52
ERDC TR-05-1
zones, and vegetation. Therefore, the end user has the flexibility to assign each of
the 16 interpreted classes into one of the three general land cover types. Of par-
ticular concern are the transition classes that will define the boundaries between
1) persistent water and intertidal zones, and 2) sparsely vegetated intertidal areas
and non-tidal vegetated marshland. Field personnel familiar with each site will
likely be needed in the end to accurately determine which groups of the 16
classes to collapse to produce three classes. Note that each of the individual mo-
saics was classified independently.
The class names were assigned based on the spectral response (i.e., cluster
spectral signature statistics) of each cluster relative to its statistically adjacent
clusters. This was a subjective process. Therefore, the end user must ultimately
assign the most accurate cluster names during the recoding of the 16 classes to
the three primary land cover types.
5.1.1
NDVI Mask for Poplar Island Restoration
Figure 31 shows an example of the NDVI image for the Knapps Narrows
site. The NDVI value for each pixel is calculated using the following equation,
which is the same equation used earlier on AISA data:
CB(4) - CB(3)
NDVI =
CB(4) + CB(3)
where
CB(4) = the infrared CAMIS band (800 nm)
CB(3) = the red CAMIS band (650 nm).
The NDVI operation outputs a single raster layer. Therefore, the pixel values
are displayed using a grayscale (black and white) lookup table. The unique value
assigned to each pixel ranges from 1 to +1. Values approaching 1 indicate un-
healthy or dead vegetation. As live vegetative biomass increases, the index value
also increases to a maximum of +1. Figure 29 also shows the resulting histogram
of pixel values derived from the NDVI image for Knapps Narrows. The NDVI
image for each reference wetland was split into two segments by visually inter-
preting the NDVI histogram, the NDVI image, and the original multispectral mo-
saic. The first image segment contained all non-vegetation areas, depicted with
darker pixels; the second contained the brighter vegetated pixels. This subjective
distinction was not completely accurate, as pixels near the center of the NDVI
histogram (i.e., near the threshold value) represented mixed pixels, or pixels dis-
playing spectral properties of both vegetated and non-vegetated features. Leaf-off
Previous Page
