847
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986
Urban-land-cover-type adequate generalization
of thermal scanner images
Peter Mandl
Institute of Geography, University Klagenfurt, Austria
ABSTRACT: A thermal infrared image showing the urban area of Klagenfurt, Austria on a cloudless day in autumn
1979 is generalized by forming radiometric and spatial classes and by clustering temperature-value-histograms
of these generalized data. The results are plotted in map resembling form and interpreted according to thermal
urban-land-cover-types. The errors appearing in such a generalization are analysed in detail. Some of these com
plex land cover classes are investigated in detail trying to explain their temperature distributions by the per
centages of overbuilt and vegetated areas within these different urban'housing quarters. The results of this
study can be used for urban climate investigations and town planning purposes.
1 INTRODUCTION
There are three points which characterize modern aeri
al remote sensing and which distinguish it from clas
sical aerial photo interpretation. The first point is
the possibility to acquire images in the non-visible
parts of the electromagnetic spectrum like the micro-
wave or the thermal infrared region. The second point
is the sensing of discrete and calibrated measurements
(e.g. by scanner) which allows the drawing of measure
ment maps and is also the base for the third point,
the utilization of digital computers to use methods
of pattern recognition, statistics, etc. for handling
and analysing the huge amounts of data. These features
of modern remote sensing have extended the applications
of this technology in the last two decades.
One field of research which got many impulses by
utilizing these new features of remote sensing (parti
cularly by the introduction of computerenhanced and
analysed thermal infrared images) is micro climatolo
gy, especially urban climatology and the application
of the findings there in town and regional planning.
A very good review on urban climate research is
D.O.Lee (1984). One main problem of these sciences is
to get measurements of climatic parameters covering
the terrain more or less continuously. The classical
method to solve this probletn is to drive measurement
traverses over the test area or to have a fix sensor
network and to interpolate the data between the mea
sured traverses or points (e.g. Oke & Hannell 1970
or Niibler 1979) .
Now remote sensing can provide urban climatology
with continuous terrain covering measurements on radia
tion temperature. Unfortunately this parameter is of
no direct use for climatology and a conversion of
radiation temperatures to surface temperatures is com
plicated because of the differnt error sources that
influence the remotely sensed data. Nevertheless spe
cial processed, enhanced and printed images of the
thermal infrared emission of urban areas have been used
to find out "heat islands" in various towns, fresh air
areas in the surrounding of towns or fresh air canals
into the central towns and to give basic informations
for planning purposes.
The aims of this study now are to show special as
pects of the heat emission field of the urban area of
Klagenfurt, Austria on a cloudless day at 10 o'clock
in the morning in autumn. The primary data for this
were thermal infrared ^mages (8 - 13 ^m wavelength)
acgiired by a Bendix M S 11-channel scanner on the
13 of September 1979. A correction of the radiomet
ric errors in the image which are intern relative er
rors (0,1-0,2 C), errors due to the atmosphere bet
ween the objects and the sensor (minus 1-3,5 C for
the correction from radiation to surface temperature),
errors due to the emission coefficients of different
surface cover types and influences of the different
viewing angles in the image on the thermal emission
distribution of various areas seemed not necessary
for the purposes of this study.
In a first part of this study images of the primary
data, geometrically rectified from the panoramic dis
tortion, sliced into five temperature classes were pro
duced (Fig. 1). The pixels then were stepwise combined
to "macropixels" consisting of 10x10, 25x25, 50x50
and 100x100 "micropixels" by calculating the arith
metic mean value of every macrdpixel. An example of
the resulting images is given in Fig. 2. All the re
sults of this first part of the study are described
in detail in Seger & Mandl 1985.
In the second part of this study which will be des
cribed in this paper three main problems are dealt
with:
1. The determination of the temperature distributi
ons of typical urban-land-cover-types and their rela
tions to the proportions of vegetation, asphalt and
houses in test areas, to find out if thermal data
can be a good indicator for such "pure" land cover
types.
2. An error investigation of different generaliza
tions of the radiometric and the spatial domain of the
thermal image data.
3. The attempt of a régionalisation of the thermal
image using temperature-value-histograms of the macro
pixels as variables and cluster analysis as classifi
cation method.
2 SOME ASPECTS OF THE GENERALIZATION PROBLEM OF
REMOTE SENSING DATA
Digital remote sensing data can be characterized by
four different types of resolution. In this study the
temporal and the spectral resolution of the data is
held constant, a thermal infrared image (9 - 13 /im)
at one acquisition date is used. The spatial resolu
tion of the primary data is 2,3m x 2,3m at the nadir
point and the radiometric resolution is 0,12 C tem
perature intervall from one data value to the other.
The thermal image of Klagenfurt consists of 803 x
3.200 pixels. Each pixel has a precision of 256 pos
sible values which means that the amount of informa
tion in the image is greater than 20,5 million bits.
In order to seperate the user important information
(signal) from the unimportant one (noise) a very well-
considered and well-balanced correspondence between
the objectiv,the degree or level of generalization and
the methods used for this have to be found. There are
