values for each of SPOT’s three 
spectral bands) to 64 habitat classes. 
Preprocessing entails applying an 
unsupervised maximum likelihood 
clustering algorithm to the data thus 
reducing the SPOT imagery from 8 
bit to 6 bit. Reflectance values 
cluster out within a three 
dimensional array defined by 
SPOT’s three spectral bands. The 
algorithm reduces the data to 64 
clusters and assigns each a unique 
value. During the final pass, every 
pixel is then given the value of the 
cluster within which it is most likely to 
fall (Lillesand and Kiefer 1987). 
GAIA software served three major 
purposes in this study: accessing 
the satellite imagery; measuring 
distance to mainland and to yhe 
nearest larger island for each island; 
and exporting the imagery as a PICT 
II file. These exported PICT II images 
were analyzed with Image software 
package (Rasband 1990). I used 
Image's particle analysis routines to 
measure the perimeter and area of 
each island as well as to tally the 
islands’ spectral classes. 
Explanation of Terms: Area. 
Perimeter, and Convolutedness. 
All data extracted from the imagery 
was in pixel units and hence are 
accurate to the 20 meter resolution of 
the SPOT MS imagery. The SPOT 
data used in this study was acquired 
at high tide so all values for area and 
perimeter represent only the 
terrestrial portions of the islands. 
In an effort to measure an island’s 
convolutedness, I derived an index 
to indicate the degree to which a 
given island’s perimeter varies from 
a circle’s of equal area. The shape 
that yields the smallest perimeter for 
a given area is a circle and can be 
expressed: 
P'=2t (1) 
V 7t 
where P’ is the smallest possible 
perimeter and A is the island’s area. 
The index of convolutedness (C) is 
simply the ratio between an island’s 
actual perimeter and its theoretical 
least perimeter: 
c-e. (2) 
p’ 
where P is the island’s perimeter, P’ 
is its least possible perimeter and C 
is the index of convolutedness. The 
more irregular an island, the higher 
its index of convolutedness. 
Landscape Richness and Diversity 
I used the total number of spectral 
classes present on an island as a 
measure of its landscape richness. 
As mentioned above, I assumed that 
an island with relatively high spectral 
class richness were also islands with 
high landscape or habitat richness. 
I then inserted landscape richness 
values into the Shannon-Weaver 
index of diversity to derive an 
island’s overall landscape diversity 
(Shannon and Weaver 1949). 
Typically this index is used to 
measure "species" diversity (Peet 
1975). Diversity (H) is calculated by 
the equation: 
H’ = - S Pi In pj (3) 
where p/ is the proportion of species 
/in a sample of s species. As the 
number of species in a system 
increases, especially if the relative 
proportions of those species are 
uniform, H will tend to be high. I 
calculated landscape diversity for 
each island of the 423 by substituting 
pixel tallies of each spectral class for 
species in the above model. In this 
way H will tend to increase when the 
number of spectral classes or 
habitats on an island is high and the 
proportions of those habitats are 
uniform. 
I calculated maximum theoretical 
diversity (Hmax) for given habitat 
richness (s) to isolate those islands 
with uneven habitat distribution. 
Maximum diversity is defined as the 
Shannon-Weaver index resulting 
from perfectly even distribution of 
pixels in spectral classes. In such a 
situation, all p/ are equal and: 
where p equals the total number of 
any given habitat class (because 
they are uniformly represented) and 
s is the total number of classes 
present. Substituting this equation 
into the Shannon-Weaver equation 
leaves: 
Hmax= • l n (^-) 
44