Please note that these changes only apply to the ECMWF Operational datasets (not the Re-analysis data).
The information provided here is also available from the ECMWF evolution of the forecasting system page.
From Summer 2003 onwards we have stopped adding to this page as all the information is available from the ECMWF web pages.
Several meteorological changes were implemented at the same time:
These modifications addressed certain problems which were noted in the Tropics over the last few months, in particular with precipitation and 2m temperature. No significant meteorological impact was expected over Europe.
Planned Changes
The 3-d variational analysis is planned for implementation in the 2nd half of 94.
It resulted in a noticeable impact on the temperature in the daytime boundary layer over continental areas. There was a reduction in the warm bias by several degrees in places, and the positive impact was also to be seen at 850 and 700 hPa. There was also a significant reduction in the bias of the dewpoint temperature at 2 metres.
Planned Changes
A change to the pre-selection of cloud motion wind (SATOB) data for the analysis will be implemented shortly.
On 12 July two data types were introduced into the data assimilation: temperature observations from aircraft, and winds derived from the METEOSAT water vapour channel.
The post-processing of 10m winds over land was changed on 23 August, resulting in a more realistic representation of model winds at observing sites. The negative bias of the wind speed has been significantly reduced, in particular during daytime.
Planned Changes
Pre-operational trials are in progress on a 3-d variational analysis system.
Changes to the model physics are under test. They include a new cloud scheme with prognostic equations for liquid water vapour content, use of mean orography and new subgrid orography parametrisation.
A revision to the definition of the tropopause for the humidity analysis was introduced on 28 March 1995.
A new version of the forecast model was implemented on 4 April, including:
Specific humidity on model levels and pressure levels was changed from spherical harmonic to quasi-regular Gaussian grid field. In addition, three new quasi-regular Gaussian grid fields on model levels were introduced: cloud liquid water content, cloud ice water content and cloud cover. Fields of low, medium, high and total cloud cover continue to be produced, but convective cloud cover is no longer produced.
This set of changes leads to a much improved representation of cloud cover, with large reductions from the previous systematic negative biases, especially during the day. A significant reduction in summertime warm bias of two-metre temperature is expected. The change should also improve the precipitation forecast, with smoother precipitation patterns and reduced spin-up problems in the early range of the forecast.
An improvement of the RMS and anomaly correlation scores in the medium-range over Europe, reflected in the overall synoptic pattern, was noted during the experimentation, particularly in Summer.
In addition, the parametrisation of the ocean surface albedo has been changed to better represent the variation of reflectivity with solar zenith angle.
Planned Changes
A 3-D variational analysis system and high resolution wave prediction model for the Mediterranean and Baltic Sea will be implemented.
Planned Changes
A 3-d variational analysis system will be implemented.
This issue features an article "A major operational forecast model change" which I can photocopy and send. Topics include discussion of a semi-Lagrangian treatment of advection, change to humidity fields (no longer transformed to spectral space), a prognostic cloud scheme and orographic effects.
On 28 November 1995, a revised form of the continuity equation was implemented in the forecast model, resulting in a reduction of noise in all lower tropospheric fields previously experienced near mountains. In particular, the fields of mean sea level pressure, geopotential height and temperature are smoother, and the near surface wind is better represented.
On 30 January 1996 ECMWF introduced a 3-dimensional variation (3D-Var) analysis scheme. 3D-Var is a new code for the analysis of model-level values of temperature, vorticity, divergence and specific humidity, and surface pressure. Minor changes to the forecast model were implemented at the same time.
On average, forecasts from 3D-Var for the Northern Hemisphere are of similar quality as forecasts from the previous Optimum Interpolation system, while forecasts for the Southern hemisphere tend to exhibit higher skill. In addition, 3D-Var gives generally better temperature verification results, especially at low levels and in the stratosphere, and better wind scores at 200 hPa and above.
The analysis and prediction of tropical cyclones appear to have improved with 3D-Var.
On 4 March 1996 the generation of initial perturbations for the Ensemble Prediction System (EPS) was changed to be based on singular vectors in the southern hemisphere as well as in the northern hemisphere.
On 25 March 1996 the amplitude of the initial perturbations for the EPS was increased by approximately 15%.
On 23 April 1996 a new definition of the sea surface temperature and sea ice was introduced. The input to the scheme is the 1 degree SST analysis from NCEP Washington and the gridded ice fields derived from SMM/I data from NESDIS.
Planned Changes
A comprehensive observation screening and quality control will be implemented in the 3D-Var assimilation system.
A technical error affecting the computation of sub-grid orography processes in the forecast model which was introduced with the change to sea surface temperature and sea ice on 23 April 1996 was corrected on 31 May.
The complete operational suite was implemented on the Fujitsu VPP700 on 18 September.
Model cycle CY15R5, introduced on 18 September, had the following changes:
Quality control and the calculation of background error variances are now based on the variational analysis rather than the earlier 01 approach. Objective verification indicates little overall sensitivity of forecast performance to these changes.
A high resolution version of the optional project global wave forecast model was introduced on 5 December 1996. The new model has a resolution of approximately 50km using a quasi-regular latitude/longitude grid with 0.5 degree intervals between latitude rows and varying numbers of points along latitudes.
Model cycle 15R7 was introduced on 10 December 1996 in both the T213 L31 10-day forecast and the EPS. The main features of the new cycle are:
On the same date, the EPS was enhanced to run at TL159 resolution (linear grid) with 50 ensemble members and the control. The model physics are computed on the quasi-regular Gaussian N80 grid. Model cycle 15R8 was introduced on 21 January 1997. This cycle corrects an error affecting the degree of balance between the mass and wind analyses.
A new version of the optional project wave models was implemented on the 13 May 1997. It consisted of a change in the representation of the wave spectrum to solve the excessive shadowing effect behind islands that was experienced before. This allowed the re-introduction of islands that were deliberately omitted in the previous version.
Several modifications to the 3D-var data assimilation system were implemented on the 14 May 1997 (model cycle 16r2):
Planned Changes
Changes in the use of satellite data and a revision in the computation of water vapour saturation pressure in the analysis and forecast will be implemented soon.
A number of modifications were made on 27 August 1997, mostly concerning the use of satellite date (model cycle 16r3):
On 25 November 1997, the first version of a four-dimensional variational data assimilation system (4D-Var) was introduced. It is based on an evolution of the previous 3D-Var system with a 6-hour cycling. Details will be given in a forthcoming newsletter article.
A number of changes to the physical parametrisation scheme were introduced on 16 December 1997 (model cycle 18r3).
Planned Changes
A number of changes to the physical parametrisation scheme were introduced on 16 December 1997 (model cycle 18r3):
The method used for the computation of initial perturbations of the Ensemble Prediction System (EPS) has been changed on 24 March 1998. From this date, initial perturbations are a combination for day D of fastest growing modes between D-2 and D (‘Evolved Singular Vectors’) and fastest growing modes between D and D+2; the meteorological impact of this change is minor (slightly increased spread in the early medium range).
The reference model spectral resolution was increased from T213 to TL319 on 1 April 1998 (improved orography, reduced diffusion, but Gaussian grid unchanged). The other modifications introduced on this occasion (Cy18r5) were:
An hourly, two-way coupling of the atmospheric and ocean-wave model was introduced on 29 June 1998. Predicted ocean waves now provide information to the atmospheric boundary layer.
Other modifications introduced at the same time (Cy18r6) were:
Planned Changes
Increase in the number of model levels from 31 to 50, with the majority of the extra levels occurring in the stratosphere, the top of the model will be moved from 10 to 0.1 hPa;
Use of TOVS and ATOVS level Ib radiance data from the NOAA satellites.
METEOSAT5/INDOEX SATOB products have been introduced in the data assimilation with effect from 12 August 1998. A formulation of uncertainties associated to physical processes (stochastic physics) was introduced on 21 October 1998 in the ECMWF Ensemble Prediction System.
Planned Changes
A version of the reference model with extended vertical domain and possibly increased planetary boundary layer resolution should also be tested soon.
On 9 March 1999, the reference model’s vertical resolution was increased from 31 to 50 levels (model cycle version 19r2) (a description of the changes was published in ECMWF Newsletter Number 82, ‘Increased stratospheric resolution in the ECMWF forecasting model’ by A.Untch et al.).
On 5 May 1999, processing and assimilation of raw TOVS/ATOVS radiance data from the NOAA-14/ NOAA-15 polar orbiting spacecraft became operational (model cycle version 21r1) (see article in this issue on page 2). The move to assimilating raw-radiance data (as opposed to the NESDIS pre-processed radiances that were previously used) has required a significant revision of the errors assigned to radiance observations in the analysis, and of the quality-control procedures; other modifications introduced at the same time were mainly technical changes (re-coding of the sea and lakes temperature prescription software, and modifications of the oceanic waves code). The meteorological impact is a moderate, but consistent, improvement in forecast skill at all ranges, the largest being observed in the southern hemisphere and lower stratosphere.
Planned Changes
A new soil moisture and temperature analysis scheme is expected to become operational this summer. An increase from 50 to 60 levels (more resolution in the planetary boundary layer) is planned for the autumn.
Planned Changes
The next change (Cy21r4) is planned for 12 October; it includes an increase in the number of vertical levels to 60 (most of the extra resolution being near the surface), with modifications to the cloud and convection scheme, and revised mean and subgrid–scale orography fields.
On 22 October 1999 the reference model’s vertical resolution was increased from 50 to 60 levels, most of the extra resolution being in the planetary boundary layer. The EPS vertical resolution has been increased in a consistent way from 30 to 40 levels.
These resolution changes have been coupled with several modifications to the atmospheric model (Cycle 21r4):
New global orography and subgrid orography fields have been introduced, based on a new high resolution dataset. The changes are mostly isolated ones except for the Greenland massif and Australia, where more substantial changes in the basic orography are evident.
The changes have been proved to have had a positive impact of the error scores (Z500 in the extratropics, the upper-level winds and temperature in the tropics). The frequency of occurrence of moderated to strong 10m winds and precipitation were also improved.
Planned Changes
The next upgrade in 2000 will be related to the data assimilation system, i.e. 12-hour cycling of 4D-Var, possibly with increased resolution of the minimised increments.
On 11 April 2000, a new version of the operational model was implemented (cycle 22r1). This is a minor scientific upgrade that includes:
A technical change has also been brought in through the use of new post-processing software (Full-Pos) developed in collaboration with Meteo-France. This change will allow the introduction of new parameters in the dissemination and MARS archive, details of which will be announced later.
This new version of the model is running on the new Fujitsu VPP5000 computer.
Other change included in this model version are:
In addition, enhancements have been made to the post-processing partly related to the new surface scheme and partly developed for application within the ERA-40 project. A major technical change has been made to the 4D-Var data handling through the implementation of purpose-built ‘ODB’ database software to deal with observations, replacing the old CMAFOC file format.
CY22r3 gives an overall improvement of upper-air scores. The cold bias in the screen temperature over spring snow conditions, in particular over Fenno-Scandia, has been removed but a moderate warm bias has been noticed in places.
On 21 November, the model resolution was upgraded from TL319 to TL511 resolution in the deterministic mode, and from TL159 to TL255 in the ensemble mode (EPS). This is roughly a reduction in grid size from 60 to 40 km (deterministic) and from 120 to 80 km (EPS). Vertical resolution remains unchanged in all model configurations. Other changes included in this model version are:
The T511 and T255 pre-operational suites have demonstrated their positive impact during the testing period on the mean scores for upper-level fields and precipitation, both in deterministic and probabalistic mode. Meteorological evaluation on individual cases has also shown an overall improvement.
The computation of the EPS stochastic physics tendencies was corrected on 15 December 2000. This removed a feature introduced with Cycle 21r2 (July 1999) which has occasionally produced non-realistic values for some EPS members (notably for near-surface temperature and humidity).
Due to the failure of the ERS-2 navigation instruments, both the ambiguous surface winds and the altimeter data became unavailable on 17 January 2001.
The amplitude of the EPS initial perturbations was upscaled by a factor of two on 5 February 2001. This rescaling was linked to the data assimilation changes introduced with cycle 22r3 in June 2000 which resulted in a lack of spread.
The ERS-2 altimeter data were switched on again in the oceanic-waves data assimilation on 6 March 2001. The ERS-2 winds remain unavailable.
Future Changes
Testing of cycle 23r4 is going on. This new cycle allows more and better uses of satellite data over land and sea ice. It also contains a minor change to the horizontal diffusion and provides more frequent calls to the radiation code in the data assimilation.
Verification has shown a noticeable reduction in cold biases of the 2m temperature around sunset; this is linked to the improved time integration of land-surface temperature. The scores for the free atmosphere show a very minor, and overall neutral, impact of the changes on the performance during the three months of testing currently available.
Future Changes
Pre-operational testing will start over the summer for an important set of changes both to the data assimilation (T255 inner loops, new Jb statistics, new humidity structure functions, the use of water-vapour radiances from Meteosat and of sea winds from QuikSCAT, an upgrade of the radiative-transfer code, upgraded quality control for mobile platforms, the use of extra channels from AMSU-A for an extended application of the data over land), and to the forecast model (finite elements in the vertical, a revised precipitation scheme, improvements in the oceanic wave model).
Future Changes
Pre-operational testing of an important set of changes for the operational system has started in October. All components of the system are affected:
More on the impact of these changes on the model performance will be reported in the next newsletter.
On 22 January 2002, an upgraded version of the model, CY24r3, was implemented. This version includes several important changes that together affect all components of the system (data assimilation, atmospheric and oceanic wave forecasts, EPS):
Following thorough testing and evaluation of the new model cycle, verification has shown a robust improvement of upper-air scores for all domains and forecast ranges, most notably over the southern hemisphere and at higher levels. Preliminary verification of tropical-cyclone tracks in September and October has also shown significant improvements, both in deterministic and probabilistic modes. Case studies have demonstrated a reduction in the number of occasionally spurious developments of small-scale cyclones over sea.
On 4 March 2002, a new blacklisting procedure was introduced to avoid using AMV (ex-SATOB) winds during time-slots likely to be affected by solar-eclipse problems.
On Mach 5 2002, a bug fix was introduced into the post-processing of gust winds in the EPS; prior to that date, the maximum was over the last 6 hours, not 12 as it should have been.
On 9 April, an upgraded version of the model, CY25r1, was implemented. This version includes a revised short-wave radiation scheme with variable effective radius of liquid cloud water, retuning of the land-surface parametrization (TESSEL) to reduce winter/spring warm biases in low-level temperatures, improved wind-gust post-processing, and the activation of new data streams in the assimilation (water-vapour radiances from Meteosat-7, SBUV and GOME ozone data, European wind profilers). A bug in the convective momentum transfer was also fixed. Verification has shown a small improvement in the upper-air scores, most notably at high levels and over warm continental land masses.
A new cycle of ECMWF model, Cy25r4, has been implemented on 14 January 2003.This version includes changes to many aspects of the data assimilation and forecasting system:
In addition the RNORM parameter scaling EPS initial perturbations was reduced from 2.0 to 1.6 as a consequence of the data assimilation statistics retuning.This has the effect of keeping the EPS spread at the same level as in the current operational version.
The impact of these changes has been found to be meteorologically positive during the test period over a wide range of areas and parameters.Tests conducted during the summer have shown a clear reduction of forecast errors generated over North America and advected towards Europe at this time of the year. The new model version removes convective instability more effectively and realistically, as was found in several case studies. Stratosphere forecasts have also been improved.
This version is being run in parallel on the Fujitsu VPP5000 and IBM HPCF, and is planned to be the one to be migrated to the new computer later this year. Return to top of page
AIRS radiances have been passively monitored since 19 February 2003.
On 4 March 2003 all the operational forecast suites were switched to run on the new IBM high-performance computing system using cycle CY25r5; this major technical upgrade went smoothly, both from the internal and external users? points of view. See also, the articles on pages 12 and 20 in this issue of the Newsletter, the article on page 11 of Newsletter 93 (Spring 2002) and:
http://www.ecmwf.int/services/computing/overview/ibm_cluster.html.
Since 25 March 2003, the ECMWF medium-range operational forecasts (T511L60 deterministic and T255L40 EPS) have been run twice per day from base times 00 and 12 UTC. However, the data-assimilation cycle has remained unchanged. The 00 UTC products are archived in MARS and are available for dissemination
On 31 March 2003, the Fujitsu computers were decommissioned. As a result (and according to plans) the old version of the seasonal-forecast suite (System 1) has also been discontinued.