897 
Figure 2. Continued 
classified as lawn areas but were primarily drier 
grasses. There was same class confusion between 
lawns and vegetation/soil mixes (sparse cover) and 
healthy vegetation (long, thick grasses). Lawns were 
also often associated with mixed pixels, either with 
inert materials in residential areas or with trees 
and shrubs in parks, cemeteries, or landscaped areas. 
Healthy, frequently watered vegetation was 
characterized on the signature plots by a steep rise 
between channels 3 and 4 and a substantial drop in 
channel 5, as shown in Figure 2(k). Oi the 
discriminant function scatter plot (Figure 1) this 
category was very high on the "greenness axis." 
Class 40 showed the most vigorous vegetative response 
of the spectral classes in CLUS67. Most of class 40 
involves healthy alfalfa fields in the agricultural 
areas, although some healthy lawn areas were combined 
into this class. Classes 20 and 21 also represented 
alfalfa fields of varying plant densities, while 
class 46 usually indicated corn fields. Class 18 was 
on the borderline between very thick and healthy 
grasses (often found in lawns or golf course roughs) 
and the slightly drier crops found in some 
agricultural fields. 
Trees and shrubs had a very similar spectral 
pattern to both lawns and moist vegetation, except 
that the response in the near infrared (channel 4) 
was usually not as high. Since trees are not as 
large as the TM sensor's IFOV (30 meters), there was 
usually some mixing between tree canopies and the 
understory materials, with both contributing to the 
pixel's response. Often, the density of tree cover 
was difficult to observe. Aerial photography that is 
slightly off nadir will show oblique views of trees, 
which are in turn hiding other surface cover 
materials, making tree canopies appear as the 
predominant land cover. With a completely vertical 
view it became apparent that tree cover density in 
the urban setting is actually quite a small 
percentage, with other cover materials contributing a 
major proportion of an individual pixel's response. 
For this reason, the land cover in virtually all of 
the tree and shrub category was mixed. However, 
trees with shrubs or dense weedy materials were major 
contributors to the spectral response. Classes 53 
and 60 were the most representative signatures in 
this category for densely wooded tree cover. Class 
19 was a borderline class between trees and other 
healthy vegetation. It often represented areas where 
grass was showing through the trees, as in city 
cemeteries or parks. Class 19 also represented 
clumps of shrubby trees and marshy weeds. Classes 26 
and 62 were primarily located in residential areas 
and most often represented treelined streets or back 
yards with large trees. Class 62 contained a 
slightly higher proportion of inert material than the 
other classes in this category. 
3 USE OF THERMAL CHANNEL SIX DATA IN RECLASSIFICATION 
In the past, very little use has been made of the 
Thematic Mappers thermal band (channel 6) in land 
cover analysis due to its coarser resolution (120 
meters) and low range of spectral variation. This is 
unfortunate, since the two parameters most 
responsible for variability of surface temperatures 
are surface moistness (moisture availability) and 
diurnal heat capacity (Carlson & Boland 1978). These 
two factors are highly related to the nature of 
surficial materials in the urban setting. 
It was observed in this study that many of the land 
cover categories that were being confused in 
multispectral classification were actually very 
different in terms of thermal properties. For 
example, coal and asphalt were classified 
interchangeably as water, and cropped agricultural 
fields were often confused with residential areas or 
natural grass. For this reason, TM channel 6 was 
used as an ancillary data layer to set thermal