Table
2. SPOT Accuracy
Assessment
Error
Matrix for Land use and
SPOT CLASS IFIC ATI0M
Land
Cover
Mapping
Ground Wl
Tru th
W2
w3
Ul
U2
CFl
MF 1
CPI
Al
A 2
FI
F2
F3
F4
F5
Total
Omission
Accuracy
UI
1ST -
0
3
—¡3—
0“
“13—
—a -
0
a—
d
0
u
0
0
161
5.0
95.0
W2
65
0
0
0
0
0
0
0
0
0
0
0
0
0
65
0.0
100.0
W3
0
~TT
29
0
0
0
0
0
0
0
0
0
0
0
0
29
0.0
100.0
Ul
0
0
U
44
1
0
11
0
0
0
2
1
0
0
0
59
25.4
58.7
U2
0
0
0
u
33
2
0
1
0
0
0
0
0
0
0
36
0.9
86.8
CFl
0
0
0
0
U
225
0
0
0
2
0
0
0
0
0
227
1.1
88.2
NF1
0
0
0
4
0
42
0
0
0
0
0
0
0
0
50
16.0
53.2
CPI
0
0
0
0
0
2
0
104
9
15
5
0
0
5
0
140
25.7
65.4
Al
0
0
0
0
0
0
0
O’
76
0
0
0
0
4
2
32
7.3
77.6
A2
0
0
0
0
0
20
0
0
T
148
0
0
0
0
0
158
11.9
80.0
FI
0
0
0
3
l
0
15
2
0
~U
139
6
0
1
0
167
12.0
79.9
F2
0
0
0
5
0
0
3
0
0
0
-a
570
6
1
3
538
3.1
94.7
F3
0
0
0
0
0
0
0
0
0
0
0
5
223
0
1
230
3.0
94.5
F4
0
0
0
4
0
0
0
16
1
0
0
1
~U
210
4
236
10.2
83.3
F5
0
0
0
0
0
0
0
0
6
0
0
0
0
— 5
188
199
5.5
90.0
Total
153
65
37
50
35
253
71
123
92
165
146
584
229
226
198
2437
Commi ss ion
0
0
0
27.2
12.3
36.4
58.0
13.
5 19.5
10.1
4.07
2.4
2.6
6.8
5.0
%
Overall Accuracy (2249;-2437 ) = 92.3%
classification (disregarding human intervention errors), individual
classes should be inspected for agreement with the ground truth data and
the extent of the commissions determined.
The deep to very deep water (Wl); moderately deep to deep water (W2);
very shallow to shallow water (W3), sparse urban and barren, rocky land
(U2); alluvial cultivated and eroded land (A2); sparse to moderately
dense caatinga forest (F2); moderately dense to dense caatinga forest
(F3); dense caatinga forest on undulating land (F4) and dense mixed
caatinga forest on hills (F5) classes produced the best results in terms
of high percentage agreement with ground truth data (80.0-100.0%) and
relatively low percent commissions (0.0-12.3%). This shows that digital
data of these classes are spectrally homogeneous and can be discriminated
readily from other classes. Other classes such as dense urban area (U1),
mixed cultivated and fallow land (NF1), and mixed cultivated and pasture
(CPI) are more spectrally heterogeneous and explain the lower overall
classification accuracies and high commission percentages.
The cotton cultivated and fallow land (CF1) and alluvial land with dense
shrubs/trees (Al) classes also yielded a high agreement relative to the
ground truth data (88.2% and 77.6%). The corresponding percentage
commissions, however, were higher at 36.4 % and 19.5%, respectively.
Most commission errors for the cotton cultivated and fallow land and for
alluvial land with dense shrubs/trees were due to confusion with the
alluvial cultivated eroded land and the mixed cultivated and pasture
classes. This was due to the spectral variability and sometimes
similarity among these classes.
The lowest classification agreement was for the dense urban area (Ul)
(58.7%), mixed cultivated and fallow land (MF1) (53.2%), mixed cultivated
and pasture (CPI) (65.4%), and for the sparse caatinga forestland (FI)
(79.9%) classes. The percent commissions for sparse caatinga forest and
rocky land was quite low (4.1%), whereas the percent omission was
relatively high (12.0 %). The errors of commission for dense urban area
354