The Version 2 of bi-monthly composites of GLOBCOVER surface reflectances at 300m resolution is available for downloading : 10 mosaïcs cover the period from May 2005 to April 2006 (example below).

Color composite of spectral MERIS surface reflectances over Valencia area, Spain

The surface reflectance is obtained from the MERIS level 1 images with a series of preprocessing steps outlined in the following scheme.

GLOBCOVER Preprocessing chain

Geometry . The geometric corrections are done using the AMORGOS tool developed by ACRI. AMORGOS uses in input MERIS level 1 images, restituted satellite attitude and orbit ephemeris files, and a Digital Terrain Model at 30" resolution, and produces in output corrected latitude, longitude and altitude for each pixel. The projection tool has been developed by the POSTEL/Medias-France team and the Cnes. The orthorectified images in output of AMORGOS and of the Projection tools have been shown to have a geolocation relative accuracy of 52 m RMS and absolute accuracy of 77 m RMS (see also the Figure below). These performances are seen as satisfactory and permit to use the MERIS images at their full resolution of 300 m.

MERIS orthorectified images geolocation accuracy

Atmospheric correction . The atmospheric correction permits to transform Top of Atmosphere radiances into surface reflectances. The effects of Rayleigh and aerosol scattering and gaseous absorption are taken into account. The algorithm uses a neural network relying on the so-called MOMO method (Fischer and Grassl, 1991) which has been selected because it was validated in the framework of another ESA project (ESA Albedo Map). The aerosol correction uses in input so far an aerosol optical depth product at 1° resolution derived from MODIS at 8 day frequency. For this Version 2, the aerosol correction is performed using a monthly aerosol optical depth product at 1 km resolution produced using 2005 MERIS-Reduced Resolution data with an algorithm developed by Brockmann Consult. This new method decreases significantly the "blue haze" effect (image below). The gaseous absorption correction uses an ozone field from ECMWF, and oxygen and H2O fields derived from the MERIS data (ratios B11/B10 and B15/B14 for O2 and H2O respectively).

Multispectral bimonthly commposite over Spain, May-June 2005 (Version 1 at left, Version 2 at right) : the "blue shade" effect is significantly reduced.

Cloud screening . Two methods are combined to screen clouds, the first using the MOMO method already quoted, and the second using thresholds of reflectance on the bands at 443, 753, 760 and 865 nm. The results of the cloud screening have been validated using ground truth data from the synoptic network of meteorological stations operated by Meteo-France over Europe and Africa, and a Cloud Toolbox developed by NOVELTIS and CNES. For this Version 2, an improved cloud screening method is used. Cloud top height is estimated for a better determination of shadows. The snow reflectance is kept at its TOA level.

BRDF Correction and Time Compositing . Two methods are combined to remove the directional effects due to variations of Sun and view geometry in the successive measurements of surface reflectances. The Mean Composite method (Vancutsem et al., 2007) is processed as a first reference time series of surface spectral reflectance over a compositing period of 51 days. Then the CYCLOPES method of Hagolle et al. (2004) is used to screen this data set and remove spurious data affected by undetected sources of noise (residual thin clouds, aerosols, shadows, ...). The time compositing is then an average of the time series of surface reflectances corrected for directional effects over various periods depending on the needs : 15 days, 2 months, or one year.