1. GEMS Kick-Off Meeting, Hamburg Aerosols: WP3 Jean-Jacques Morcrette, Olivier Boucher With contributions at ECMWF from: Angela Benedetti: background error statistics Erik Andersson: Overseeing the assimilation of GEMS observations

2. GEMS AEROSOL Global Monitoring System Leader: O. Boucher

3. GEMS Aerosols AER_1: Implementation of the direct physical aerosol model in the ECMWF model (O.Boucher, J. Feichter) > implementation of parametrisations for tropospheric aerosols > implementation of parametrisations for stratospheric aerosols > implementation of new emission inventories > implementation of aerosol optical properties > production of test simulations HC-MO, MPI-M, CEA-LSCE, ECMWF, SA-UPMC AER_2: Refinement of aerosol emission sources (M.Sofiev) > update and assimilation of the anthropogenic emission inventories of aerosol and its precursors > assimilation of information on wild fires > quantification of the wind-blown dust emission from desert areas > quantification of the wind-blown sea salt emission > sources of stratospheric aerosols FMI, CEA-LSCE, MPI-M, SA-UPMC

4. GEMS Aerosols AER_3: Aerosol data assimilation (J-J Morcrette) > adaptation of RT codes for SW and LW radiances in nadir geometry > preparation and harmonisation of aerosol satellite data sets > error covariance matrices > test of a 1D-Var system using aerosol products > test of a 1D-Var system using aerosol radiances ECMWF, CEA-LSCE, HC-MO, SA-UPMC AER_4: Evaluation of the model and analyses (C.ODowd, I. Chiapello) > assessment of diagnostics and skill scores > evaluation of aerosol radiative properties and associated radiative fluxes > evaluation of aerosol physico-chemical properties > analysis of model results with respect to air quality NUIG, CNRS-LOA, CEA-LSCE, DWD, ECMWF, MPI-M, RMIB, SA-UPMC

6. GEMS Aerosols: products @ end of contract Analysis of aerosol-related observations at ERA40 resolution (T L 159 L60 [1.125 deg] 2 or better) twice a day Total optical thickness at ~0.55 m OCEAN Angstrom coefficient (or at~ 0.865 m) Total optical thickness at ~0.55 m LAND From model 12-hour forecasts used in assimilation cycle.Up to 15 mixing ratio profiles of aerosols, every 3 hours sea salt 3 bins 1 st stage desert dust 3 bins organic 2 bins 2 nd stage black carbon, carbonaceous 2bins sulfate fly ash stratospheric.Corresponding 2D-fields for sources and sinks

7. GEMS-AEROSOL: initial steps at ECMWF JJMorcrette, A.Benedetti, S.Serrar, A.Beljaars, P.Bechtold, A.Untch Introduce aerosol prognostic variables in the Integrated Forecast System and assimilate global aerosol information Instruments: MERIS, MODIS x 2, MISR, SEAWIFS, POLDER R/T 6S Modelling LMDZ Sources/Sinks LMD-Inca Data Assim. 1DVar Variables, radiances Validation AERONET

8. Error covariance matrices Assumptions: We will deal with unbiased Gaussian distributions, that can therefore be completely characterized by their covariance functions (or matrices in the discrete case) We need to deal with: –Model error covariance, describing covariance between errors in the model at any two locations and two time instances –Background error covariance: related to correlation between errors of different model variables (e.g., geostrophic balance introduces a strong correlation between temperature and wind errors) –Observation error covariance: includes measurement errors, imperfections in the design of the observation operators and representativeness errors (linked to the different scales represented by the model and the observations)

9. GOALS THIS YEAR (from RD Memorandum by Andersson and Benedetti, http://w3ec2.ecmwf.int/publications/library/do/references/list/3 ) 1. May 2005: Run IFS with simulated aerosol data a. Establish first ODB format for aerosol total optical depth data (AB, DT) b. Introduce new observed variable in IFS (AB) 2. May 2005: Enable aerosol forecast - no physics a. Introduce aerosol as one of the GFL variables (AU, done) b. Initial data from climatological aerosol in grib (JJM, SS) 3. July 2005: Aerosol appearing in JO-table, simulated data a. Implement an aerosol observation operator, called from hop.F90 (AB) b. Calculate departures for simulated aerosol observations (AB) c. Use new generic GOM-arrays introduced in 29r2 (MH, AB) 4. Sept 2005: Establish an aerosol 1D-Var a. Based on existing cloud/rain 1D-Var (AB) b. To be used intermittently for Jo and Jb testing and validation (AB) 5. Nov 2005: Calculation of aerosol Jb cost function a. Make Jb codes generic for GEMS variables- reproduce current result for ozone (MF, RE) b. Run a set of aerosol forecasts (for Jb calibration), with or without physics and sources/sinks depending on progress with the AER work (AB, JJM) c. Derive Jb statistics, using the lagged forecast approach (NMC-method) (MF, RE, AB) 6. Dec 2005: Functional assimilation test (AB) a. Communication and interpolation of GEMS variables between the job steps of incremental 4D-Var (increments and trajectory) (LI, YT) b. Scripts generalisations for GEMS variables (JH) c. Generalisation of grid-point variables (LI)

10. AN AEROSOL DATA ASSIMILATION SYSTEM Observation define Format BUFR ODB Observations into ODB/IFS Observation operator Model and Parameterizations B-matrix Calculation (1 st estimate) TL/AD Model and Parameterizations Control variable Interface TL/AD Extend Jb Nearly completed tasks Work in progress

11. BACKGROUND ERROR STATISTICS (1 st estimate) Use NMC method (differences between 48 and 24h forecasts) Three aerosol species (now extended to six species with comparable results): - desert dust - seal salt - continental particulate (- background tropospheric ) (- background stratospheric) (- black carbon) Forecasts initialized from climatology Thirty days of 5-day forecats for January and July 2001 (now extended to 4 months, with comparable results). Little seasonal differences. Parameters computed: vertical and horizontal covariances and error correlations (global fields, zonal fields and integral averages over whole globe, Tropics, Northern Hemisphere and Southern Hemisphere) for total aerosol mixing ratio.

12. Variance for total aerosol mixing ratio, zonal average, January 2001

13. Vertical correlations, zonal average, level 1000 hPa, January 2001

14. Level 1000 hPa Average vertical correlations, January 2001 Level 200 hPa Level 500 hPa

15. MODIS observations MODIS (MODerate Resolution Imaging Spectroradiometer): Provides high radiometric sensitivity in 36 spectral bands from 0.4 µm to 14.4 µm. Resolution varies between 250 m and 1 km, depending on the channel. Level 2 sample data were downloaded and can be found in ec:/oparch/gems/modisaerosol/terra[aqua]/200505/05 These files contain retrieved products over ocean and land including optical depth at several wavelengths and Angstrom exponent. The resolution is 10x10 km and the data are grouped in 5-min granules. Suggested MODIS products for use at ECMWF (assimilation/verification): Reflectance over land/ocean Aerosol optical depth over land/ocean (all available wavelengths) Angstrom exponent Satellite/atmospheric parameters (i.e. zenith angle, viewing angle, etc.)

16. AEROSOL OPTICAL THICKNESS (AOT) at 0.55 µm from MODIS Kaufman et al., Nature, 2002 COARSE PARTICLES 0.0 0.1 0.2 0.3 0.4 0.0 0.15 0.3 0.45 0.6 FINE PARTICLES

17. ANGSTROM EXPONENT Expresses the spectral dependence of aerosol optical thickness with the wavelength of incident light. Large particles generally have low values of Angstrom exponent, while small particles have large values. Chang et al., 2005 Large values of Angstrom exponent indicating the presence of small aerosol particles from a recent forest fire.

18. FUTURE PLANS (from RD Memorandum by Andersson and Benedetti, http://w3ec2.ecmwf.int/publications/library/do/references/list/3) Q1, Q2 and Q3 2006: Monitoring of real aerosol data and validation a.Acquire real aerosol data from MODIS (and other sources depending on availability) b.Run with (parts of) aerosol physics depending on progress in AER c.Contribute to model and physics validation, based on comparison with real aerosol data d.Perform data monitoring for a period of time of the main data sources e.Adopt monitoring software (e.g. obstat and SATMON, Metview??) f.Study data quality g.Assess biases Q4 2006 and Q1 2007: Preliminary assimilation experiments a.Implement quality control thresholds b.Decide on data selection, i.e. which data to use actively c.Re-tune, massage and re-calibrate Jb, possibly several times

19. ITEMS FOR DISCUSSION Better definition of collaboration with partners on specific tasks Examples for WP_AER_3 (Aerosol data assimilation): Task 3.2: Preparation and harmonisation of aerosol datasets: it has been suggested that CNRS-LOA reprocesses MODIS data and ECMWF gets data from LOA. Need to coordinate with ECMWF –specific processing needs. Task 3.3: Error covariance matrices: from the proposal it appears that error covariance matrices will be specified from partners CEA-LSCE and CNRS-LOA. Great for observational error covariance matrices, caveats for background error covariance matrix – this needs to be tailored to ECMWF assimilation system.

20. 6S: Second Simulation of the Satellite Signal in the Solar Spectrum: Vermote et al., 1997, version 4.1 –allows the computation of radiances in the shortwave channels of most of the present satellites: AVHRR, GOES, HRV_spot, METEOSAT, MODIS, POLDER, TM_landsat –under the proper satellite/sun/target geometry –and various specifications of the surface homogeneous vs. inhomogeneous without and with directional effect with possibly different target and environment characteristics with different surface representations: Hapke, Verstraete et al., Roujean et al., Walthall et al., Minnaert, Iaquinta and Pinty, Rahman et al., Kuusk allowing sophisticated descriptions of the surface including structural parameters for the canopy. Simulating the SW radiances

21. What still has to be done with the 6S RT code expected before the end of 2005 Speed-up the direct code Develop and test the TL and AD

22. Likely problems: vertical distribution –With an assimilation of satellite optical thickness or radiances, only the total optical thickness (vertically integrated distribution) of aerosols is likely to be available. Is this information enough for the model? How to distribute it on the vertical? How to distribute it between different types of aerosols in terms of: – amounts – optical properties Pandoras box - 1

23. Pandoras box - 2 Likely problems: Surface effects –Surface reflectance: the derivation of total aerosol thickness from satellite observations cannot be properly done without knowledge of the surface directional spectral reflectance below these aerosols. This holds for aerosols over land and over ocean. –How to make sure that the retrieval of aerosol and surface properties use consistent approaches? Do we need considering a simultaneous variational retrieval of surface and aerosol properties? A particular problem might be that, up to now, the aerosol remote sensing community is quite distinct from the surface albedo remote sensing community. –Various algorithms have been attempted in the past: TOMS: Re UV ~0, get aer : problem with vertical positioning of aerosol layer within the Rayleigh column (Seftor et al., 1997, JGR) MODIS: 2.1 m Re aer ~0, get Re surf, from 0.67 and 0.47 m channels, get aer : problem with spectral redistributing of aerosol (and surface) optical properties, particularly for desert aerosols, and over bright surfaces (Kaufman et al., 1997, JGR) POLDER: can model polarization by surface, get aer : not all surfaces well represented, and problems with little polarizing by big aerosols (Deuze et al., 2001, JGR; Tanre et al., 2001, GRL)

24. Pandoras box - 3 Likely problems: Surface effects –How to make sure that the retrieval of aerosol and surface properties use consistent approaches? Do we need considering a simultaneous variational retrieval of surface and aerosol properties? A particular problem might be that, up to now, the aerosol remote sensing community is quite distinct from the surface albedo remote sensing community.

25. END thank you for your attention!

26. Sea salt, global map at 1000 hPa, day-5, 1 Jan. 2001 Units are mg/kg – max of color scale is 1

27. Desert Dust, global map at 1000 hPa, day-5, 1 Jan. 2001 Units are mg/kg – max of color scale is 1

28. Continental particulate, global map at 1000 hPa, day-5, 1 Jan. 2001 Units are mg/kg – max of color scale is 1

29. Sea salt, zonal average, model levels, day-5, 1 Jan. 2001 Units are mg/kg – max of color scale is 1

30. Desert Dust, zonal average, model levels, day-5, 1 Jan. 2001 Units are mg/kg – max of color scale is 1

31. Variance for total aerosol mixing ratio, zonal average, January 2001

32. Variance, global averages, January 2001

33. Horizontal correlations, level 1000 hPa, 200 km, January 2001

34. Horizontal correlations, level 1000 hPa, 1000 km, January 2001