Geoscience Reference
In-Depth Information
1600
−
50
500
PD
MPS
PSCOV
1400
−
60
400
1200
1000
−
70
300
800
−
80
200
600
400
90
−
100
200
0
−
100
0
82
84
86
88
90
92
94
96
82
84
86
88
90
92
94
96
82
84
86
88
90
92
94
96
−
30
150
−
15
PSSD
ED
MSI
40
−
−
20
120
−
50
−
25
90
−
60
−
30
−
70
60
−
35
−
80
30
−
40
−
90
−
100
0
−
45
82
84
86
88
90
92
94
96
82
84
86
88
90
92
94
96
82
84
86
88
90
92
94
96
140
−
2
10
AWMSI
MPFD
AWMPFD
120
−
3
8
100
−
4
80
6
60
−
5
40
4
−
6
20
−
7
2
0
82
84
86
88
90
92
94
96
82
84
86
88
90
92
94
96
82
84
86
88
90
92
94
96
0
200
5
MNN
MPI
SDI
−
10
150
0
−
20
100
−
5
−
30
50
−
10
−
40
0
−
50
-50
−
15
82
84
86
88
90
92
94
96
82
84
86
88
90
92
94
96
82
84
86
88
90
92
94
96
Figure 15.3
The relative errors of 12 selected landscape indices for the landscape (y-axis) against the overall
accuracy (x-axis).
(the greater overall accuracy, the smaller REA). However, overall accuracy and REA explained
some aspects of classification errors but did not explain other possible sources of classification
errors (e.g., the spatial distributions of misclassifications). Therefore, these accuracy measures
alone were not adequate to assess the accuracy of the MNN and MPI, which have particularly
strong spatial features.
The variations of landscape index errors were different among different landscape indices. For
example, the errors of MPDF, AWMPFD, and SDI at the landscape level were within a range of
10%, whereas the errors of PD, PSCOV, ED, AWMSI, and MPI for entire landscapes or forest
patches exceeded 100%. The former group of landscape indices was not as sensitive to image data
classification and the errors of these landscape indices were not controlled by classification accuracy
measures. Landscape indices in this group were unreliable despite the image classification accuracy
values. The latter group of landscape indices was sensitive to image data classifications, and
therefore a small difference in classification accuracy resulted in a large difference in landscape
index values. In this case, classification accuracy was always superior when accuracy-sensitive
landscape indices were used. Intermediate indices exhibited intermediate sensitivity to image data
classifications. The rule of higher overall accuracy and smaller absolute values of REA was
particularly applicable to this intermediate group. Further systematic studies are needed to determine
which landscape index belongs to these sensitive groups.