ECMWF plans to upgrade the horizontal resolution of its analyses and forecasts.The upgrade will have a horizontal resolution that translates to about 9 km for HRES and the data assimilation (the outer loop of the 4D-Var) and to about 18 km for the ENS up to day 10. A new cycle of the IFS will be introduced to implement the horizontal resolution upgrade. This cycle is labelled 41r2, and includes a number of enhancements to the model and data assimilation llisted herein. The detailed specification of the resolution upgrades included in IFS cycle 41r2 are:
These upgrades
|
The page will be updated as required. It was last changed on dd.mm.yyyy
For a record of changes made to this page please refer to Document versions .
Further information and advice regarding the upgrade can be obtained from User Support.
The planned timetable for the implementation of IFS cycle 41r2 is as follows:
Date | Event |
---|---|
4 Nov 2015 | Initial announcement to Member States and other customers |
early Dec 2015 | Availability of test data in dissemination |
end Mar 2016 | Expected date of implementation |
The timetable represents current expectations and may change in light of actual progress made.
The 2016 horizontal resolution upgrade has been developed with a trade-off between resolution and computational costs in mind. A number of options of how to produce the most effective combination of horizontal resolutions between 4D-Var, EDA, HRES and ENS have been tested to establish computing costs and to derive possible efficiency gains.
The most viable option found was to change from the current linear (TL) to cubic (TC) spectral truncation. With the cubic spectral truncation the shortest resolved wave is represented by four rather than two grid points. While keeping the spectral truncation unchanged, the resolution is increased in grid-point space to more accurately represent the physical processes and advection. In the current operational configuration of the IFS a build-up of energy at the shortest scales is mitigated by a lower-than-nominal resolution of the orography, strong horizontal diffusion and a de-aliasing filter. In IFS 41r2 this is much less of an issue. The TC configuration also substantially improves mass conservation.
In order to reduce the computational cost further, the use of a new octahedral grid with spectral truncation denoted by TCO has been investigated. The octahedral grid applies a new rule for computing the number of points per latitude circle and is globally more uniform than the previously used reduced Gaussian grid. It is based on a new mesh that also allows for future implementations of a hybrid spectral – grid point model. The computational cost is reduced by about 25% compared to the cubic grid as fewer grid point calculations are needed.
Data assimilation:
Compute scale-dependent hybrid B (background error covariance) by adding samples from latest EDA forecast to static climatological B with increasing weight of today's EDA for smaller wavelengths (30% up to T63, growing to a maximum 93% at T399).
The EDA now cycles its own background error and covariance estimates, rather than using climatological estimates.
Change to use the Sonntag equation for saturation vapour pressure in humidity observation operators to improve saturation calculation for very cold temperatures (colder than -40C).
Satellite:
GPSRO (radio occultation) observation errors based on a physical error propagation model are increased by 25% to account for missing sources of error (e.g. obs error correlations, forecast model error). Improves lower stratosphere/tropopause winds and temperatures.
Activated SSMIS F-18 humidity sounding channels over ocean and extended all-sky assimilation to snow covered land surfaces.
Improved specification of AMSU-A observation errors based on satellite (due to instrument noise characteristics and ageing) and situation (cloud, orography) thereby increasing the number of observations assimilated.
Improved aerosol detection and screening for IASI infrared satellite data.
Increased use of Atmospheric Motion Vectors (AMVs), including extension in latitudinal coverage from geostationary platforms from 60 to 64 degrees zenith angle and addition of Meteosat mid-height AMVs derived from infrared imagery.
Revised data selection (screening) of cold-air outbreaks in low-peaking all-sky microwave channels to allow more data to be assimilated.
Updated microwave observation operator coefficient files (54-level RTTOV files with latest spectroscopy)
Numerics:
Increased semi-lagrangian departure point iterations from 3 to 5 to remove numerical instabilities near strong wind gradients, particularly improving East Asia (downstream of the Himalayas) and improved representation of tropical cyclones.
Changed formulation of the horizontal spectral diffusion to a spectral viscosity with significantly reduced damping at the small scales.
Removed dealiasing filter on rotational part of the wind as no longer needed for cubic grid (no aliasing).
Reduced diffusion in the sponge layer near the top of the model (above level 30) scaled by grid resolution rather spectral resolution, due to new cubic grid.
Physics:
Improved representation of radiation-surface interactions with approximate updates every timestep on the full resolution grid leads to a reduction in 2m temperature errors near coastlines.
Introduced resolution dependence in the parametrization of non-orographic gravity wave drag, reducing with resolution and improving upper stratospheric wind and temperature for HRES and ENS.
Increased cloud erosion rate when convection is active, to reduce cloud cover slightly and improve radiation, particularly over the ocean.
Correction to solar zenith angle for the sunshine duration diagnostic. For clear sky days the sunshine duration increases by 2 hours, now in good agreement with observations. For cloudy days, sunshine duration may now be overestimated due to an existing underestimation of cloud optical thickness.
Ensemble:
Modified SKEB (Stochastic Kinetic Energy Backscatter) necessary for the new cubic grid, removing the numerical dissipation estimate from the dissipation rate. Reduces spread slightly, but this is then consistent with reduced RMSE in the new cycle.
Modified singular vector calibration to compensate for increased variance from the higher resolution EDA.
Information about the meteorological impact of the new cycle will be provided at a later date.
See IFS cycle 41r2 resolution changes for a summary of the resolution changes for the atmospheric, wave and ocean component models of HRES and ENS.
IFS cycle 41r2 introduces a new form of the reduced Gaussian grid, the octahedral grid, for both HRES and ENS. See Introducing the octahedral reduced Gaussian grid for further details.
The figures below show the grid spacing and orography fields in a cylindrical projection for HRES and ENS Legs 1 and 2 over a region covering the Alps (43.0° S to 49.0° N, 4.5° E to 17.0°).
Old: N640 original reduced Gaussian grid (~16 km resolution) | New: O1280 octahedral reduced Gaussian grid (~9 km resolution) |
---|---|
ENS
Old Leg 1: N320 original reduced Gaussian grid (~32 km resolution) | New Leg 1: O640 octahedral reduced Gaussian grid (~18 km resolution) |
---|---|
Old Leg 2: N160 original reduced Gaussian grid (~64 km resolution) | New Leg 2: O320 octahedral reduced Gaussian grid (~36 km resolution) |
The octahedral reduced Gaussian grid has a slightly lower resolution than the corresponding original reduced Gaussian grid with the same number of latitude lines. For example, the O640 grid has a resolution of ~18 km whereas the resolution of the N640 grid is ~16 km. |
The increase in horizontal resolution for HRES and ENS is reflected in changes to the GRIB headers. For the model reduced Gaussian grid these are documented in the table below.
GRIB 1 Section 2 Octets | GRIB 2 Section 3 Octets | grib_api key | HRES | ENS Leg 1 | ENS Leg2 | |||
---|---|---|---|---|---|---|---|---|
Old | New | Old | New | Old | New | |||
9-10 | 35-38 | Nj | 1280 | 2560 | 640 | 1280 | 320 | 640 |
11-13 | 47-50 | latitudeOfFirstGridPointInDegrees | 89.892 | 89.946 | 89.784 | 89.892 | 89.570 | 89.784 |
18-20 | 56-59 | latitudeOfLastGridPointInDegrees | -89.892 | -89.946 | -89.784 | -89.892 | -89.570 | -89.784 |
21-23 | 60-63 | longitudeOfLastGridPointInDegrees | 359.859 | 359.929 | 359.718 | 359.860 | 359.437 | 359.722 |
26-27 | 6-7 | N | 640 | 1280 | 320 | 640 | 160 | 320 |
33-nn | 35-nn | pl | N640 | O1280 | N320 | O640 | N160 | O320 |
The GRIB model identifiers (generating process identification number) for the new cycle will be changed as follows:
GRIB 1 Section 1 Octets | GRIB 2 Section 4 Octets | grib_api key | Component | Model ID | |
---|---|---|---|---|---|
Old | New | ||||
6 | 14 | generatingProcessIdentifier | Atmospheric model | 145 | 146 |
Ocean wave model | 111 | 112 | |||
HRES stand-alone ocean wave model | 211 | 212 |
None.
The size of the grid point fields produced have increased by a factor of 3 while the size of the spectral fields remains unchanged. When retrieving data via MARS or dissemination, if no spectral truncation or grid resolutions are specified, fields are provided at model resolution.
In particular, users should be aware of the increase in memory and CPU time needed to process the increased resolution (grid-point) fields and adjust their programs and batch scripts appropriately. Increase in data volume provides a summary of the field sizes at the new horizontal resolutions.
The ECMWF software stack has been upgraded to support the octahedral grid. This includes upgrades to EMOSLIB, MARS, Metview, GRIB_API and product generation (dissemination).
All software versions are subject to change depending on the results of ongoing testing. This page will document the current status of the software versions needed to process fields on the octahedral grid. |
MARS has been adapted to support retrieval of data on the octahedral reduced Gaussian grid. Users can test retrieval of data from the pre-operational e-suite on ECMWF systems with the command:
mars <retrieve request> |
The MARS post-processing keyword 'GRID' now accepts values comprising a letter denoting the grid name followed by an integer (the grid number) representing the number of lines from pole to equator. For example, use:
Not all grid names and grid number combinations are supported. MARS requests specifying an unsupported grid will fail. For example, a retrieval request with GRID=O512 will return an error. A list of supported grid names and grid numbers will be provided.
In addition, the MARS post-processing keywords GAUSSIAN=REDUCED and GAUSSIAN=REGULAR are deprecated.
MARS requests specifying GRID=AV will return the model grid. After the implementation of IFS cycle 41r2, this will be O1280 for HRES and O640 / O320 for ENS Leg 1 /Leg 2.
The MARS keyword RESOL=N128 can be used to truncate products from ENS Leg 1 to the N=128 original reduced Gaussian grid of Leg 2 prior to interpolation to regular latitude-longitude grids. Following the horizontal resolution upgrade, anyone using this intermediate interpolation method should use RESOL=O320 to truncate O640 Leg 1 products the O320 octahedral grid of ENS Leg 2.
Magics 2.24.7 provides preliminary support for the octahedral grid.
Metview 4.6.0 provides preliminary support for the octahedral grid. Registered users can test this version on ECMWF systems.
EMOSLIB 000420 provides preliminary support for octahedral grids. This includes full support for interpolation of input octahedral grids to any of the previously supported grids as well as spherical transforms to the new octahedral grids.
For more information please check version 000420 changes..
This version has a known performance issue when interpolating from Octahedral to lat/lon grids. We are currently working on improving the speed. |
Starting from EMOSLIB 000410 calls to GRIBEX are no longer supported. Any call to GRIBEX will result in ABORT'ed code with the following error message:
Any code that still relies on GRIBEX should be migrated to use GRIB API at your earliest convenience. |
Preliminary support for the octahedral grid is provided from grib_api 1.14.2. Users of the grib_find_nearest routine are strongly advised to upgrade to this version.
Older versions of grib_api can decode the octahedral grid successfully
Information about technical changes to dissemination will be provided at a later date.
Information will be provided at a later date for users wishing to test their time-critical option 1, 2 or 3 applications with the higher resolution data .
Test data from the cycle 41r2 e-suite are available in MARS. The data are available with experiment version is 0069 (MARS keyword EXPVER=0069).
One full day (00 UTC and 12 UTC) of forecast data for both HRES and ENS, including ENS model level fields, has been stored for 10 August 2015.
The data can be accessed in MARS from:
The data are provided to enable users to test the handling of the new grid and testing their programs. In particular, changes to meteorological fields may be needed depending on the outcome of ongoing testing.
The data are intended for testing technical aspects only and should not be used for operational forecasting. All registered users of ECMWF computing systems will be able to access the test data sets in MARS. Users of our operational real time forecasts without access to our systems can contact our Data Services to obtain test data. |
Please report any problems you find during your testing to User Support.
Test data from the pre-operational e-suite will be available through test dissemination at a later date.
Date | Reason for update |
---|---|
4 November 2015 | Initial version |