High Spatial Resolution Digital Imagery
63
5.2.2
IKONOS Image
5.2.2.1 Histogram Matching. As stated above, no field data were acquired at
the time of the acquisition of the four-band multispectral, 4-m spatial resolution
IKONOS image. The classification of these data, therefore, had to rely on analyst
interpretation to develop the necessary unique vegetation signatures. An attempt
was made to use the signatures acquired from the CAMIS mosaics. This proce-
dure required that the spectral histograms from the July 2002 IKONOS image be
matched to the December 2001 CAMIS spectral histograms. An exhaustive
analysis of the two sets of four histograms (i.e., blue, green, red, and near infra-
red) was unsuccessful in developing IKONOS bands with 8-bit distributions
similar to those of the CAMIS bands. The primary reason for the failure of this
technique is assumed to be the date of acquisition of each image set. South Flor-
ida does experience a subdued senescence of vegetation during the winter
months. The CAMIS images were captured during this period. The IKONOS sat-
ellite data were collected during high summer, when the vegetation is actively
growing. In addition, water levels within Lake Okeechobee significantly impact
the distribution of vegetation types, particularly the areal extent of water hyacinth
and water lettuce. Water levels had changed between the two acquisition dates.
The differences in plant phenology and species distributions resulted in a very
poor matching of the IKONOS histograms to the CAMIS histograms. The super-
vised classification technique was abandoned.
5.2.2.2 Unsupervised Classification. An unsupervised classification tech-
nique was employed to develop a class map from the IKONOS data. The cluster
techniques applied to the IKONOS imagery over Lake Okeechobee were similar
to those applied to the CAMIS mosaics for the Poplar Island Reference Wetlands
Mapping project. At Poplar Island, the images were first segmented into two
primary data sets using the Normalized Difference Vegetation Index (NDVI).
One segment represented primarily non-vegetated surface features, while the
other segment represented primarily vegetated pixels. Then each segment was
classified using an unsupervised clustering routine within the Imagine software.
Next, the classes from the two independent segments were labeled with vegeta-
tion or non-vegetation types using visual interpretation of the original multis-
pectral CAMIS frames. Finally, the segments were combined and the classes
were logically aggregated to create final land cover maps.
For the IKONOS scene over Lake Okeechobee, an NDVI image was again
used to stratify the multispectral image, but this time into four segments. The
objective of increasing the number of segments was to enhance the separability
of the classes derived from the unsupervised classifications. The first segment
included the lowest NDVI values and thus represented features with the least