TOTELY
1mfield,
ia, USA
, containing
ty of spatial
zed aircraft
lor infrared
refinement
xed much of
‘the surface
improve the
s additional
vork will be
Virginia are
- factors and
ted by deep
llis drainage
ob - Seneca
surrounding
ities change
'scribes the
rough oak-
rs). Recent
research by
; related the
oh available
'woods and
and maple)
followed by
'S) and pure
rs). Oaks
drier slopes,
t the higher
|, pines may
alds on the
ed in some
ain Institute,
organization
ultures, has
Except for
undisturbed
y red spruce
oods on the
d just north
Red pine
be found in
ing of open
west. What
st clearings
1051 meters
(3448 feet) to 1287 meters (4223 feet) and various aspects are
provided by the slopes on either side of Big Run, along the sides
of the ridges on the west and by hollows north and south of The
Mountain Institute on the east side.
DATA
To explore the effect of spatial and spectral characteristics on the
ability to differentiate forest communities, a number of data types
will be examined in the overall project. For these preliminary
results we have concentrated on two of the most widely used data
products. A Landsat Thematic Mapper (TM) scene acquired on
July 17, 1993, with six visible, near infrared and mid infrared
bands at 30 meter spatial resolution covered the study site and
was of excellent quality. In the same month color infrared aerial
photography was flown over the area at a scale of 1:12,000 by the
US Forest Service. A frame covering the intensive study site was
digitized using color filters on a Nikon AX1200 Flatbed Scanner
at the Marshall University Remote Sensing Applications Center
to provide a three-band (green, red and near infrared) digital
image with 3 meter spatial resolution.
Digital elevation model data for the Spruce Knob 7.5 minute
topographic quadrangle was obtained from the US Geological
Survey (USGS).
Ground truth data consisting of tree species, trunk diameter and
precise location were gathered for eight sites of 60 meter diameter
located using Global Positioning System (GPS) in July and
September of 1995. In addition a windshield survey was
performed for all driveable roads within the study area, stopping
at 0.1 mile intervals to identify the canopy tree species and
understory at over 200 locations in September. Two of the
detailed field measurement sites and 16 of the windshield sites lie
within the intensive study site. Ten observations of more distant
tree stands have been added to expand the coverage within this
subset to more community types.
ANALYSIS TECHNIQUES
Analysis was performed using the PCI EASI/PACE Image
Analysis Software in the Laboratory for Terrestrial Physics (LTP)
Computer Facility at the NASA Goddard Space Flight Center.
TM data for the Spruce Knob quadrangle were geographically
referenced to the aircraft image using 12 ground control points
with a residual error of less than 0.5 pixels. The DEM elevation
data were registered to the same base using the same transform
equation. Data layers for aspect and slope were derived from the
elevation data and also registered to the aircraft base.
Unsupervised classifications were then performed on the digitized
color infrared aircraft image, the TM image (six bands) for the
corresponding area and a three-band version of the TM image,
containing those bands which most closely resemble color
infrared photography (Bands 2, 3 and 4). Layers for the elevation
and aspect data were added separately to the aircraft and TM (six
bands) images and unsupervised classifications were again
performed. The parameters (allowed numbers of clusters,
merging and splitting criteria and number of iterations) remained
the same for all classifications, resulting in from 56 to 71 clusters.
These clusters were then assigned to 5 or more forest types,
Pasture or road, based on spectral signature, geographic location,
247
proximity to other similar classes and appearance on the aerial
photograph.
The PCI software allowed another interesting approach to
exploring multilayered data sets termed "Modeling". A new
image could be created by assigning a value to the combination
(logical and) of a range of values in two or more bands (e.g., red
spruce clusters 1-5 and elevation 1000-1050 meters = 25 in the
new image). The classified aircraft and TM images were mapped
to a series of 12 and 14 classes, respectively, so that forest types
which were a little different spectrally could be treated separately.
The elevation and aspect data were divided into ten intervals over
their range in the image (increments of about 20 meters and 36
degrees, respectively). Images were then created combining
classes from the aircraft image classification with elevation and
with aspect, respectively.
The locations of the ground measurements were digitized from
the locations recorded on the Spruce Knob quadrangle using
Global Positioning System (GPS) data for the detailed
measurement sites and the windshield observations. The
positions were transferred to the aircraft image base and samples
of 100 pixels and 1 pixel, respectively, were extracted from the
aerial photograph and TM based classifications, and compared
with the ground truth. The results are discussed below.
RESULTS
Classification based on spectral, spatial and terrain
information
Classification images based on aircraft and TM (six bands) are
presented in Figures 2 and 3. The forest types include
riparian/brushland in light medium grey, two types of hardwood
in white and medium grey, red pine plantations in dark grey, and
red spruce or red spruce and yellow birch in very dark grey. The
pasture adjoining The Mountain Institute is shown in light grey
and portions of the roads are in black.
A number of variations were observed within the hardwood and
riparian/brushland classes which will be investigated in future
field work. In particular, there were two distinctly different
groupings of hardwood signatures. Hardwood 1 (in white)
appears to be the dominant type east of the creek and on the lower
slopes west of the creek. Hardwood 2 (in medium grey) borders
the west of the pasture and occurs as small stands within the
pasture and just north of it and as two large stands on the ridges
west of the creek. These latter areas were harvested but have to
some degree grown back. Since hardwood 1 occurs at lower
elevations, it is possible that it represents primarily maples while
hardwood 2 contains greater proportions of beech along with
maples. It is also possible that hardwood 1 represents more
mature or denser stands than hardwood 2.
The aircraft based classification is, of course, far more detailed
and textured due to its finer spatial resolution. The two hardwood
types are interspersed among each other to a far greater extent
than in the TM classification. The bushes and saplings associated
with the creekbed (riparian/brushland) are mingled with
hardwood. Lines of red spruce can be seen in the red pine
plantations.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
