92
ERDC TR-05-1
marsh grass communities illustrates this dilemma found in all high spatial resolu-
tion digital imagery. Homogeneity with respect to species composition may not
mean homogeneity with respect to biomass distribution and foliar density. For
example, S. patens can exist as uniform mats composed of both live and dead
stems, as well as widely spaced clumps. Both densities of S. patens would be
classified as salt meadow hay from ground surveys. However, the spectral re-
sponse of a pixel within the high-density S. patens is significantly different from
a pixel within the low-density S. patens. The spectral mixture of the low-density
S. patens is easily confused with low-density stands of the other marsh grass spe-
cies.
The best solution to classify spectrally similar species is to collect large
numbers of ground truth plots, stratified primarily by composition and density.
Additional variables needed for stratification to ensure a representative sample
may include background materials, namely soils and water quality (i.e., clarity).
From a practical standpoint, a comprehensive vegetation inventory over a small
study area, perhaps no greater than few hundred acres, could supply the stratified
sample needed for a quality image classification. However, for an area as large as
the Refuge, such a sample is relatively impractical to provide enough useful
ground truth data for a useable thematic map from high-resolution imagery. A
more practical approach may be to divide the Refuge into a set of smaller areas,
or tiles, concentrating ground truth data collection within project specific man-
agement or study sites.
The implementation of high resolution airborne digital remotely sensed im-
agery, both multispectral and hyperspectral, is advancing with the enhancements
to both the sensor systems and the computer hardware and software needed for
data processing. The obstacles encountered in this applied research effort signifi-
cantly affected the final product, in terms of both the accuracy of the thematic
map and the time required to complete the project. As these technologies im-
prove, it will be important for Corps personnel to understand the basic principles
of how the imagery are collected, how the imagery are post-processed, and how
the final map product is created.
Previous Page
