Data assimilation and forecast model changes (Cy28r3)

29 September 2004

Generic news item image

The data assimilation, forecast and ensemble system used for medium-range forecasts has been upgraded successfully on 28 September 2004. At the same time the operational suite has been transferred to the IBM Phase 3 High Performance Computing Facility. Several new sets of observations have been activated (Meteosat-8 radiances, SCIAMACHY ozone products from KNMI, Early orbits from ATOVS as retransmitted by EUMETSAT-EARS); the numerics of the convection scheme has been significantly improved; the targeting of tropical cyclones for the EPS was refined, and a Gaussian sampling of all initial EPS perturbations was introduced in replacement of the old selection/rotation procedure.

A new version of ECMWF model Cycle 28r3 is in the final phase of pre-operational testing. Start date of the parallel experimental suite was 1 August 2004. Sample data of the deterministic and EPS runs can be retrieved from MARS with expver=15.

Changes from the currently operational version (Cycle 28r2) include:

  1. Physics
  • Revised convection scheme numerics and calling of the cloud scheme
  • Use of tangent linear and adjoint of vertical diffusion in the 1st minimization of 4D-Var
  • Reduction of radiation frequency to 1 hour in the high-resolution forecasts
  • Improved numerics of surface tile coupling
  • Post-processing of total- column liquid water and ice.

2.  Satellite

  • RTTOV-8
  • Minor revisions to ATOVS and AIRS usage
  • Assimilation of MSG clear-sky radiances and GOES 9 BUFR AMVs
  • Assimilation of SCIAMACHY ozone products from KNMI
  • Correct error in AMSU-B usage over land
  • Activate EARS data.

3.  Data assimilation

  • Blacklist SYNOP humidity data at local night time (solar elevation SOE less than zero).
  • Increased use of radiosonde humidity: use RS90 to -80C, RS80 to -60C, other sensors to -40 C
  • Proper cycling of the information from the wave altimeter and the surface data analyses (FG forecasts moved from 00 and 12 UTC to 06 and 18 UTC).

4.  EPS

  • Gaussian sampling for extra-tropical singular vectors (instead of selection and rotation)
  • Revision of initial condition perturbations for tropical cyclones (TC)



  • Initial condition perturbations extended to latitude belt 40 S 40 N (from 25 S 25 N);
  • Tropical singular vectors are computed in the subspace orthogonal to the leading 25 extra-tropical singular vectors;
  • New algorithm to determine optimisation regions based on predicted TC tracks from previous EPS run;
  • The Caribbean remains an optimisation region if no TC in the vicinity;
  • Ortho-normalisation applied to set of all tropical SVs.

Verification results gathered to date from the experimental suite and from several months of research experiments in winter and summer indicate that, in general, the impact of the new cycle is neutral to slightly positive in the extra-tropical troposphere in the deterministic forecasting system for the total sample of experiments. Wind and temperature scores in the tropical troposphere show small improvements, but stratospheric temperature biases are slightly larger. This issue is being addressed.

There are, however, noticeable changes in the EPS. The revision of initial condition perturbations for TCs, including the use of tropical cyclone tracks from the previous EPS run, improves the TC track forecasting giving a better spread and skill in the strike probability. The revised numerics in the convection scheme and calling of the cloud scheme has reduced the performance differences between the T511 forecast and the T255 control. The previously experienced tendency in the EPS to develop spuriously deep lows has been reduced, providing better synoptic guidance with a smaller spread without affecting the overall skill of the EPS. Users also need to be aware of the changes in the generation of initial EPS perturbations. With the introduction of the Gaussian sampling the one to one positive and negative perturbations will no longer be present.

The change to the new cycle should be transparent for users apart from a change in Model identification number for the atmospheric model in GRIB headers, which will be set to 124 (Note: after reaching the end of the defined range of possible identifications numbers in cycle 26r1 with 203 it was agreed that the numbers will be recycled). The wave model identification numbers will not change and will continue to be 115 for the Global Wave Model and 215 for the Mediterranean Wave Model.

There will be only minor technical changes in cycle 28r3. The main differences will be:

  • archive and disseminate step 0 for the deterministic forecast (as this is used in the generation of weather parameter products)
  • omit production of step 0 for BC project forecasts (was available in FDB only)
  • omit production of step 3 and 9 for eps control forecasts (to overcome a inconsistency between the control forecast and the perturbed forecasts and allow comparison of cumulative and mean fields)
  • archive tropical cyclone tracks in BUFR format in MARS