Two input maps were used for comparing the FSM sequential implementation
to the HK implementation in Berry et al. (1994) and to the FSM parallel
implementation on the CM-5. The first map is
a 
ERDAS/Lan map (FIRE) containing fire patterns from
the Southern Yellowstone National Park. The FIRE map is composed of
ten map classes each with different pixel densities, number of clusters,
and cluster sizes. Table 5
lists the map class statistics of
the FIRE map which was supplied by Dr. Robert Gardner
of the Oak Ridge National Laboratory, Environmental Science Division
(Hargrove et al., 1995).
The second map is a 
ERDAS/Lan map (FORD) containing
pre-classification values, obtained from applying a
classification function on four or five of the inputs from thematic
mapper imagery (Ford et al., 1994). The original inputs
are sensor readings (wavelengths) from different parts of the
electro-magnetic spectrum. The
pixel values in the input map are arbitrary values that represent map class
membership.
The values are logically significant, but not numerically. A data value of 8 is
not necessarily closer to a data value of 9 than it is to any other value (0-31).
Table 6 lists the map class statistics
for the FORD map, which was supplied by Dr. Ray Ford from the Department
of Computer Science at the University of Montana and the
Wildlife Spatial Analysis Laboratory directed by Dr. Roland Redmond
at the University of Montana.
Table 5: Statistics for the 9 map classes of the FIRE map (
pixels).
Table 6: Statistics for the 31 map classes of the FORD map
(
pixels).