Geoscience Reference
In-Depth Information
1000
0
500
PD
MPS
PSCOV
800
−
20
400
600
−
40
300
400
−
60
200
200
−
80
100
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
150
150
−
15
PSSD
ED
MSI
−
20
100
120
−
25
50
90
−
30
0
60
−
35
−
50
30
−
40
−
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
500
0
20
AWMSI
MPFD
AWMPFD
400
−
2
15
300
−
4
10
200
−
6
5
100
−
8
0
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
1000
30
MNN
MPI
%LAND
800
20
20
600
10
10
400
0
0
200
−
10
−
10
0
−
20
−
200
−
20
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 15.4
The relative errors of 12 selected landscape indices for forest class (y-axis) against the overall
accuracy (x-axis).
15.5 CONCLUSIONS
The uncertainties or errors associated with landscape indices vary in their responses to image
data classifications. Also, the existing statistical methods for assessing classification accuracy have
different controls relative to the uncertainties or errors of landscape indices. Assessing accuracy of
landscape indices requires combined knowledge of the overall accuracy (means of user's accuracy
and producer's accuracy) and the REA (differences between user's accuracy and producer's accu-
racy). To reliably characterize landscape conditions using landscape indices, our results indicate it
is necessary to use maps with high overall accuracy and low absolute REA. The selections of
landscape indices are also important because different landscape indices have different sensitivities
to image data classifications. Based on commonly achievable levels of classification accuracy, the
magnitudes of errors associated with landscape indices can be higher than the values of landscape
indices. Comparisons between different thematic maps should consider these errors. Assuming that
the distribution of errors identified by the error matrix is representative of the misclassifications
across the area of interest, the total land area of different class categories can be revised with REA
and the errors of this landscape index can be lowered. Revised values of %LAND should be used
when quantifying landscape conditions.
