Accessibility to higher resolution earth observation satellites suggests an improvement in the
potential for fine scale image classification. In this comparative study, imagery from three optical
satellites (WorldView-2, Pléiades and RapidEye) were used to extract primary land cover classes
from a pixel-based classification principle in a suburban area. Following a systematic working
procedure, manual segmentation and vegetation indices were applied to generate smaller subsets to
in turn develop sets of ISODATA unsupervised classification maps. With the focus on the land cover
classification differences detected between the sensors at spectral level, the validation of accuracies
and their relevance for fine scale classification in the built-up environment domain were examined.
If an overview of an urban area is required, RapidEye will provide an above average (0.69 κ) result
with the built-up class sufficiently extracted. The higher resolution sensors such as WorldView-2
and Pléiades in comparison delivered finer scale accuracy at pixel and parcel level with high
correlation and accuracy levels (0.65-0.71 κ) achieved from these two independent classifications.