66
+
+ + +
10 16 20
1
notion scatter
equentially
n image display
ity or
there is a
rials, so
play very similar
sses contained
e than one cover
eonfusion caused
ielding similar
s of the
luster diagrams,
ial photography,
ry was best
oteristics of
^tral classes the
jure 2 by
s. Ground
arficial
spectral
э, represents the
Dttom) , and light
concrete, metal,
er conditions
ions in both of
49, 2, and 48
al component
st of the light
represented
a small amount
ing. Classes 52,
drop between
between channels
nany of the flat
fell as saline or
agetation.
are soil with the
Ef, and were
з of dry farms
Ltion, usually
<rere mixed
aleted concrete
3es 50 and 42
construction
Many of these
m trailer parks
scraped land,
jht colored soils
sand covered
Water was expected to have a low spectral response
in all four channels, especially in band 5, the
water-absorption band. There was difficulty,
however, in distinguishing between water and other
dark cover categories by four channel mutlispectral
classification alone. Class 37 displayed a typical
pattern for open water and was found in lakes, sewage
treatment ponds, and in other deep ponds. Class 54,
though, was comprised of equal proportions of open
water and coal piles. By training on individual
pixels in both water and coal cover categories it was
noted that the spectral responses frcm the visible
and near infrared bands were essentially identical,
while the middle infrared channel (band 5) showed a
slightly lower response for the water. This,
however, did not present enough variability to keep
the two classes from being merged in the hierarchical
clustering process. Classes 64, 28, and 27 also
showed a high propensity for water within their
spectral classes. The spectral curve for class 43
displayed a rather odd shape for water, yet it
characterized the Jordan surplus canal with it silty
water and its mix with embankment materials.
The very dark materials in Figure 2(b) were
primarily considered to be open coal or slag piles,
with occasional sites of tar or very black asphalt.
As mentioned earlier, there was considerable
difficulty in clearly distinguishing these cover
types from water surfaces by spectral characteristics
alone. Other methods were later tested to better
differentiate these cover types. Class 54
represented the darkest coal and slag piles in a
smelter location. Lighter colored materials were
represented by classes 64 and 28 and were found in
scattered coal piles throughout the industrial areas
of Salt Lake.
Asphalt surfaces cover much of the urban Salt Lake
area and are good indicators of cannercial and
transportation land uses. A distinction was made
between light and dark asphalt surfaces since they
often represent different land use activities or have
different effects on the urban environment. The dark
inert materials cover type mostly included blacktop
areas and dark material mixed with soil (such as
railroad yards). Classes 12, 38, and 27 represented
this cover category although class 27 also
characterized some water bodies. The other classes
were also occasionally confused with water in the
surplus canal. The light asphalt surfaces were
generally asphalt mixed with gravel to form roads and
parking lots. Several of the pixels in these classes
also represented roofing materials in the commercial
areas composed of tar and gravel. Classes 58 and 10
especially characterized these areas devoid of
vegetation. Classes 59 and 5, on the other hand, had
a higher proportion of vegetation cover mixed in, as
evidenced by the flatter slopes between channels 4
and 5 in their spectral curves shown in Figure 2(d).
Classes 15 and 35 primarily represented parking lots
and road networks. Although class 63 was also mostly
transportation, there was a little mix with natural
grass areas.
There were large portions of the study area with
mixed pixel responses. This generally occurred in
residential areas where a large variety of
heterogeneous surface materials were spaced very
closely together, usually into an area smaller than
the spatial resolution of a pixel. The mixed
responses from surfaces such as lawns, concrete,
asphalt shingles, trees, metals, etc. made the
hybridized signature curves displayed in Figure 2 (e-
f). While these classes occupied virtually the same
region on the discriminant function scatter diagram
as other non-mixed land cover categories, they
reflected distinct differences in the shapes of their
spectral curves. This was observed by comparing
Figure 2 (e-f) with Figure 2 (g-i). Class 56 had a
very bright response and was found in many trailer
courts. The major contributor to this response was
the shiny roofs of trailers with sparse lawns and
asphalt mixed in. This class also expressed some
confusion with agricultural fields having bare ground
and stubble remaining. The shape of the spectral
curve for class 36 was almost identical to class 56
yet it was slightly darker in reflectance. Class 36
was also primarily shingle roofs with small mixes of
lawns and trees. This class was found among light
roofed condominium complexes and was also
occasionally confused with stubble fields. Classes 8
and 34 were found in condominium complexes and other
high density residential areas fringing the central
business district (CBD). There were three distinct
cropped fields, however, that were also identified as
class 8. Class 45 was not as commonly associated
with residential areas. This class more often
represented the mixed pixels where asphalt borders on
natural grass areas (e.g., along freeways, railroad
tracks, and in cortmercial or industrial areas) .
Other mixed pixel locations showed a higher
percentage of vegetation contributing to the
spectral response. Classes 65 and 4 represented
surface materials that were approximately 50 percent
covered by vegetation (usually lawns and trees).
Class 1 contained about 60-70 percent vegetation
cover, while classes 14 and were generally over 75
percent healthy vegetation. These mixed pixels with
high vegetation components were again found primarily
in residential locations where surface cover was very
heterogeneous within a small area. These mixed
classes also showed a very large within-class
spectral variance (over 6.5 times the mean variance
for all remaining land cover classes).
A large portion of the study area was not developed
and was covered by senesced annual grasses and weeds.
A wide variety of species and cover densities were
grouped within a few signatures for this category.
Figure 2(g-h) illustrates two distinct patterns in
these natural grasslands. Classes 44, 47, 13, and
30 displayed a pattern of healthier vegetation, as
indicated by the steepness in curves between channels
3 and 4. These were usually weedy fields or vacant
lots where the plants were not completely senesced.
In classes 13 and 30 the soil was often wet,
contributing to the darkness of the response. There
was seme confusion between this cover category and
fields in areas where crops were newly planted and
the soil was still contributing a major portion of
the response. Classes 33 and 22 still had a
considerable amount of vegetative response and were
quite reflective, containing a high proportion of
light soil. Classes 51 and 57 represented darker
colored soils in natural, undeveloped areas, while
classes 25, 7, and 11 were often found as idle fields
in dry farm or irrigated areas.
Cropped or sparsely watered agricultural fields
represented in Figure 2(i), were typified by mixed
soil and vegetation responses. The surface materials
contributing to these spectral curves mostly included
cropped fields with seme stubble and sane new growth
showing through; newly planted or young crops;
pasture areas, irrigated or subirrigated, with weeds
or bare patches; and occasionally short cropped
grasses with same soil showing through. Since this
category was a blend of the vegetation and soil
responses there was some confusion with other similar
cover categories. Pixels representing these spectral
classes were often found in areas classed as lawns,
natural grasses, or in residential areas. Classes
23, 55, and 67 were usually correctly identified as
pastures or cropped fields, while classes 24 and 9
were frequently confused with sparse lawns.
Distinguishing lawns from other vegetative surface
materials was also rather difficult. Various lawns
were quite different from each other in terms of
moisture content, amount of thatch, shortness of the
grass, and grass vigor. Fairways of golf courses and
school playfields were most characteristic of the
lawn cover type. Classes 66 and 17 from Figure 2(j)
showed the short cropped, but healthy grass found at
parks, playgrounds, and many golf courses. Class 39
identified lawns that often had trees or shrubs
nearby. Several other spectral classes were