Every day, ECMWF produces various global Analyses and Forecasts and archives them in MARS. The first ECMWF numerical model in 1979 was a grid-point model with 15 levels in the vertical and a horizontal resolution of 1.875 degrees in latitude and longitude, corresponding to a grid length of 200 kilometres. A number of major changes have occurred since ECMWF's activity started:

• In April 1983, the grid-point model was replaced by a T63 spectral model (i.e. a spectral representation in the horizontal with a triangular truncation at wave-number 63). The number of levels in the vertical was increased to 16.
• In May 1985, the spectral truncation was extended to wave-number 106. The number of levels was increased to 19 in 1986.
• In September 1991, a much higher resolution spectral model was put into operation. The spectral truncation was extended to wave-number 213 and the number of levels was increased to 31 (i.e. T213L31).
• In 1992, a wave model used for ocean wave forecasting (the WAM model) became operational, followed after a few months by a Mediterranean implementation.
  In December the Ensemble Prediction System (EPS) with 32 members and a resolution of T63L31 became operational.
• In November 1997, 4d-Var (four-dimensional variational Analysis) became operational.
• In April 1998, the spectral truncation was extended to wave-number 319.
• In June 1998, the atmospheric model and the wave model were coupled in order to take advantage of the boundary conditions they represent to each other.
• In March 1999, the number of levels was increased to 50.
• In October 1999, the number of levels was increased to 60.
• In September 2000, 4d-Var cycling was increased to 12 hours. Type First Guess has been discontinued.
• In November 2000, the spectral truncation was extended to wave number 511.
• In June 2004, the Early Delivery Forecasting System was implemented, comprising two main 6-hour 4d-Var analysis and forecast cycles for 00 and 12 UTC and two additional 12-hour 4d-Var analysis and first guess forecast cycles.
• In February 2006 the resolution of the deterministic model has been increased to T799 and 91 model levels. The resolution of the Ensemble Prediction System (EPS) went up to T399 and 62 model levels.
• In November 2006 the EPS has been upgraded to 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 and T255 L62 for day 11 to day 15.
• In March 2007 the operational seasonal forecasting system has been upgraded from System 2 to System 3 and integrated in the Multi-Model Seasonal Forecast stream. The resolution of the atmospheric model has been increased to T159L62.
• In March 2008 VarEPS and monthly forecasting were combined into a single system. On Thursday of each week, the 00 UTC VarEPS forecast are extended from 15 to 32 days at a resolution of T255 L62 with ocean coupling introduced from day 10.
• In January 2010 the horizontal resolution of the deterministic model has been increased to T1279. The resolution of the EPS went up to T639/T319 for Leg A/B respectively.

• In June 2013 IFS cycle 38r2 introduced higher vertical resolution in the high-resolution (T1279) forecast and data assimilation of the operational runs at 00 and 12 UTC (HRES) as well as the 06 and 18 UTC cycles of the Boundary Conditions (BC) optional programme. The number of model levels increases from the current 91 levels (L91) to 137 levels (L137).

• In November 2013 the vertical resolution and the vertical extent used for the medium-range and monthly ensemble forecasts has changed: the number of levels of the ENS has increased from 62 to 91 with the model top raised from 5 hPa to 0.01 hPa. The MARS streams DCDA, DCWV, ENDA and EWDA were discontinued. 

In addition, various other projects run on a regular basis, such as Seasonal Forecast, Multi-Analysis Ensemble, Monthly Forecast and Sensitivity forecast, or have run in the past, for example the ECMWF Re-Analysis. All their outputs are available in MARS. The majority of Observations used as model input are also available in MARS.

A comprehensive list of changes in the ECMWF model can be found on our main website, giving detailed documentation on significant changes to the operational forecasting system.

Products

In order to know the data available from MARS, users need to be familiar with ECMWF's activities. The overview given below is not exhaustive. It is rather meant as an introduction to ECMWF's activities and the most common products. Users wanting to learn more are encouraged to study the User Guide to ECMWF Forecast Products.

The datasets pages on the main website provide a good entry point to browse the archive content in particular for operational and reanalysis data. From there you will also find links into the MARS catalogue, which allows you to browse the entire archive content.

Operational data produced daily at ECMWF

Atmospheric deterministic models  

• Analysis:global analyses for the four main synoptic hours 00, 06, 12 and 18 UTC. They are the best gridded estimate of the state of the atmosphere (best fit to observations). For each of the synoptic hours, data is produced at the following levels:
  
  • Surface fields represent the meteorology at the surface.
  • Model levels are used in ECMWF's forecast model to resolve the atmosphere in the vertical.
  • Pressure levels are interpolated by the model from its Model Levels.
  • Isentropic levels are either potential vorticity or potential temperature.
    
    
• Forecast:global 10-day forecasts based on the 00/12 UTC Analysis (the 00 UTC run started on March 2001 as an experimental suite for severe weather prediction). Forecast products are classified in the same level types as Analysis data: Surface, Model levels, Pressure levels, and Isentropic levels. Meteorological parameters are written output for every forecast time step, 3-hourly intervals from 00 to 72 hours, and 6-hourly from 72 to 240 hours.
• First guess:a forecast with base time from the previous synoptic hour and a forecast time step of (usually) 6 hours. Note that since the change to 12 hour-cycling 4d-Var in the year 2000, this type of data has been discontinued.
• Initialised Analysis:the best 'balanced' gridded estimate of the state of the atmosphere (initial state of the forecast). With the implementation of the three dimensional variational Analysis (3d-Var) on January 1996, the Initialised Analysis is no longer produced, and the Analysis is the best and the best 'balanced' gridded estimate of the atmosphere.
• Four-dimensional variational Analysis:analyses using observations over a time window (e.g. a 6 hour 4d-Var at cycle hh will contain observations from hh-02:59 to hh+03:00).
• Errors in Analysis:the assumed uncertainty of an observation is combined with the assumed uncertainty of the First Guess, resulting in an estimate of the total uncertainty in the Analysis
• Errors in First Guess:result of the uncertainty of the First Guess compared with observations
• Four-dimensional variational Analysis increments:the low resolution increment which is added to the first-guess after each inner loop minimisation.

Wave deterministic models  

Since 1998, ECMWF's deterministic atmospheric model is coupled with a wave model.

• Analysis: global analyses of the surface of the oceans for the four synoptic hours.
• Forecast: global 10 day forecasts based on 00/12 UTC Analysis.

ECMWF runs the WAM model at a finer resolution for a limited area. It used to be called the Mediterranean model but is now known as the European Wave Model. This model is run to 5 days and produces Analysis and Forecast.

Ensemble Prediction System   

ECMWF's Ensemble Prediction System has a coupled atmospheric and wave model. On 28 November 2006 the Variable Resolution Ensemble Prediction System (VarEPS) has been introduced by extending the forecast range from 10 to 15 days with a resolution of T399L62 for day 1 to day 10 (Leg 1) and T255L62 for day 11 to day 15 (Leg 2). On 11 March 2008 the Monthly Forecasting System, running once a week, has been integrated with the VarEPS. The new monthly forecast products were produced for the first time on 13 March 2008.

• Control forecast: an unperturbed forecast at a lower resolution than the main 10-day deterministic forecast. Forecast runs to 15 days, with lower resolution from truncation step 240 onwards. Data is available on Surface, Model levels, Pressure levels and Isentropic levels.
• Calibration/Validation forecast: VarEPS includes two constant-resolution forecasts for calibration and validation purposes which run for both resolutions from day 1 - 15. Data is available on Surface, Model levels, Pressure levels and Isentropic levels.
• Perturbed forecasts: different forecasts to 10 days with perturbed initial conditions. They are numbered from 1 to N depending on the EPS setup. Data is available on the Surface and on Pressure levels and Isentropic levels.
• Initial condition perturbations: the initial conditions for the EPS are designed to represent the uncertainties inherent in the operational analysis. They are created by adding perturbations to the operational analysis which produce the fastest energy growth during the first two days of the forecast period, defined using the singular vector technique.
• Forecast probabilities: a statistical distribution of the weather parameters from all ensemble members is used to produce probabilistic weather forecasts. With the introduction of VarEPS this data type has been discontinued.
• Event probabilities: provide the probabilities of the occurrence of weather events at each grid point. The probabilities are calculated on the basis that each ensemble member is equally likely.
• Ensemble means: are means of the ensemble forecast members.
• Clusters: similar ensemble members are grouped together into clusters. The mean and standard deviation of these clusters are computed (as well as the mean and standard deviation of the overall ensemble). Five sets of clusters are computed, one for the entire European area, and four for smaller areas.
• Tubes: another clustering method which averages all ensemble members which are close to the ensemble mean and excludes members which are significantly different.
• Extreme Forecast Index(EFI): measures the difference between the probability distribution from the EPS and the model climate distribution.

Boundary conditions  

Also known as short cut-off. Four additional Analyses are run for 00, 06, 12 and 18 UTC with a cut-off time of 4 hours, followed by global 4-day forecasts to provide some Member States with boundary conditions for their limited area models. All Analyses data but only the Forecast from 00 UTC are archived.

Since the 14. March 2006 the BC project has been merged with the main deterministic operational forecast suite. In the new configuration, only the 06 and 18 UTC runs are part of the BC suite. For the 00 and 12 UTC runs data are provided from the main deterministic forecast suite. Now all four data assimilation cycles are based on 6h 4d-Var with a cut-off time of 3 h. Temporary data storage for 3 months will be continued in MARS for 06 and 18 UTC data with STREAM=SCDA.

Multi-Analysis Ensemble  

Every day ECMWF receives Analyses from four centres, NCEP, The Met Office, Météo-France and Deutscher Wetterdienst, and runs 5 forecasts, 1 based on each different analyses plus one compound of all the analyses (consensus) including ECMWF's Analysis.

Seasonal Forecast  

ECMWF started an experimental programme for seasonal prediction in 1995, which attempts to predict seasonal changes by coupling three models: atmospheric, wave and ocean models. More information on the Seasonal Forecast can be found on the ECMWF Website.

Monthly Forecast  

With the same setup as the seasonal forecast, this project runs every week producing forecasts to 32 days, again, coupling atmospheric, wave and ocean models. More information on the Monthly Forecast can be found on the ECMWF Website.

Ocean Analysis  

The main purpose of the ocean analysis is to provide initial conditions for the extended range forecasts (seasonal and monthly). There are two streams: The re-analysis stream spans the period 1959 up to 11 days behind real time whereas the real-time stream started in August 2006. More information on the Ocean Analysis can be found on the ECMWF Website.

Monthly and climatology datasets

ECMWF maintains an archive of monthly means data from the atmospheric and wave model archive. The resolution and internal representation of the archive may change according to changes in ECMWF's operational practice.

• Atmospheric Analysis monthly means are averaged over the calendar month for each of the synoptic times 00, 06, 12 and 18 UTC.
• Atmospheric Forecast monthly means: are the average of all the forecasts for a particular step that verify in the selected calendar month. All monthly means are archived at model resolution.
• Wave Analysis monthly means: are averaged over the calendar month for each of the synoptic times 00, 06, 12 and 18 UTC.
• Wave forecast monthly means: are the average of all the forecasts for a particular step that verify in the selected calendar month. Wave Monthly means are archived at 1.5 degree resolution.

There is a limited climatology data set, which contains the Geopotential and Temperature on Pressure Levels for each month of the year. The data originally came from NCEP, Washington, and it was processed to store it in MARS. Although it is available to users, it is recommended that modern applications use the Monthly means archive or the Re-Analysis Monthly means archive for climatology purposes.

Special datasets

ECMWF Re-Analysis  

The ECMWF Re-Analysis (ERA) project has so far produced three datasets:

• ERA-15, a validated 15 year data set of assimilated data for the period 1979 to 1994
• ERA-40, a validated 45 year data set of assimilated data for the period 1957 to 2002
• ERA-Interim, a validated set of assimilated data starting from 1979, extended to the present in near real-time

These available datasets include: analysis, forecast and forecast accumulations as output from atmospheric model, as well as analysis and forecast from a wave model re-analysis. There is also a Monthly Means data set containing data at the resolution of the data assimilation and forecast system used by each Re-Analysis.

TIGGE  

The Interactive Grand Global Ensemble, is a key component of THORPEX: a World Weather Research Programme to accelerate the improvements in the accuracy of 1-day to 2 week high-impact weather forecasts for the benefit of humanity. Global ensemble forecasts to around 14 days generated routinely at different centres around the world. Currently data from ECMWF, JMA (Japan), Met Office (UK), CMA (China), NCEP (USA), MSC (Canada), Météo-France, BOM (Australia), CPTEC (Brazil) and KMA (Korea) is archived.

ENSEMBLES  

The EU-funded ENSEMBLES project intends to develop an ensemble prediction system for climate change based on the principal state-of-the-art, high resolution, global and regional Earth System models developed in Europe, validated against quality controlled, high resolution gridded datasets for Europe, to produce for the first time, an objective probabilistic estimate of uncertainty in future climate at the seasonal to decadal and longer timescales. A large set of seasonal, annual and decadal hindcasts is available which have been produced with different forecast systems run by ECMWF, Météo-France, Met Office, IfM, INGV and CERFACS. These systems addressed the important problem of the impact of model uncertainty on forecast error by using the multi-model, stochastic physics and perturbed parameters approaches.

DEMETER  

is the acronym of the EU-funded project "Development of a European Multi model Ensemble system for seasonal to inTERannual prediction". The objective of the project is to develop a well-validated European coupled multi-model ensemble forecast system for reliable seasonal to inter annual prediction. Six comprehensive European global coupled atmosphere-ocean models are being installed at ECMWF, those of: ECMWF, Météo-France, LODYC, Met Office, MPI, INGV, INM-HIRLAM and CERFACS.

TOST  

TOST is the THORPEX Observing System Test, an experiment carried out at the end of 2003 to evaluate targeted observations in an Ensemble Prediction System. There is output from 3 different models, ECMWF, Météo-France and United Kingdom Met Office.

PROVOST  

stands for Prediction Of climate Variations On Seasonal to inter annual Time scales. They are a set of experiments from four centres: ECMWF, Météo-France, EDF and Met Office. The experiments are 120 day runs from 9 consecutive starting days, with write-ups every 24 hours of Pressure level and Surface data.

ECSN-HIRETYCS  

ECSN is the European Climate Support Network. HIRETYCS is the High Resolution Ten Year Climate Simulation. This data set consist of 10-year climate simulations produced at three centres: Centre National de Recherches Météorologiques (CNRM), Max-Planck Institute (MPI) and United Kingdom Met Office.

Research experiments

A vast amount of data is archived daily containing IFS (Integrated Forecast System) experiments produced by ECMWF's Research Department or by Member States' users at ECMWF. Basically, an experiment can address any area of meteorology and it is archived accordingly. Users wanting to retrieve Research experiments need to know in advance the name of the specific experiment and its nature. For this information please, contact User Support.

Member States projects

Some Member States also archive output other than from IFS into MARS. One example of such activity is the COSMO-LEPS forecast suite, running daily at ECMWF.

Input data to ECMWF models or for verification/scores

Fields  

ECMWF archives a selection of products, both Analysis and Forecast, from other Centres: Exeter, Melbourne, Montreal, Offenbach, Toulouse, Tokyo and Washington. Most of these products exist on the GTS and they are combined to create global fields with ECMWF local GRIB header extensions added. A convention exists to create empty (dummy) fields when there are missing data (e.g. if problems arise at one of the Centres or on the GTS).

There are other fields which are used as input for the ECMWF forecasting system (e.g. Sea Surface Temperatures from Washington which are used by ECMWF's Analysis).

Images from Meteosat and GOES Satellites are coded in GRIB and archived in MARS.

Observations  

Observations used as input to the assimilation system are also archived in MARS, as well as the Feedback on how the observations are used in the analysis. There are, amongst others:

• Surface data, conventional and satellite data covering: synoptic observations, soil and earth temperature, buoy information and PAOB observations (pseudo surface pressure observations in the Southern Hemisphere).
• Vertical soundings, conventional and satellite data: PILOT, TEMP, ROCOB, RTOVS, ATOVS, SATEM.
• Upper-air data, conventional observations such as AIREP and ACARS, together with satellite upper-air SATOB, high resolution winds and geostationary radiances.

MARS also holds the observations in the form of files as they are presented to the Analysis (Analysis Input), as well as the feedback files (Analysis Feedback). These are mainly used to allow to reproduce in the future any past operational run. There is one set of files per synoptic time.

Data deficiencies

There are some known deficiencies in the archives (suspect or missing data). Those are listed in the Data Services pages and in the Archive Changes pages.  

Data formats

Archived data is stored in two WMO formats: GRIB for fields and BUFR for observations. In addition, observation feedback is archived in the ECMWF/IFS format ODB (Observational Data Base). In general, the retrieved data is returned in the archive format. However, with the Data Server or the Web API it is possible to request fields to be returned in netCDF format.   

WMO FM 92-IX Ext GRIB

GRIB (GRid In Binary) is a WMO defined format for meteorological field data, or (more generally) any regularly spaced gridded data. All ECMWF model output is in GRIB format with ECMWF local extensions in their headers. The GRIB format is handled via the GRIB  software. Fields are archived in one of the following spatial coordinate systems:

Spherical Harmonics (SH) mainly for upper air fields

Gaussian Grid (GG) mainly for surface data, although some upper air fields as well

Latitude/Longitude (LL) other centre's data, wave and ocean data

For the correspondence between the three types of grid resolutions see table 1.

Table 1: Correspondence between resolutions of ECMWF grid types.

The GRIB format is handled via the GRIB  software.

WMO FM-94 BUFR

BUFR (Binary Universal Form Representation) is a WMO defined format for point data (irregularly spaced), which is used for archived observations. The BUFR format is handled via BUFREX and other EMOSLIB subroutines as described in the BUFR User's Guide.

ODB

In the IFS observations are handled by ODB (Observational Data Base).

ODB is a

• Hierarchical in-core database with a data definition and query language: ODB/SQL
• A data format
• ...

ODB Observation Feedback (ofb) data is archived in MARS to improve its representation in the MARS meta data. ODB also introduces SQL capabilities to request feedback data.

• To improve the handling of observations, ODB will be further integrated into ECMWF systems.