Towards a variational data assimilation system for NEMO_ The .._1_ by malj


									   Ocean Re-analysis within the
  European ENSEMBLES project

Contributors: CERFACS, ECMWF, INGV, Met Office

           Baseline experiment
 Ocean reanalyses produced for the period 1960-
  2005 in order to initialize seasonal to decadal
  climate “predictions” (hindcasts).
 Data for assimilation
  – Temperature and salinity profiles from the quality-
    controlled ENSEMBLES in situ data-base (EN3v2;
    Ingleby and Huddleston 2007, J. Mar. Sys.)
 Surface forcing
  – ERA40 (with bias-corrected precipitation) up to 2002.
  – ECMWF operational analyses thereafter.
 Experimental set-up (spin-up, relaxation,…) close
  to that of ENACT.
Baseline experiment

ENSEMBLES (EN3) quality-controlled in situ ocean
      temperature and salinity profiles
   Data sources: WOD05, GTSPP, Argo GDAC
   Processing and QC designed for data assimilation:
     – Thinning: 1hr in time; +/- 0.2o lat/lon
     – “Superobbing”: daily averages of some buoy data

          Data assimilation systems
  –   OPA (ORCA2) model; 2o x 2o with equatorial refinement, 31 levels
  –   3D-Var FGAT with 10-day window, and IAU
  –   Bck. error covariances: multivariate balance, variances ~ dTb/dz
  –   Obs. error covariances: uncorrelated, variances using Fu et al. method
  –   Refs: Weaver et al. (2003), MWR; Weaver et al. (2005), QJRMS
  –   HOPE-E model; 1o x 1o with equatorial refinement, 29 levels
  –   OI with10-day window and IAU
  –   Altimeter data as well as in situ data
  –   Balance operators (S(T), geostrophy)
  –   Online model bias correction
  –   Ref: Balmaseda et al. (2007), ECMWF Tech. Memo 508

         Data assimilation systems
  – OPA (ORCA2) model; 2o x 2o with equatorial refinement, 31 levels
  – Reduced-order OI with 14-day window
  – Bck. error covariances: seasonally-dependent bi-variate (T-S) EOFs
    computed from the model
  – Obs. error covariances: uncorrelated, constant variances for T and S
  – Ref: Bellucci et al. (2007), MWR
 Met Office (GloSea)
  –   Unified Model, 1.25o x 1.25o with equatorial refinement, 40 levels
  –   Analysis Correction method (OI-type)
  –   Online model bias correction (pressure correction method)
  –   S(T) adjustment
  –   Ref: Troccoli and Haines (1999), JAOT; Bell et al. (2004), QJRMS;
      Martin et al. (2007), QJRMS.
        Data assimilation systems
 Met Office (DePreSys)
  – HadCM3 (coupled model)
  – Assimilate observed anomalies added to coupled model climatology
  – Relax coupled model to ocean and atmospheric (ERA40) analyses
  – Monthly ocean analyses produced offline using 4D multivariate OI of T
    and S profiles and SST (objective analysis – no ocean model)
  – Error covariances: based on EOFs computed from the coupled model
  – Ref: Smith et al. (2007), Science

 Ocean reanalysis production is complete for some
  partners, ongoing for others.
 Some participants have produced ensembles of
  ocean reanalyses.
  e.g., 5 sets for ECMWF; 9 sets for CERFACS
  – Ensembles have been produced using perturbed
    atmospheric forcing fluxes (windstress, SST, precipitation)
  – Forcing perturbations produced for ENSEMBLES by
    ECMWF, and based on differences between
        ERA40 and CORE for windstress
        ERA40 (bias-corrected) and NCEP for precipitation
        Reynolds 2DVAR and OI for SST (1982-)
        NCEP ERSST and HadSST (pre 1982)
Early comparison between CERFACS and
             INGV analyses
  (same model but different assimilation methods)

            Heat content anomalies 0-300 m
      Global                    Northern Hemisphere

Impact of ocean data assimilation on ECMWF
       seasonal forecasts (System 3)
                               (courtesy: M. Balmaseda, ECMWF)
                                             NINO4 SST rms errors
                                                 76 start dates from 19870101 to 20051001
                                                 Ensemble sizes are 3 (esj6) and 3 (esj6)

                                Fc esj6/m1         Fc esj6/m0               Persistence         Ensemble sd
                 0.8                         S3 No data                     S3 Assim
   Rms error (deg C)




                           0       1         2               3              4               5       6         7
                                                    Forecast time (months)

                               Data assimilation improves seasonal forecast skill                                 10
     Archiving and dissemination
 Archiving of output on an ENSEMBLES public data-
  server at ECMWF is planned.
 Monthly means of a common set of 2D and 3D
  variables (including transports).
 Output on a common (regular) grid (Levitus-like) as
  well as the native model grid.
 CF compliant NetCDF format.
 Possibly all ensemble members to be made


To top