Ten years of forecasting atmospheric composition at ECMWF

Johannes Flemming, Vincent-Henri Peuch, Luke Jones


On 17 May 2007, the first graphs showing daily forecasts of reactive gases such as carbon monoxide and tropospheric ozone were published on the ECMWF website. This marked the beginning of daily atmospheric composition forecasting at ECMWF, which is now run operationally by the EU-funded Copernicus Atmosphere Monitoring Service (CAMS) operated by the Centre.

Initiated by the late Tony Hollingsworth, the GEMS project started in 2005 to build the capacity for a regional and global forecasting and data assimilation system of atmospheric composition. The idea for the global component of the GEMS system was to extend ECMWF's Integrated Forecasting System (IFS) in such a way that it could also be used for data assimilation and modelling of atmospheric composition in the troposphere and stratosphere.

Carbon monoxide forecast
Carbon monoxide forecast. Carbon monoxide (CO) at the surface on 22 June 2017 as predicted by the global CAMS system using the IFS. The map was generated using the interactive plotting tools available at Intensive forest fires in Portugal caused increased emissions of CO and other air pollutants in Europe at that time.

First steps

A noteworthy aspect of atmospheric composition developments at ECMWF is that they have been carried out in close collaboration with European partners, such as the Dutch national meteorological service (KNMI), the French national meteorological service (Météo-France), the National Center for Scientific Research (CNRS) in France and the Forschungszentrum Jülich in Germany.

First building blocks for the assimilation of stratospheric ozone were already in place before GEMS. They were further developed for the assimilation of reactive gases, aerosols and greenhouse gases. The inclusion of modelling components to simulate emissions injection, deposition and chemical conversion started as a collaborative effort in the GEMS project, was pursued in the series of MACC projects and is now continued in CAMS.

In 2005, directly including a complex atmospheric chemistry mechanism consisting of 50–100 tracers and 200–400 chemical reactions in the IFS was regarded as an uncertain venture. Instead a coupled approach was suggested: by means of the OASIS 4 coupler, three global chemical transport models (MOZART 3.5, MOCAGE and TM5) were coupled to the IFS. In the coupled system, the IFS would only transport and assimilate five key chemical species (CO, NOx, SO2 CH2O and O3), while the modelled source and sink terms were provided by the coupled chemical transport model. The first forecasts in May 2007 were generated by the coupled system IFS-MOZART without the assimilation of atmospheric composition observations.

A major step in the development of the CAMS global system was the introduction of data assimilation of aerosol optical depth and selected reactive gases in operational forecasts in July 2008. Other important upgrades of the system include improvements in the prediction of ozone hole chemistry and the use of daily observed biomass burning emissions (GFAS).

Towards an integrated system

The coupled system was finally retired in September 2014 because of its low computational efficiency. Following the example of aerosols, a chemistry scheme (CB05) had been integrated into the IFS within the MACC projects. The computational efficiency of the integration into the IFS finally enabled an upgrade in the horizontal resolution of the CAMS forecasting system from T255 (80 km) to T511 (40 km) in June 2016. Still, including chemistry and aerosols in the IFS increases the computational cost of a forecast by a factor of five, and that of a complete data assimilation cycle by a factor of two.

CAMS composition forecasts are run in the operational environment used for numerical weather prediction. After including new atmospheric composition developments in the latest IFS cycle, a preparatory system (e-suite) is run to evaluate forecast performance. The e-suite is later turned into the operational CAMS forecast suite (o-suite). Composition forecasts initialised only from the previous forecasts without data assimilation, as they were run ten years ago, are still run in research mode. They have proven to be very useful to evaluate the impact of the assimilation of atmospheric composition observations on forecast results.

CAMS global operational products have many uses, starting from the CAMS ensemble of regional air quality models in Europe, which use the global forecast as boundary conditions. The global CAMS system predicts dust storms and aerosols or plumes of air pollutants from wildfires. As part of the European project PANDA, CAMS global forecasts have demonstrated significant skill for air quality forecasting over China. Ozone and aerosol prediction are used to underpin the CAMS UV forecast and to assess the solar power potential for the energy sector. Several scientific aircraft campaigns to measure air pollution have used specially tailored CAMS composition forecasts to assist in the flight planning.

A planned upgrade of the CAMS global forecast suite will see an increase of the vertical resolution to the 137 level configuration of ECMWF's operational weather forecasts. A high-resolution (9 km) forecast of the greenhouse gases CO2 and methane will also be added to the CAMS product portfolio. Other upgrades in the pipeline include more advanced options to represent chemical and aerosol processes to improve the model components of the IFS. Another important development effort will be to enable the assimilation system to correct surface emissions. The assimilation of new atmospheric composition observations from the Copernicus space component, such as the polar-orbiting Sentinel 5P (later Sentinel 5) and the geostationary Sentinel 4 satellites, will bring opportunities to further improve atmospheric composition analyses and forecasts at ECMWF.