis error, while we expect a
substantial benefit in the forecast.


                                                       Figures 7.3.2.6, 7.3.2.7 and 7.3.2.8 present the
        Figure 7.3.2.4.: Temperature RMS error.        annual mean solution obtained in the Tropical
                                                       Pacific: mean sea surface elevation (figure
                                                       7.3.2.5), mean zonal velocity (figure 7.3.2.6) and
                                                       mean temperature section (figure 7) along the
                                                       equator. The true ocean, the false ocean and the
                                                       assimilation solution (IAU analysis) are
                                                       compared, demonstrating the ability of the
                                                       assimilation scheme to correct the wind model
                                                       error successfully and drive the false model close
                                                       to the true ocean. A closer look to the behaviour
                                                       of the scheme, assimilation cycle after
                                                       assimilation cycle, whows that this is due to the
                                                       ability of the IAU increment (on temperature,
salinity, and velocity) to compensate precisely the model error due to the wind. In the deep ocean, it is the
temperature and salinity increment that is most important, while close to the surface, the velocity
increment is required.

This study demonstrates an implementation of the IAU initialisation scheme into an OPA free surface
global data assimilation system. Before delivery of this system (at month 18) it will be evaluated using a
        Figure 7.3.2.5: Sea surface elevation RMS error
                                                         1-year hindcast experiment (i.e. using real data)
                                                         with a higher resolution regional model
                                                         (NATL4). During this second phase of IAU
                                                         evaluation, we will focus on the diagnostic of the
                                                         model forecast from IAU initial conditions. This
                                                         is indeed where we expect the best improvement
                                                         with respect to previous results.




                                                     22
       Figure 7.3.2. 6: Annual mean sea surface elevation for the
        true ocean (top left), the false ocean (top right) and the
                     assimilation solution (bottom).




     Figure 7.3.7.: Annual mean zonal velocity along the
     equator for the true ocean (top left), the false ocean
      (top right) and the assimilation solution (bottom).




23
                             Figure 7.3.2.8.: Annual mean temperature along the equator for the true
                                                  ocean (top left), the false ocean
                                         (top right) and the assimilation solution (bottom).




7.3.2.2 In situ/satellite data assimilation scheme

During the first year of the project, a data assimilation scheme assimilating jointly in situ and satellite
observations into OPA free surface global configuration has been implemented (including the IAU
scheme to initialize the model forecast).

a. The assimilation system

The assimilation system is here the same as in subtask 7.3.2.1. Again, the ORCA2 model is used for
assimilation development purpose. The assimilation scheme is also the SEEK filter, together with the
IAU initialisation scheme.

The first difference lays in the observation space which now includes temperature and salinity profiles in
addition to altimetry. The purpose is indeed to develop and evaluate this system (analysis scheme and
initialisation protocol) to assimilate jointly in situ and satellite observations. For that purpose, we will go
a little further in the realism of the simulations that we are going to perform: we will still evaluate the
system using twin experiments, but we will not assume anymore that we know the true model when we
estimate the SEEK forecast error covariance matrix.


                                                              24
In that context, the SEEK error parameterization will be less reliable, and the local SEEK approximation
will be required. Indeed, when a reduced order representation of a covariance matrix is used, the
confidence interval for small correlation coefficients is always the largest. As a consequence, long range
correlation coefficients (which are usually small) are often overestimated, inducing spurious long-range
observation influence. The local approximation of the SEEK filter has been designed to eliminate these
lon-range influence by setting the corresponding correlation coefficients to zero.

In practice, this means that the model state vector is divided in a number of subsystems for which
separate statistical analyses are performed. Each analysis is applied on the set of observations that is
inside the influence bubble of the corresponding subsystem. These influence bubbles must be
parameterized; they define the spatial domain outside which the correlation coeffcients are set to zero.

                                                       In the implementation of the local SEEK filter
                                                       applied in MERSEA, the subsystems of the model
                                                       state are specified as the water columns of the 3D
                                                       ocean model. The influence bubbles of every
                                                       subsystems are then specified as a floating mask (of
                                                       adjustable dimension) centered successively on every
                                                       water columns of the ocean model. As a consequence,
                                                       an upgrade of the algorithm was needed to apply this
                                                       scheme to a global model with grid periodicity
                                                       conditions. For instance, a subsystem close to the
                                                       Western edge of the grid must be associated to an
                                                       influence bubble extending also close to the Eastern
                                                       edge of the grid, which corresponds to the same
                                                       geographical region. Hence, when the floating mask
                                                       defining the influence bubble is pasted onto the
  Figure 7.3.2.9.: Correction obtained from a single   model grid around such subsystem, care must be
   observation close to the Eastern edge of the grid.  taken to cut the part of the mask extending further
                                                       than the Western (resp. Eastern) limit of the grid and
paste it at the Eastern (resp. Western) limit of the grid. Figure 7.3.2.8 shows an example of the increment
generated by the analysis of one observation close to the Eastern edge of the grid with this enhancement
of the local SEEK algorithm. This procedure is obviously necessary to ensure the continuity of the
analysis accross the grid cut.

b. Twin assimilation experiments

This assimilation system (the SEEK filter assimilating altimetric and in situ data within ORCA2) with the
IAU initialisation scheme is being tested using twin experiments. Again, as in subtask 7.3.2.1, the
experiment setup is designed in such a way that the only source of error in the system is model error. But,
here, instead of doubling the wind in the tropical regions, we decided to introduce perturbations in key
parameters of the bulk formulation for the air-sea fluxes of heat and fresh water. This kind of simulation
will also be used in subtask 7.3.2.3 (from months 19 to 30) to evaluate the possibility to correct the model
atmospheric forcing by controlling the bulk formulation parameters. Here, we use the same kind of
experimental setup with an assimilation scheme that corrects the model state vector only.

The reference simulation (the true ocean) is the standard ORCA2 interannual simulation for the year
1993, with the original bulk formula. Synthetic altimetric, temperature and salinity (profiles) observations
are then sampled from this reference simulation to be assimilated in a modified simulation (the false
ocean) in which the bulk formulation has been modified. Four possible perturbations have been
                                                     H W (Tw-Ta
specific heat, W is the wind speed, Tw and Ta are the sea and air temperatures) can be simplified using
constant bulk coefficient CH instead of the default complex parameterization, (ii) the latent heat flux
QL         E W max(0,qs-qa
speed, qs and qa are the surface and atmospheric specific humidities) can be simplified using constant bulk
                                                     25
coefficient CE instead of the default complex parameterization, (iii) the precipitation can be multiplied by
a constant factor CP, and (iv) the cloudiness can be multiplied by a constant factor CC. In this experiment,
the false ocean is defined by the set of parameters: CH=10-3, CE=1.12 10-3, CP=1, CC=1.

In this experiment, we do not assume anymore (as in the twin experiments presented in subtask 7.3.2.1)
that the true model is known to parameterize the forecast error covariance matrix. In order to generate a
sensible parameterization, we still apply the same kind of procedure as in subtask 7.3.2.1, only replacing
the true model where it is used by an ensemble of models with different values for the bulk parameters.
Table 1 summarizes the ensemble of bulk parameters that have been used. The dispersion of the
parameters in this ensemble should be consistent with a priori information on the parameters error
covariance. We also keep an ensemble of initial conditions (the same as in subtask 7.3.2.1) to increase the
size of the sample of 5-day forecast differences. From this sample, we retain only the first 20 EOFs to
parameterize the SEEK forecast error covariance matrix.


                                     Member      CE        CH   CC CP
                                         1    1.12 10-3 1. 10-3 1 1
                                         2    1.00 10-3 1. 10-3 1 1
                                         3    0.90 10-3 1. 10-3 1 1
                                                                                Table 1.
                                         4    1.20 10-3 1. 10-3 1 1
                                                                                Ensemble of
                                         5    1.30 10-3 1. 10-3 1 1             bulk
                                                                                parameters
                                         6    1.12 10-3 0.90 10-3 1 1
                                         7    1.12 10-3 0.80 10-3 1 1
                                         8    0.50 10-3 0.50 10-3 1 1
                                         9    0.10 10-3 0.10 10-3 1 1
                                        10    0.60 10-3 0.60 10-3 1 1



c. Evaluation of the scheme

First test experiments are currently being performed. The evaluation of the scheme on the twin
experiment presented above must be continued in the next 6 months, so as to obtain a scheme usable by
the subtask 7.3.2.3 between month 19 and month 30. Then, the developments in substask 7.3.2.2 will be
suspended until month 30. At that time, the work in task 7.3.2.2 will resume: combining the results obtain
in subtask 7.3.2.1 and 7.3.2.3, we will (i) perform a reanalysis with ORCA2, (ii) evaluate the possibility
for the assimilation in a higher resolution model (i.e. NATL4) to benefit from previous assimilation in a
lower resolution model (i.e. ORCA2), and (iii) perform a reanalysis using a regional higher resolution
model (i.e. NATL4) using all developments obtained in task 7.3.2.




Milestones reached
None
Deliverables obtained
None
                                                      26
Deviations from work plan (if any)
None.

Table 1: Deliverables List

Del     Deliverabl   Workpa      Date      Actual/F    Estimated     Used           Lead contractor
 .       e name       ckage      due       orecast     indicative   indicati
no.                    no.                 delivery     person-        ve
                                             date      months *)    person-
                                                                    months
                                                                       *)
7.1.   First         WP7.1     31/05/0    31/05/200 32.5            32.5       CNRS/
1      version of              5          5                                    LEGI
       the global
       1/4° model
       ORCA-
       R025
*) if available

Table 2: Milestones List
Milestones are intermediate steps leading to Deliverables, but distinct from them


Milestone Milestone name             Workpackage    Date due         Actual/Forecast     Lead
    no.                                  no.                          delivery date    contractor
7.1.1     First version of the       WP7.1       31/05/2005          31/05/2005      CNRS/
          global 1/4° model                                                          LEGI
          ORCA-R025




                                                  27

								
To top