However, comparison of thematic maps
based only on such a spectrally centred approach does not
make full use of the additional information on small-scale
variability in the higher spatial resolution data. Such variability
could allow discrimination of spectrally similar
habitats (such as macroalgal, seagrass, and coral communities
with similar photosynthetic pigments) that differ in
their homogeneity. For example, seagrass habitats tend to be
internally relatively homogeneous compared to much patchier
coral reef habitats. To make fuller use of the additional
textural information in the IKONOS imagery, which has
around 25–50 pixels for each SPOT XS or Landsat TM
pixel, a 5 by 5 pixel variance filter was passed over each of
the three IKONOS depth-invariant bands (bands 1/2, bands
1/3, and bands 2/3) to create three textural layers for use in
supervised classification. To avoid creating artefacts around
edges of masked out areas such as land and clouds, the filter
was designed to ignore zero values. The area operated on by
the filter was chosen to coincide with the 20-m mean
boundary spacing between fine-level habitats in the area
(Mumby et al., 1999). Similar texture algorithms were also
created and implemented for CASI (2020 m area) and
Landsat TM data (9090 m area). After classification, a
single contextual decision rule was applied to areas of
fringing reef in which pixels classified as seagrass habitats
(classes 12 and 13 in Table 1) were converted to the
spectrally similar habitat class 2 (Table 1).