s “treads are still
4 23
ng”. Older was
zone, ruts visible,
Trace was defined
tation is growing
don, Ft. Bliss). In
re coincident with
these, 76 were
ler, and 555 were
s were generated
se points.
uring the week of
n two objectives:
aneuver areas and
of use, and 2) to
nt in the satellite
es were met with
Bliss staff.
llowing steps:
> of two criteria:
or presence of a
pact was roughly
3liss staff. Sites
ry five-level scale
impact, moderate
rmediate classes.
' features were
rared (CIR) and
7 derived from
lune 1992. The
especially useful
rrain which was
features in the
road extent in the
ical land cover.
using GPS.
essional Global
provided by the
) the field sites.
to 8 satellites,
after differential
| site a GPS file
-processing. The
cond; a minimum
| for each point.
on the 1:50,000
(DMA) maps of
ie June 1992 TM
3) Describe the site.
Written descriptions made at each site included
information on the dominant vegetation cover, soil
characteristics (e.g., color and texture), and other
aspects of the site. Impacts such as lack of
interdunal vegetation and tracks were also noted.
Photographs of the area around the sites were
taken from a high vantage point (e.g., on top of a
dune or car hood) for future reference.
4. PREPROCESSING
4.1. Image Rectification
The 4 June 1992 and 29 October 1990
Landsat scenes and the 6 October 1990 SPOT
scene were rectified by digitizing ground control
points from 7.5 minute quads. Two Landsat
scenes, dated 19 September 1993 and 28 October
1984, were rectified using image to image
registration. In both cases the master image used
was the rectified 4 June 1992 Landsat scene. An
affine transformation was used to rectify the
images to Universal Tranverse Mercator (UTM)
Zone 13, North American Datum 1927 (NAD27)
map projection. The Landsat images were
resampled to 30 meter pixels, while the SPOT
images were resampled to 20 meter pixels. The
nearest neighbor algorithm was used to preserve
the spectral integrity of the data. The root mean
square error was less than one pixel for all of the
image rectifications performed.
4.2. Spectral Calibration
Comparison of multi-temporal imagery, as
in change detection studies, requires that the
images be radiometrically calibrated. Linear
spectral calibration reduces undesirable changes in
sun angle, atmospheric scattering, and sensor drift
(Eckhardt ef al., 1990; Coppin and Bauer, 1994).
The process has two major requirements. First,
one image must serve as a reference image to
which all others will be calibrated. Second, bright
and dark targets must be identified which are not
influenced by phenological or other natural
changes, such as differences in moisture. The
reference image was arbitrarily chosen as the 19
September 1993 Landsat scene. The goal of this
processing was to derive slope and intercept values
from calibration targets so that the 28 October
1984 and the 29 October 1990 Landsat scenes
would more closely match the digital numbers of
the reference scene. The images were visually
surveyed, along with various maps of the area, for
features which could serve as calibration targets.
Very bright and dark features influenced primarily
by atmospheric scattering only were chosen as
calibration targets.
An industrial facility called Smeltertown in
southeastern El Paso was selected as the dark
127
target. A firing range pad on the eastern edge of
MA 8 and an area of bright wind blown sand
across the US-Mexico border, southwest of El
Paso were chosen as bright targets.
Spectral signatures were extracted from
the calibration targets. The signatures for each of
these targets were plotted and examined to ensure
that their shape did not vary erratically between all
three dates. Radical changes in shape would
indicate that the target was not stable over time
and could not be used in this analysis; however, all
the targets met this criterion.
For each band, the mean digital number
for each target was plotted for the signatures
obtained from the 1994 imagery against the those
obtained from the image being calibrated. The
slopes and intercepts were found by linear
regression. The slopes and intercepts were used to
linearly transform each band of the other Landsat
images so that new spectrally calibrated imagery
was obtained.
The rectified and calibrated satellite
imagery were then subsetted to cover only the
study area. The subsets were rectangular with a
northwest UTM coordinate of 3,566,486N,
363,728E and a southeast UTM coordinate of
3,548,426N, 385,418E.
4.3. GPS Processing
Given the resolution of the SPOT and TM
imagery, differential processing was required to
obtain sub-pixel accuracy of the field locations.
Unprocessed GPS files yield an approximate
locational accuracy of 200 meters. Differential
correction improves this accuracy to within 2-5
meters. An Arc/Info point coverage was generated
from these corrected coordinates.
5. IMAGE PROCESSING
This study was broken down into three
sub-projects: detection, monitoring and
classification. The first sub-project was an image
interpretation effort to identify how various
training impacts appear on satellite imagery as
well as aerial photos. The second sub-project, was
a change detection investigation, using multi-
temporal satellite imagery of the area. This would
provide information on the ability of these systems
to monitor change due to training impacts. The
third sub-project was a correlation of a classified
image with ground-based impact data. This was
an initial investigation into quantifying levels of
impact.