High Spatial Resolution Digital Imagery
67
were used for these data. The algorithm calculates a separability statistic for all
pair-wise combinations of signatures.
The final signatures used for classification for the southern (7 June) and the
northern (30 June) mosaics, respectively, were generated by a meticulous and
iterative evaluation of the class statistics for each individual training site, which
identified these mean signatures as the most representative of the spectral prop-
erties within the northern and southern mosaics. During this evaluation the sepa-
rability analysis was repeated many times. The output of each iteration identified
the statistically indistinct training sites. Non-separable classes were either drop-
ped from the data set if they were determined to be outliers, or merged with other
training sites. The results of the last transformed divergence tests suggested that
all of the pair-wise comparisons were at least moderately separable. However,
because this analysis was based on an extremely small sample size (i.e., total
number of pixels in each training class relative to the total number of pixels in
the mosaics), it was expected that an inspection of the final class map would
reveal some overlap between relative similar vegetation classes. Figures 39 and
40 display the spectral curves of the 37 class signatures used for the southern mo-
saic. Note the overlap of many of the signatures, particularly in the marsh grass
communities.
Figure 39. Spectral plots of low reflectance signatures used to classify
the southern mosaic.
Previous Page
