The CMA global ensemble prediction system was upgraded on 26 December, 2018. The first updated CMA outputs for TIGGE archive have been ingested only since 1 June 2020.

The main changes in this upgrade are:

• The forecast model system is changed from TL639 L60 to GRAPES_GFS, correspondingly, the horizontal resolution is changed to 50km, and the model top is changed to 3hPa.
• The initial perturbation method is changed from Bred Vector to Singular Vector.
• The SKEB is introduced in the model perturbations.
• The ensemble size is changed from 15 to 31.

Full details are available in the page Models.

2024 Jan 24 - Extended coupling between ICON model parameters and data assimilation building upon the 10M-wind assimilation

2023 Mar 15 - icon-2.6.5-nwp1

• Extension of model-DA coupling for optimizing T2M and RH2M scores
• Modification of Charnock parameterization for ocean surface roughness
  Sep 6 - icon-2.6.6-nwp0

2022 May 11 - version ICON-2.6.4-nwp4

• Extended coupling between ICON model parameters and data assimilation building upon the T2M assimilation

2022 Jul 4 - version ICON-2.6.4-nwp5

• Improvement of model-DA coupling for surface evaporation

2022 Nov 23 - resolution upgrade ICON-EPS (ICON deterministic unchanged)

• horizontal resolution: global / 40km to 26km, EU / 20 km to 13 km EU domain.
• number of model levels: global / 90 => 120, EU / 60 to 74

The horizontal resolution of the ECCC contribution to TIGGE has been increased from 1.0 to 0.25 degree since 12Z 14 Apr 2021 Also the GRIB 2 packing has been changed to simple instead of complex one (that will generally  increase file sizes but also  significantly reduce time needed to process such data).

The ECCC global ensemble prediction system (origin cwao in the TIGGE archive) was upgraded on July 3 2019.

The main changes in this upgrade are:

• The horizontal resolution of the Global Deterministic Prediction System (SGPD) is reduced from 25km to 15km, and the number of vertical levels is increased from 80 to 84 levels.
• The atmospheric component of the Global Ensemble Prediction System (GEPS) is now coupled with the ocean-ice model.
• The horizontal resolution of the Regional Ensemble Prediction System (REPS) increases from 15km to 10km, with an increase in the output frequency from 2 to 4 runs per day.

Full details of the changes made to the various systems are available in this official page.

The ECCC global ensemble prediction system (origin cwao in the TIGGE archive) was upgraded to version 5.0.0 on 18 September 2018.

The main changes in this upgrade are:

• The 800x400 global Gaussian grid at 50 km resolution is updated to a Yin-Yang grid at 39 km resolution
•   The model top is raised from 2 hPa to 0.1 hPa.
•  In the ensemble Kalman filter assimilation component, the digital filter initialization is replaced by an incremental analysis update (IAU) procedure.
  

Further documentation of the operational upgrade can be found at  http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/changes_e.html

The ECCC global ensemble prediction system was upgraded to version 2.0.2 on 17 August 2011.

The main changes included in this upgrade are:

• The number of members in the ensemble Kalman filter is doubled from 96 to 192.
• The resolution of the medium-range forecasts is changed from 100 to 66 km.
• The model top is raised from 10 hPa to 2 hPa.
• A new dynamical model, with a Charney-Phillips vertical grid, is used.

Further documentation of the operational upgrade can be found at http://collaboration.cmc.ec.gc.ca/cmc/CMOI/product_guide/docs/changes_e.html

The ECMWF forecast system IFS was upgraded to the version 48r1.

With IFS Cycle 48r1, the horizontal resolution of the  medium-range ensemble  (ENS) will increase from 18 to 9 km. This will bring the ENS to the same horizontal resolution as the high resolution forecast (HRES). The vertical resolution for both ENS and HRES will remain at 137 model levels and the ENS will continue to have 51 members.

For TIGGE, the IFS data is interpolated to the same resolution O640 as up to now.

More details can be found in Implementation of IFS Cycle 48r1

The ECMWF forecast system IFS was upgraded to the version 47r3.

Cycle upgrade 47r3 will bring improvements to the assimilation and observations usage and a significantly improved physical basis for moist processes, necessary to facilitate further development of the Integrated Forecasting System (IFS) and future application at convection-permitting resolutions.

More details can be found in Implementation of IFS Cycle 47r3

The ECMWF forecast system IFS was upgraded to the version 47r1.

This cycle includes changes in the treatment of observations and improvements in the data assimilation and to the model. Quintic vertical interpolation in the semi-Lagrangian advection scheme has been introduced as well as the inclusion of a better surface albedo climatology making use of more data from the MODIS instrument.

New Metrics of Tropical Cyclone (TC) “size” will supplement the existing forecasts of TC track and intensity.

Full details can be found in Implementation of IFS Cycle 47r1 page.

The ECMWF forecast system IFS was upgraded to the version 45r1.

The main changes included in this upgrade:

Assimilation

• Weakly coupled sea-ice atmosphere assimilation applied with the use of OCEAN5 sea-ice (instead of OSTIA) in the surface analysis of the high-resolution (HRES 4d-Var) and the ensemble of data assimilations (EDA) analyses;
• Relative humidity increments calculated using temperature instead of virtual temperature;
• Weak constraint model error forcing applied at every time step instead of every hour to avoid shocks in the model integration.

Observations

• Assimilation of non-surface-sensitive infra-red (IR) channels over land;
• Assimilation of all sky micro-wave (MW) sounding channels over coasts;
• Use of direct broadcast FY-3C MWHS2 data for better timeliness;
• Introduction of RTTOV-12 and new microwave instrument coefficients;
• Activation of constrained variational bias correction (VarBC);

• Retuning of the radiosonde observation error, and introduction of a scheme to account for radiosonde drift;

• Introduction of temperature bias correction of old-style AIREP observations; aircraft temperature varBC predictor upgraded to a three predictor model (cruise, ascent, descent); reduced thinning of aircraft data;

• Assimilation of JASON-3 and Sentinel-3A altimeters, and use of new altimeters for wave data assimilation;

Model

• Coupling of the 3-dimensional ocean and atmosphere: introduction of the coupling to the NEMO 3-dimensional ocean model also in the high-resolution forecast (HRES), with the same ocean model version used in the medium-range/monthly ensemble (ENS): NEMO3.4 in ORCA025_Z75 configuration; upgrade of the NEMO-IFS coupling strategy in both ENS and HRES to a full-coupling in the tropical region (partial-coupling-extra-tropics);
• Improved numerics for warm-rain cloud microphysics and vertical extrapolation for semi-lagrangian trajectory;
• Increased methane oxidation rate to improve (increase) water vapour in the stratosphere;
• Improved representation of super-cooled liquid water in convection, and minor convection updates;
• Improvements in the tangent forward and adjoint models linked to the convection scheme;
• Correction of soil thermal conductivity formulation and addition of soil ice dependency;
• New extended output parameters have been added. See below.
• Modified parameter for non-orographic gravity-wave drag scheme for 91 levels;
• Model error changes:
  • Stochastically perturbed parametrization tendency scheme (SPPT): improved flow-dependent error representation via reduced spread in clear skies regions (due to unperturbed radiative-tendency in clear sky), activation of tendency perturbations in stratosphere, and weaker tapering of perturbations in boundary layer; amplitude reduction of the SPPT perturbations patterns (by 20%);
  • EDA: cycling of stochastic physics random fields in the EDA, and adoption of the same SPPT configuration in EDA as in ENS;
  • Stochastic kinetic energy backscatter scheme (SKEB): deactivation of the stochastic backscatter (SKEB) scheme due to improved model error representation by the SPPT scheme (see above), leading to a 2.5% cost saving in the ENS;

Click  here for full details.

The ECMWF forecast system IFS was upgraded to the version 43r3.

The main changes included in this upgrade:

Assimilation

• • Improved humidity background error variances directly from the EDA like for all other variables.
  • Revised wavelet filtering of background error variances and revised quality control of drop-sonde wind observations in 4DVAR to improve tropical cyclone structures.

Observations

• • Increased use of microwave humidity sounding data by adding new sensors (SAPHIR, GMI 183 GHz channels).
  • Activation of 118 GHz channels over land from MWHS-2 instrument on-board FY-3C.
  • Harmonised data usage over land and sea-ice for microwave sounders (adding MHS channel 4 over snow, adding some ATMS channels, lower observation errors for MHS data over land).
  • Improved screening of infrared observations for anomalously high atmospheric concentrations of hydrogen cyanide (HCN) from wildfires.
  • Improved quality control for radio occultation observations and radiosonde data.

Model

• • New, more efficient radiation scheme with reduced noise and more accurate longwave radiation transfer calculation.
  • New aerosol climatology based on ‘tuned’ CAMS aerosol re-analysis including dependence on relative humidity.
  • Increased super-cooled liquid water at colder temperatures (down to -38C) from the convection scheme.
  • Visibility calculation changed to use ‘tuned’ CAMS aerosol climatology.

Click  here for full details.

The ECMWF forecast system IFS was upgraded to the version 43r1.

A major upgrade is in the dynamical ocean model (NEMO):

• Resolution increase from 1 degree and 42 layers to 0.25 degrees and 75 layers
• Implementation of interactive sea-ice model (LIM2)
• Ocean and sea-ice components of the ENS initial conditions provided by new ocean analysis and reanalysis suite ORAS5

Click  here for full details.

The ECMWF forecast system IFS was upgraded to the version 41r2.

The main changes included in this upgrade:

• Introduction of a new form of the reduced Gaussian grid, the octahedral grid, for HRES, ENS and ENS Extended;
• Horizontal resolution of the HRES increased from T_L1279 / N640 to T_CO1279 / O1280, where subscript C stands for cubic and O for octahedral;
• Horizontal resolution of the ENS increased from T_L639 / N320 to T_CO639 / O640 for ENS (Days 0 - 15) and from T_L319 / N160 to T_CO319 / O320  for ENS Extended (Days 16 - 46);
• For the medium-range ENS there will no longer be a decrease of resolution at day 10: the ENS Days 11 - 15 will be run at the same T_CO639 / O640 resolution as ENS Days 0 - 10;

Click  here for full details.

The ECMWF model was upgraded to cycle 36r1 on 26 January 2010.

The main changes included in this cycle are:

• Deterministic forecast and analysis horizontal resolution is increased from T799 to T1279, i.e. from 25 km to 16 km.
• EPS resolutions are increased from T399/T255 to T639/T319 for Leg A/B respectively. That is to say, the first 10 days of the EPS (Leg A) will run at T639 (32 km) and the extension beyond day 10 (Leg B) will run at T319 (63 km).
• Correction of short-wave radiation interaction with clouds.

The ECMWF model was upgraded to cycle 33r2 on 30th September 2008 12Z. The main changes included in this cycle are:

• The OSTIA (Operational Sea Surface Temperature and Sea Ice Analysis) high-resolution sea surface temperature, produced by the Met Office, and corresponding sea ice analysis (from EUMETSAT Ocea and Sea Ice SAF)
• Conserving interpolation scheme for trajectory fields in 4D-Var
• New variational bias correction (VARBC) bias predictors to allow the correction of infrared shortwave channels affected by solar effects
• Cleaner cold-start of AMSUA channel 14 bias corrections
• Changes to physics for melting of falling snow, albedo of permanent snow cover (e.g. over Antartica), diurnal variation of sea surface temperature, and linear parametrization schemes
• Convective contribution added to wind gusts in post-processing
• Monitoring of MERIS total-column water vapour data

The ECMWF model was upgraded to cycle 33r1 on 3rd June 2008 12Z. The main changes included in this cycle are:

• Improved moist physics in tangent linear/ adjoint model used in 4D- Var assimilation.
• Re-tuned entrainment in convection scheme.
• Bug fix to scaling of freezing term in convection scheme.
• Additional shear term in diffusion coefficient of vertical diffusion.
• Increased turbulent orographic form drag.
• Fix for soil temperature analysis in areas with 100% snow cover.
• Change in surface roughness for momentum, and change in post-processing of two-metre temperature and specific humidity.
• Assimilation of AMSR-E and TMI radiances in 1D+4D-Var; assimilation OMI ozone data.
• Usage of all four wind solutions for QuikSCAT in assimilation, rather than only two previously.
• Extended coverage and increased resolution for the limited area wave model.
• Improved shallow water physics and modified advection scheme for ocean wave models.
• Introduction of two new wave model parameters: maximum wave height and corresponding wave period.

The ECMWF model was upgraded to cycle 32r3 on 11th March 2008 12Z. The main changes included in this cycle are:

• Integration of the Monthly Forecasting System with the medium-range Ensemble Prediction System (EPS).
• Use of persisted SST anomalies in all atmospheric forecasts
• Daily ocean-coupling of days 10 to 15 of 0000 UTC EPS forecasts
• Monthly Forecast run once per week from 0000 UTC on a Thursday as an extension of the 15 day EPS forecast from this base time
• Modified EFI products using the new unified re-forecasts
• New GRIB description for all Monthly Forecast products, analogous to the current medium-range EPS data

The ECMWF model was upgraded to cycle 32r3 on 6th November 2007 12Z.

The main changes included in this cycle are:

• New formulation of convective entrainment and relaxation timescale
• Reduction in free atmosphere vertical diffusion
• New soil hydrology scheme
• New radiosonde temperature and humidity bias correction
• Increase in number of radio occultation data from COSMIC
• Assimilation of AMSR-E, TMI, SSMIS window channels (clear sky)
• Assimilation of SBUV (NOAA-17, NOAA-18) and monitoring of OMI ozone data

The main changes to the EPS included in this cycle are:

• initial perturbation amplitude reduced by 30%
• singular vectors targeted on tropical cyclones are computed with the new moist physics package in the tangent-linear and adjoint models (as used in the operational in 4D-Var since cycle 32r2).

The ECMWF model was upgraded to cycle 32r2 on 5th June 2007 12Z

The main changes included in this cycle are:

• Three-minimization version of 4D- Var assimilation scheme (T95/T159/T255) with improved moist linear physics (cloud and convection).
• Improved parametrization of the heterogeneous ozone chemistry.
• New short-wave radiation scheme (RRTM-SW), plus McICA cloud-radiation interaction and MODIS albedo.
• Retuned ice particle size.
• Revised subgrid-orography scheme.
• Explicit numerical treatment of convection in the moist tangent linear model used in the calculation of tropical singular vectors.

ECMWF upgraded its Ensemble Prediction System (EPS) and introduced the Variable Resolution Ensemble Prediction System (VarEPS). In particular, the forecast range was extended to 15 days using the VarEPS system with a resolution of T399 L62 for day 1 to day 10 (leg 1) and T255 L62 for T+246 to day 15 (leg 2).

The JMA updated their Global EPS on March 30 2021.

The major changes are:

• Increase of the number of vertical layers from 100 to 128 for GSM and GEPS
• Increase of the number of ensemble members in GEPS from 27 to 51
• Revision of the global snow analysis
• Introduction of the global soil moisture analysis

The JMA updated their Global EPS (previously named One-week EPS) at 00UTC of 19th  January 2017.

The major changes are:

• The vertical resolution of the Global EPS model is increased from
• TL479L60 to TL479L100
• Perturbations from LETKF are introduced for initial perturbations.
• Perturbations to sea surface temperature are introduced.
• The name of the EPS is renamed from One-week EPS to Global EPS

The resolution of one-week ensemble prediction model of JMA was upgraded on 21 November 2007.

Major changes are:

• Increase in the resolution from TL159L40 to TL319L60 with a topmost level raised from 0.4hPa to 0.1hPa.
• Use of a new high-resolution analysis of sea surface temperature and sea ice concentration as ocean surface boundary conditions.
• Use of surface snow depth data from the domestic dense observational network in the global snow depth analysis.
• Introduction of a convective triggering scheme into the deep convection parameterization.
• Introduction of a new 2-dimensional aerosol climatology derived from satellite observations for the radiation calculation.
• Increase in the resolution of inner loop model of the four-dimensional variational (4D-Var) data assimilation system from T106L40 to T159L60.
• Introduction of a sigular vector method to make the initial perturbations.

Notes:

• The resolution and parameters of TIGGE data from JMA remains unchanged.
• Orography and Land Sea Mask in TIGGE data are slightly different from the previous data associated with the model change.

The orography was added to KMA control forecast outputs (for step 0 only).

https://jira.ecmwf.int/servicedesk/customer/portal/4/SD-84207

The KMA has changed their model used for TIGGE contribution from UM (Unified Model) ensemble to  KIM (Korean Integrated Model) ensemble, v. 3.7, since the 1^st of  July 2022.

KMA has been operating 2 ensemble models based on UM and KIM. The latter one, KIM, was developed during 9 year project from 2011 to 2019 and became an operational model in 2020.

The Met Office Global model was upgraded to OS45 with effect from the 12UTC cycle on 5 May 2022.

• The Global Atmosphere-Land Model is now fully coupled to a ¼ degree interactive Ocean Model in both 10km deterministic and 20km MOGREPS-G ensemble configurations, with both the atmospheric and ocean models exchanging information hourly.
• The atmosphere-land model has been upgraded from GA7.2.1GL8 to GA8GL9.
• New observations have been introduced from radiosonde descents, Sentinel, Aeolus and Mode-S in Europe.

UM upgraded to version PS43 for the 12UTC cycle on 4 December 2019.

The PS43 upgrade to the Met Office global modelling capability introduces a new data assimilation system to the global ensemble and a significant upgrade to the global model physics. MOGREPS-G is now driven by an ensemble of analyses, each with its own data assimilation cycle (En-4D-en-VAR), replacing the original ETKF (Ensemble Transform Kalman Filter) method. The result is a more sophisticated ensemble with increased spread and stability.

The global model physics is upgraded to the GA7.2 configuration which is focused on improvements to physical processes. This has several significant benefits including improvements to 250hPa winds, 850hPa tropical winds and temperatures and improved tracks of Tropical Cyclones. The UM is upgraded to UM11.2.

UM upgraded to version PS39 from 12UTC cycle on 11 July 2017.

TIGGE global ensemble (the Met Office MOGREPS-G) was upgraded to 20km (from 33km), N640L70, 280x960 lat-long grid, 0.1875 N-S x 0.28125 E-W degrees;  the MOGREPS-G will continue to run four times a day (00, 06, 12 and 18UTC), but the number of ensemble members per run has increased from 12 to 18, with the last two runs being combined to create a 36-member time-lagged ensemble.

Forecast length remains at 7d 6h.

Relevant associated deterministic/data assimilation changes:

• Replace RTTOV9 by RTTOV11
• New ground GPS observation operator
• Improved snow analysis reconfiguration
• Upgrade to UM10.6

UM upgraded to version PS38 from 6UTC cycle on 8 November 2016.

The changes include predominantly a satellite applications package to include more satellite channels, re-tuned observation error estimates and better quality control of observations. This is expected to improve the headline NWP index scores by about +0.5 to +1.0 against analyses, observations, and independent ECMWF analyses.

Satellite package includes:

• Improved use of infrared and microwave satellite data over land, providing more temperature information at lower levels.
• Improved global humidity forecasts, due to more weight being given to satellite humidity information in our analysis.
• Incorporation of more Chinese polar orbiting satellite data, following our initial introduction in March 2016.
• The introduction of data from a new Japanese Geostationary satellite, Himawari-8.

UM upgraded to version 7.4:

• Horizontal resolution upgraded to N216 (60km) from N144 (90km).
• Vertical resolution upgraded to 70 levels from 38 levels.
• SKEB2 upgraded with N216 tuning and velocity potential wind increments
• Bug fixed in wind gust diagnostic.
• CAPE and CIN formulations changed to bring into line with those of other Centres.
• Perturbations now formulated using vertical localisation to improve spread near the surface.
• Implementation was for the 12UTC cycle on 9 March 2010.

Implementation of E-suite 4. The model was upgraded to UM7.3 (released 26/3/09). The upgrade from UM6.3 represents 2 major and 5 minor releases of the model and brings the UK 15-day ensemble into line with its parent system and deterministic forecast system. There are many changes to the science schemes. RMS errors are improved in all areas with particular benefit noted in the northern hemisphere in the forecast range 6-12 days. The stochastic physics schemes SCV and SKEB1 were also replaced by the new SKEB2 scheme. This results in increased ensemble spread at all lead times (e.g. up to 10% in PMSL spread at T+10 days).

Implementation of E-suite 3. The model was upgraded to UM6.3 (released as a portable version 23/7/08). The upgrade from UM6.1 was primarily the implementation of the new configuration management system, although code restructuring in several schemes such as radiation and convection resulted in small improvements to RMS errors particularly in the tropics.

The UKMO TIGGE suite (MOTHS) was upgraded on 27 November 2007 00Z. The model version and stochastic physics schemes were not upgraded. UM Science Pack was upgraded from cycle 39 to cycle 44 (representing the operational deterministic model active as of May 2007 and the current parent 3-day ensemble forecast system).

The main features are:

• Change to the convective cloud which allows the cloud to decay away more slowly and more likely to influence the radiation scheme.
• Advanced adaptive detrainment for mid-level convection.
• Bug fix to freezing level calculation in convection.
• Bug fix to MSLP calculation over orography.
• Addition of biogenic aerosol climatology.
• Two corrections in the representation of convective clouds in the radiation scheme.
• Using MODIS observations of the surface albedo.
• Change to soil runoff formulation.
• Seasonally varying Leaf Area Index.
• SPARC ozone climatology instead of Li and Shine.
• A change from fixed SSTs to SSTs which evolve based on the climatology but with the SST anomaly at the start of the run persisted.

The UKMO TIGGE suite (MOTHS) was upgraded on 13 June 2007 00Z.

Model Version

• The UM was upgraded from Portable UM (PUM) version 6.0 to 6.1.

Stochastic Physics

• Random Parameters Scheme (RP) upgraded to version 2.
• Stochastic Convective Vorticity (SCV) upgraded from version 6.0 to 6.1.
• Additionally a (portability) bug was fixed in SCV that resulted in anomalous results in the tropics.
• Stochastic Kinetic Energy Backscatter (SKEB) was introduced at version 11.3.

UM Science Pack was upgraded from cycle 34 to cycle 39 (representing the operational deterministic model active as of June 2006). The main features are:

• Small improvements to mid-level convection.
• Introduction of the adaptive detrainment scheme.
• Changes to the formulation of the thermal roughness length.
• Other miscellaneous boundary level improvements.
• Updated soil parameter ancillary fields.

GEFS v12 upgrade, effective 2020 Sep 23 12z:

• FV3 dynamical model, GFDL MP, Stochastic physics, two-tier SST with NSST, No TS relocation
• increased ensemble size from 21 to 31
• increased horizontal resolution for TIGGE from 1 to 0.5 degree

NAEFS upgrade, effective 12Z:

• GFS bias correction
• Combination of GFS & GEFS forecasts
• Probabilistic NAEFS forecasts
• Downscaled NAEFS forecasts

NAEFS Upgrade, effective 12Z:

• Ensemble size increased from 14 to 20 members

Ensemble Upgrade and NAEFS first Implementation, effective 12Z:

GEFS upgrade:

• increasing ensemble size from 10 to 14 members
• adding ensemble control for 0600 GMT, 1200 GMT and 1800 GMT
• introducing ET to breeding method
• NAEFS implementation:
• bias corrected forecast
• ensemble weights
• forecast anomalies

Ensemble Upgrade, effective 12Z:

• Increased resolution to T126 for all members between 180 hrs and 384 hrs (16 days)
• Changed initial pertubations from 24 hr breeding cycle to 6 hr cycle
• Added perturbed tropical storm vortex relocation

Effective 12Z: Bias-corrected QPF and PQPF forecasts

Effective 12Z: Ensembles available four times daily with T126 resolution out to 180 hrs

Effective 12Z: Mask Rescaling

Effective 12Z: Increased (T126) resolution for all members until 84 hrs

Effective 12Z: Change in regional rescaling procedure

Effective 12Z: Increased resolution for all members until 60 hrs

Effective 12Z: Increased membership at 12Z (10 members)

Effective 12Z: Increase in initial perturbation amplitude size

Effective 12Z: Change in regional rescaling procedure for setting initial perturbation amplitudes

Effective 12Z: New seasonally varying analysis uncertainty estimates introduced into regional rescaling procedure

NEPS-G upgrade to version 3:

The global model has been upgraded to UM11.2 (of Met Office, UK) and the model physics has been upgraded from GA6.2 to GA7.2 configuration in order to improve the representation of  the physical processes. The upgraded physics configuration includes improved treatment of gaseous absorption in the radiation scheme, JULES multi-layer snow scheme and improved treatment of warm rain and ice clouds.

Full details are available in the xls document linked in the page Models.  

Changes to Meteo-France contribution to TIGGE since 20 Jun 2022 :

• the last step in 18Z runs is changed from +108H to +102H

Further details can be found in the page Models.  

Changes to Meteo-France contribution to TIGGE since 08 Jan 2021 :

• 4 runs from 0/6/12/18Z instead of 2 from 6/18Z
  • there are different final steps for different runs:
    • 0/12Z: +48H
    • 6Z: +90H
    • 18Z: +108H
• resolution increase to 0.5 degree
• simple packing instead of JPEG one in GRIB 2 files
• tmin/tmax fix

Further details can be found in the page Models.