849
Table 1. Portions of overbuilt and vegetated areas and mean values of the radiation temperature of different
urban-land-cover-types in % of the total test area; area delineation and calculation by Karin Fera.
Test area
area
in ha
over
built
area
vege
tated
area
buil
dings
traf
fic
areas
other
overb.
areas
trees
bushes
mea
dows
mean
radiat.
temper.
1
Medieval inner city
8,8
88
12
56
21
11
5
1
6
26,03°C
2
Renaissance inner city
8,3
95
5
52
33
10
3
0
2
25,29°C
3
Old suburb buildings
1,6
73
27
42
11
20
11
3
13
25,57°C
4
House-blocks end 19th
century
3,6
69
31
43
4
22
15
3
13
24,82°C
5
Better single-family-
house area
7,9
33
67
14
10
9
21
11
35
22,90°C
6
House-blocks 1939/40
12,0
46
54
22
11
13
14
5
35
21,42°C
7
Simple single-family-
house area
5,1
35
65
17
10
8
24
8
33
21,35°C
8
New suburban house-
blocks area
17,4
34
66
12
14
8
5
2
59
22,66°C
9
Old village kernel with
new single-family-houses
7,9
30
70
14
6
10
18
6
46
23,16°C
only very few rules of thumb, no procedures and it is
mostly left to the scientist's knowledge and luck to
find the right mixture of these three things to get
the desired results.
3 RADIATION TEMPERATURE DISTRIBUTIONS OF DIFFERENT
URBAN-LAND-COVER-TYPES
Pure land use classes covering large areas like forest
or fields are easy to delineate in thermal images
because of their homogeneneous temperature distribu
tions. Not so are the manyfold mixed structures in
urban environment. To find out something about these
complex classes nine test areas representing nine
types of urban-land-cover were selected. With the help
of largescale infrared aerial photographs the areas
of the pure cover classes overbuilt areas (subdivided
into buildings, traffic areas and other overbuilt
areas) and vegetated areas (subdivided into trees,
bushes and meadows or grass) were delineated and mea
sured. The results are listed in table 1. On the other
hand the same test areas were delineated in the thermal
image and radiation-temperature-value-histograms
aggregated into one degree temperature classes (30
classes because the sensor was calibrated between
10 and 40° C) and the arithmetic mean values were
calculated (table 1).
Applying cluster analysis (SPSS-X procedure Cluster
and Quick Cluster) to group the nine test areas into
urban structure types using three variable groups
(temperature-value-histograms, percentage overbuilt
and vegetated area, percentages of the six pure cover
classes) no corresponding classification could be
found (see table 2). This fact is due for the most
part to the effects of shadows, different roof materi
als and differing viewing angles. To the variable
groups "percentage overbuilt vegetated area" and
"percentage of pure cover classes" the three error
reasons are not applicable. The differences in clus
ter membership between these two variable groups are
due to the varying percentages of the pure classes
in the vegetated areas. In this example the level of
generalization of the 30 temperature classes in the
histogramm data, which were strongly influenced by the
three error sources mentioned above, was not adequate
to answer the question if it was possible to explain
the overbuilt - vegetation proportion of urban-land-
cover-types by the radiation temperature distribution
of the same test area. Also the method (the cluster
procedure) was not optimal.
A better method and a better level of generalization
to show such a relation is to use the arithmetic
mean value of every test area and to correlate it
with the proportions of the land-cover-classes. The
results of this procedure are listed in table 3. We
see that the proportions of overbuilt and vegetated
Table 2. Clustermembership of the urban-land-cover-
types named in table 1, using different variable sets
for clustering into 4 groups.
No. of
temperature
% overbuilt
% of six
ur-la-
value-class
and vegeta-
pure land-
co-type
histogramms
ted area
cover types*
1
1
1
1
2
1
1
1
3
1
2
2
4
1
2
2
5
2
3
3
6
3
4
3
7
3
3
3
8
4
3
4
9
2
3
4
* buildings, traffic areas, other overbuilt areas,
trees, bushes, meadows.
Table 3. Correlation between mean radiation tempera
ture values of the 9 test areas and the 8 land-cover-
class percentages (see table 1).
Land-cover-class
corr.
coeff.
R2
signifi
cance
Overbuilt area
0,86
0, 74
0,00145
Vegetated area
-0,86
0,74
0,00145
Buildings
0,86
0,75
0,00132
Traffic areas
Other overbuilt
0,41
0, 17
0,13853
areas
0,49
0,24
0,09243
Trees
-0,53
0,28
0,07036
Bushes
-0,60
0,36
0,04349
Meadows
-0,72
0,52
0,01412
areas in the test regions are highly correlated with
the mean radiation temperature values. The six pure
cover classes are one generalization step too fine
to explain the mean temperature values in an optimal
way. So we see that the optimal description parameters
of the selected testareas and the affiliated level of
spatial generalization are the mean radiation tempera
ture values and the percentages of the rough urban-
land-cover-types "overbuilt" and "vegetated".
