EM 1110-2-2907
1 October 2003
properties in the vegetation. This further helped to discriminate among the vegetation
classes.
(2) HyVista pre-processed the digital data. Pre-processing included a smoothing algo-
rithm to reduce the signal to noise ratio (SNR) across the scene, to an impressive >500:1.
The data were geographically rectified using ground control points identified on a geo-reg-
istered USGS orthophoto. The geo-positional accuracy was determined to be within 3 pix-
els across 95% of the scene. This was established by comparing the image with a high-
resolution orthophoto. A digital orthophoto was then over laid on top of the digital hyper-
spectral data to verify geo-positional accuracy.
e. Study Results. Analyst used KHAT (Congalton, 1991), a classification statistic used to
test the results of supervised versus unsupervised classification (Equation 6-1). KHAT con-
siders both omission and commission errors. Statistically it is "a measure of the difference
between the actual agreement between reference data and the results of classification, and
the chance agreement between the reference data and a random classifier" (see
KHAT values usually range from 0 to 1. Zero indicates the classification is not better than a
random assignment of pixels; one indicates that the classification maintains a 100% im-
provement from a random assignment. KHAT values equaled 0.69 in this study, well within
the 0.6 to 0.8 range that describes the class designation to be "very good" (≥ 0.8 is "excel-
lent"). For this study, KHAT indicated good vegetative mapping results with the supervised
classification for distinguishing plant species and for mapping surface water vegetation. The
KHAT also verified the potential value of image classification to map submerged aquatic
vegetation using HIS data.
observed accuracy - chance agreement
k=
1 - chance agreement
(6-1)
r
r
N S xii - S ( xi + x + i )
i =1
i =1
r
k = N 2 - S ( xi + x + i )
i =1
where
r
=
number of rows in the error matrix
xii
=
number of observations in row i and column i (the diagonal)
xi+
=
total observations of row i
x+i
=
total observations of column i
N
=
total of observations in the matrix .
The estimated time savings of the mapping project as compared with the manual analysis
using color infrared was calculated to be a factor of 10 or better. Additional benefits include
a digital baseline for change detection and managing restoration. The study did not establish
under which conditions HSI did not work. HSI processing and analyses was shown to be a
generally valuable tool in a large-scale riparian restoration.
6-4
Previous Page
