Geoscience Reference
In-Depth Information
Classified Data Sets (Plate 19.1)
POINT_ID
SUBPIXEL_IS%
REFERENCE_IS%
24944
100
7
24945
20
11
24946
20
0
24947
20
0
24948
40
39
24949
40
30
24950
40
8
24951
40
43
24952
40
27
24953
30
0
24954
90
63
24955
90
82
25093
40
28
25094
90
36
25095
40
16
25096
50
0
25097
100
45
25098
50
31
25099
30
4
25100
100
34
25101
90
31
25102
50
5
25103
90
56
25104
90
93
Note:
The continuous REFERENCE_IS% field data
are subsequently “rounded up” to coincide with
the Mid-Atlantic RESAC classification structure
(Table 19.1). Data from the SUBPIXEL_IS%
and REFERENCE_IS% fields are utilized in the
accuracy assessment.
for each of the six sample replicates. From the six replicates we computed the mean and coefficient
of variation (defined as the standard deviation divided by the mean, expressed as a percentage) at
both the pixel and whole-area measures of accuracy for each of the unit areas.
Additionally, per Ji and Jensen (1999), we performed a per-pixel rank Spearman correlation
test between the subpixel estimates and the reference data for all cells in the Dead Run watershed
(no smaller unit areas were processed for the rank test).
19.3 RESULTS AND DISCUSSION
an accuracy assessment of the Mid-Atlantic RESAC
subpixel classification, but rather a discussion of alternative assessment methodologies that may
be more compatible with the characteristics of subpixel classified data.
The classified data utilized for this study were in preliminary form and were not meant for
external distribution for use in watershed assessments. Although we did not quantitatively assess
the registration displacement of the two datasets, a manual review showed an approximate and
unsystematic 1-m difference throughout the Dead Run subwatershed.
Our results indicated that the pixel-based methods of determining the accuracy of the subpixel
estimates yielded results that were consistently lower than the whole-area method of determining
accuracy (Table 19.3 and Table 19.4). The whole-area estimate of impervious surface percentage
We wish to stress that this study is not
