For the 2000 LCLUmap, stratified randomsamplingwas used and 30
samples were selected for each of the six semi-automatically classified
classes (strata), except for the Built class. Given the importance of the
Built class for urbanization and population analyses, 60 samples were
selected for the accuracy assessment. The accuracy of the classes that
had beenmapped through visual interpretation andmanual digitization
was not assessed, as thesewere visually interpreted and verified before
being digitized. Mapped areas within the four-region study area that
were selected for accuracy assessment were limited to: (1) patches of
2000 LCLU where 5 × 5 blocks of pixels (150 × 150 m) contained the
same class and represented consistent LCLU and (2) areas that
contained cloud-free 2000 SPOT 10 m (10 scenes) and 5 m (one
scene) panchromatic imagery to be used for visual reference for the accuracy
assessment. To determine which portions of the SPOT imagery
contained clouds and could not be utilized as reference data, bright
clouds were identified using object-based image classification methods
with eCognition software and thin clouds not picked up using
eCognition were manually identified and delineated.