LandCover CCI Pre-processing
Kirches, Grit1; Arino, Olivier2; Brockmann, Carsten1; Defourny, Pierre3; Boettcher, Martin1; Bontemps, Sophie3; Danne, Olaf1; Kalogirou, Vasileios2; Lamarche, Celine3; Paperin, Michael1; Radoux, Julien3; Smollich, Susan1; Zuehlke, Marco1
1Brockmann Consult GmbH, GERMANY; 2ESA, ITALY; 3UCL, BELGIUM

The Land Cover CCI (LC CCI) project aims at generating a multi-sensor global land cover ECV dataset, using MERIS FRS, MERIS RR, SPOT-VGT and ASAR instruments. It will produce three combined land cover products for the years 2000, 2005 and 2010. The project builds upon the state-of-the-art technology established during the GlobCover Project. The main Land Cover processing chain is characterised by its pre-processing and classification parts. The completed automated pre-processing chain generates global 7-day composites of surface reflectances using MERIS and SPOT-VGT in a series of pre-processing steps. These steps include geometric correction, pixel identification, atmospheric correction with aerosol retrieval as well as compositing and mosaicing. Improved algorithms have been developed and validated for these pre-processing steps. So a new method of pixel identification is being used within LC CCI, particularly in relation to cloud detection. As already demonstrated through the GlobCover experiences, two main issues are encountered when working with MERIS FR as the main input data source, namely, the lack of acquisitions and the lack of SWIR bands. As a result, multi-sensor approaches which use available MERIS RR and SPOT-VGT time series, are integrated in the processing chain for the LC CCI products. The occasional temporal and spatial gaps in the coverage of MERIS FR data is complemented with MERIS RR products which have a pixel size of 1.2 km at nadir. The merging of the 300m MERIS FR pixels with lower resolution RR pixels (limited to some problematic areas) downgrades the geometric resolution. In order to minimize this negative impact, the merging has to ensure that the positioning accuracy of the RR and FR pixels (i.e. the matching in space) does not have any systematic error. Only the blurring effect of the coarser resolution is allowed to remain. These points are two of the major issues in the LC CCI project when considering matters related to the pre-processing of the satellite data.
The amount of input data that have to be processed and classified in less than one year is 200 TB, which means that the available time period for the pre-processing is 6 months. The volume of input data in itself poses specific challenges to the project. An accompanying quality management has also been established for the pre-processing. This includes the quality assessment of the input products as well as of the global 7-day surface reflectance products. In particular the quality assessment of the input data requires an automatic as well as a time expensive visual checks.
The algorithms developed and the experiences gathered from the LC CCI project can make a significant contribution to the pre-processing chain for land cover applications being developed within the scope of future satellite missions like Sentinel 2 and Sentinel 3.