.cut River. Photo- 
686-692. 
& J.C. van Huis 
.1 dune manage- 
ITC Journal 
C. Davis 1969. 
'hotograiranetric 
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
Spectral characterization of urban land covers 
from Thematic Mapper data 
iy of recreational 
:higan Academy of 
XVII, 1962, 
lman 1973. 
:ries. Tijdschrift 
rrafie, 64, no 1: 
and) 1977. 
Friese meren. 
onderzoek. 
h to flourishing 
rphotostudy of 
nor peninsula, 
pretation, work- 
rzeewerken 1977. 
ek naar patronen 
evaart op het 
aan de hand van 
>ta nr 291, Lely- 
rzeewerken 1979. 
n van een lucht- 
op de randmeren 
ion geography of 
iew, vol LXII, no 
urist market re- 
erkehr, no 3: 89- 
ion and Resources 
9, part 3s 467- 
s in the uplands. 
York Moors. 
ng Institute, 
onderzoek Hel- 
cten. Afdeling 
hogeschool, 
en vanuit vaste 
vincien Gebied. 
universiteit 
f recreational 
ial photographs. 
ties and their 
of the islands, 
erference. In: 
Flora and vege- 
and coastal 
erdam. 
nvironmental Con- 
c lecture given 
e 20th, 1985 at 
Douglas J.Wheeler 
Utah State University, Logan, USA 
ABSTRACT: Using Salt Lake City, Utah, as a test case, this study evaluates the capabilities of Landsat~~b 
Thematic Mapper (TM) digital data for distinguishing urban land cover materials. This was accempfShed by 
using a newly developed hierarchical clustering algorithm which statistically derived spectral classes from TM 
channels 2, 3, 4, and 5 (visible, near infrared and middle infrared). The relationships between spectral 
groups were further analyzed using three statistical evaluations: principal components analysis, cluster 
analysis, and discriminant analysis. Through the use of component scores, cluster linkage diagrams, and 
canonical discriminant function scatter plots; as well as TM spectral curves, aerial photography, and ground 
investigation; the spectral classes were grouped into twelve predetermined land cover categories. The accuracy 
of classification was assessed at approximately 80 percent (0.05 significance level). A significant 
improvement in classification accuracy (91.5 percent) was achieved by stratifying the multispectral 
classification with thresholds from the TM thermal channel introduced as ancillary data. 
INTRODUCTION—DERIVING SPECTRAL CLASSES FROM TM 
With a high proportion of the world's population 
living in cities it is increasingly important to 
understand the very complex ecological interactions 
that are taking place within the urban environment. 
One key to better understanding these relationships 
is to characterize the land cover materials that 
influence radiational and micro-climatological 
balances. By monitoring successional changes in land 
cover materials one may observe its effect on urban 
ecosystem processes. 
The value of Landsat's multispectral scanner (MSS) 
in detecting land cover has been established (Todd 
1978). Landsat 5's advanced multispectral scanner 
called Thematic Mapper (TM) , with improved spatial 
and spectral resolution over MSS, could be an even 
better tool for detecting land cover characteristics 
in complex urban environments. Although very little 
is published at the present time on the use to TM 
data in urban analysis, preliminary studies using 
Thematic Mapper Simulator (IMS) data flown from 
aircraft indicated that the 30 meter pixel size of TM 
might be ideal for mapping urban land cover elements 
(Clark 1980, Welch 1982). Two studies using TM data 
for Mobile, Alabama, show promising results 
(Quattrochi 1983, Wang 1985). 
The urban region of Salt Lake County, Utah, was 
chosen as the study area for this evaluation of TM 
digital data due to the diversity of land activity 
found within a relatively small area. The majority 
of Salt Lake County's 1984 population of 650,000 is 
found within a 15 kilometer wide strip stretching 
north-south along the base of the Wasatch Mountains. 
Salt Lake City contains the diversity of land 
activities usually associated with metropolitan areas 
of a much larger size. There are many heavy and 
light industrial and commercial activities; diverse 
multifamily, single-family, and rural residential 
areas; extensive irrigated and nonirrigated 
agricultural lands; as well as natural vegetation and 
water features distributed throughout the county. 
The central valley portion of the county consists of 
several urban areas that are coalescing due to 
ongoing urban development and suburbanization. Even 
within the city limits of the communities in the Salt 
Lake valley there are many agricultural and natural 
areas that are being filled in with urban land use 
activities. 
A strip 15 kilometers wide and 25 kilometers long 
was selected for the study area, extending through 
the urban corridor of the valley. From within this 
urban study area, 10 test windows were selected to be 
used for generating spectral signatures from the raw 
TM data. The principal data source for analysis wa c 
the digital tape of a Thematic Mapper scene dated 
July 27, 1984. Using test windows to break up the 
study area into manageable size units made the field 
investigation and computer processing more cost 
effective. The 10 windows were carefully chosen from 
aerial photography to represent the broadest spectrun 
of cover materials found within the Salt Lake urban 
environment. The 10 test windows accounted for 
approximately 16 percent of the study area. 
After the test windows were selected, several trips 
to the field were made to identify land cover 
materials. At that time, ocular estimates were made 
of the percentages of surficial materials found in 
association with one another, and which combinations 
comprised various cover types. It was decided that 
12 particular land cover classes would be desirable 
to detect from the TM data of Salt Lake City. The 12 
preliminary land cover classes decided upon for this 
study included: (1) open water, (2) light inert 
materials (e.g., bare soil, concrete, and reflective 
metals and glass), (3) coal and slag, (4) dark inert 
materials (e.g., railroads and blacktop surfaces), 
(5) light asphalt-gravel surfaces, (6) mixed pixels 
with mostly inert cover and little vegetation cover, 
(7) mixed pixels with high vegetation cover, (8) 
senesced weeds and natural grass, (9) healthy moist 
vegetation, (10) drier or sparse vegetation with soil 
showing, (11) short cropped grasses (lawns), and (12) 
trees and shrubs. 
A large portion of the computer processing of TM 
data was accomplished using ELAS digital image 
processing software obtained from NASA's Earth 
Resources Laboratory. The TM data tape was 
reformatted into an ELAS data file format to be 
processed on a Prime 400 computer. Of the seven 
original TM channels it was decided to generate 
statistic files from data in spectral bands 2, 3, 4, 
and 5, giving representation from the visible, near 
infrared, and middle infrared wavelengths. The 
thermal channel (band 6) was not used in determining 
spectral classes because of its lower spatial 
resolution (120 meters compared to 30 meters) and the 
lack of differentiation in spectral values. While 
there appear to be many opinions on which TM bands 
are optimum for processing, there is considerable 
893