COMPARISON OF MERIS-BASED LAND COVER PRODUCTS OVER THE ARABIAN PENINSULA

Accurate high-resolution spatially-distributed land cover data is required for parameterizing a wide range of applications in many scientific disciplines ranging in scale from global circulation models to watershed scale soil erosion models. Many remote sensing-based land cover products at the global, continental, regional and national scales have been produced using different remotely-sensed data, classification systems, spatial resolution, and classification algorithms. The MERIS-based GlobCover 2009 and GlobCover 2005-2006 global land cover products are the most detailed with 22 land cover classes defined with the UN Land Cover Classification System (LCCS) at a fine spatial resolution of 300-m. The objective of this paper is to compare the agreement between these widely used MERIS-based global land cover products over the Arabian Peninsula. Both areal and per-pixel comparisons including overall accuracy, Kappa statistics and individual class accuracy were calculated. The results shows that: (1) overall agreement between the two products over the Arabian Peninsula is 99 % with Kappa statistics of 82%; (2) Individual class accuracy ranges from 16% for open broadleaved deciduous forest class to 100% for bare areas class; (3) More than 96% of the Arabian Peninsula surface is barren; (4) Differences within individual classes ranges from 188% for the mosaic vegetation/ Croplands class to -80% for the open broadleaved deciduous forest class.