High Spatial Resolution Digital Imagery
61
rability among several signatures suggested that different species associations
with similar plant canopy architectures (e.g., height, density) maintain closely
related vegetative textures. In addition, plant associations with similar amounts of
senescent biomass tended to have similar spectral responses. The following
bullets summarize the observations from the signature separability analyses:
Floating mats of water hyacinth mixed with other species are spectrally
similar to mixed vegetation types without hyacinth.
Tall reeds, such as Phragmites and Typha, are spectrally similar to
shorter species that have senescent biomass, including herbicide treated
water hyacinth.
Sample sites that include both green vegetation and open water are
spectrally similar regardless of the species present.
Dense mats of pure water hyacinth showed the same reflectance
characteristics as dense stands of smartweed.
With nearly 5000 ha (~ 12,000 acres) depicted within the three CAMIS mo-
saics, 36 training sites (i.e., sample plots) provided a relative limited vegetation
characterization.
Two options were available to minimize the negative effects associated with
the inseparability of the vegetation signatures described above. One option was
to aggregate the signatures to create fewer classes. This step should improve the
separability of the training statistics. However, lumping classes together would
reduce the potential detail in the final thematic map. A second option was to em-
ploy all 24 signatures in the maximum likelihood classifier and then combine the
spectrally similar thematic classes. The second option was selected in an attempt
to maintain the greatest level of spatial detail in the final thematic product.
The 24 classes in the supervised classification product were aggregated to a
final 18 classes.
5.2.1.3 Minimum Mapping Unit. A minimum mapping unit (MMU) filtering
routine was also applied to the full resolution class map. This filtering algorithm
first delineates all of the raster polygons throughout the thematic image. A raster
polygon is defined as a group of adjacent (i.e., connected) pixels with the same
thematic class value. The adjacency criteria define the polygon using pixels
joined along the four flat edges of the square pixel and joined at the corners. The
second step is to define an appropriate MMU. For this research effort, a mini-
mum mapping unit of 144 pixels (144 m2) was selected. This size is unusual be-
cause it covers the same two-dimensional area as nine 4- 4-m IKONOS pixels.
The next step is to delete all raster polygons that are less than 144 pixels. This
operation is synonymous with applying a polygon sieve to the thematic image,
Previous Page
