In: Wagner W., Szekely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B
433
and visible red bands and positive correlations among the
visible bands because the spectral characteristics of land use are
such that as the vigour or tone of the feature increases, the red
reflectance diminishes and the near-infrared reflectance
increases.
4.2 Landuse Classes
Table 4, 5, and 6 contain summaries of the results accuracy
assessment generated from the three images on the five land
use. The overall, user as well as producer accuracies of
individual classes were consistently high for the three
imageries. The accuracies for Landsat, NigeriaSat-1 and SPOT
were 66.5%, 81.2% and 82.8% respectively. The Kappa
statistics were 0.87, 0.97 and 0.89 respectively. The user and
producer accuracies of individual classes in Landsat ranged
from 51.9% to 87.3%, whereas in NigeriaSat-1, they varied
from 60% to 97.3% and in SPOT from 74.7% to 89%. On
Landsat image for the ‘built-up area’ category of land use, the
producer accuracy is 82.8% and the user is 87.3%. This means
that more than 80% of the built-up area in the image was
correctly defined and mapped. The other four land use classes
i.e. bare rocks, farm land; secondary forest regrowth forest and
water body had relatively low producer and user accuracies. For
instance secondary forest regrowth has a producer accuracy of
66.5% and user accuracy of 51.9%. The four categories are thus
not as well defined as the built-up areas. The ‘built-up area’ of
NigeriaSat-1 had the highest accuracy on producer accuracy
(97.3%), it also had user accuracy of 96.9%. On the SPOT
image, the statistics on built-up area category are 81.98% for
producer accuracy and 91.58% for user accuracy.
It appears that Landsat has lower value for accuracies than
either SPOT or NigeriaSat-1. Different reasons may be
responsible for this outcome. One reason may have to do with
the intrinsic characteristics of the images. For instance Landsat
has a spatial resolution of 30metres, NigeriaSat-1 32 metres and
the SPOT lOmetres. Chen, et al., (2004) has shown that these
can variously have effect on the levels of accuracies obtained
from the images. Another reason could be that, spectral
characteristics among the different land cover types (e.g. built-
up, bare rock) are similar, while spectral variation within the
same land cover type or even within the same image might be
high (Cushine, 1987).
TABLE. 4 ACCURACY ASSESSMENT OF LANDSAT TM IMAGERY
Sat
ellit
Cla
ssifi
Reference Data
Re
f.
Cl
Nu
mb
PA
%
UA%
Kappa
e
Ima
ge
ed
Data
Bu
ilt-
up
Ar
%
Ba
re
Ro
ck
%
Fa
rm
lan
d
%
Se
c.
Fo
res
t
%
Wa
ter-
bod
y
%
To
tal
To
tal
Cor
rect
%
Lan
Buil
82.
5.6
4.6
1.1
0.6
10
94.
82.
82.
87.
1.00
dsat
t-up
Area
79
0
7
5
7
0
88
79
79
26
00
Bar
8.2
58.
12.
8.2
4.5
10
91.
58.
58.
63.
1.00
e
rock
6
84
13
1
2
0
96
84
84
98
00
Far
7.0
19.
52.
12.
0.9
10
91.
52.
52.
56.
1.00
mlan
d
2
14
15
29
0
0
50
15
15
99
00
Sec.
1.7
9.4
29.
66.
21.
10
128.
66.
66.
51.
0.86
fores
t re
grow
th
9
1
21
49
14
0
04
49
49
93
01
Wat
0.1
8.0
1.8
h.
72.
10
94.
72.
72.
76.
1.00
er
body
5
2
4
87
76
0
64
76
76
89
00
Overall Landsat Classification Accuracy = 66.47% (i.e. 82.79+58.84+52.15+66.49+72.76 ),
Overall Kappa Statistics = 0.8714
501.02
TABLE. 5 ACCURACY ASSESSMENT OF NIGERIA SAT-1 IMAGERY
TABLE . 6 ACCURACY ASSESSMENT OF SPOT IMAGERY
Land Cover Types
Built-up Area
Bare rock
Farm land
Sec forest regrowth
Water body
10 e Id 20 30
Fig2: Classified print of the Landsat TM image