Geoscience Reference
In-Depth Information
Table 10.1 Accuracy
of agricultural and
nonagriculture land cover
classifications
Agricultural
Overall accuracy
(230/261)
88.1226%
Kappa coefficient
0.8561
N
¼
261
Nonagriculture
Overall accuracy
(294/331)
88.8218%
Kappa coefficient
0.8615
N
¼
331
Table 10.2 Error matrix for agricultural and non-agricultural classifications
Agriculture classification
Class
Bare soil Green ag
Forest/shrub Water Wet veg
Early ag
Total
Bare soil
86.21
0
0
0
0
0
19.16
Green ag
0
79.25
5.56
0
0
11.11
19.16
Forest/shrub
5.17
20.75
94.44
0
0
3.7
19.16
Water
0
0
0
93.75
0
0
11.49
wet veg
0
0
0
6.25
100
0
11.49
Early ag
8.62
0
0
0
0
85.19
19.54
Total
100
100
100
100
100
100
100
Non-agriculture classification
Class
Grass
Wet veg
Forest/shrub
Impervious Water
Bare soil
Total
Grass
94.17
5.26
2.56
4.88
0
0
30.82
Wet veg
0
76.32
2.56
0
0
0
9.06
Forest/shrub
5.83
18.42
94.87
0
0
0
15.11
Impervious
0
0
0
85.37
0
23.33
14.8
Water
0
0
0
0
100
0
15.11
Bare soil
0
0
0
9.76
0
76.67
15.11
Total
100
100
100
100
100
100
100
and impervious. Bare soil was classified as impervious in 23% of instances and
impervious classified as bare soil in about 10% of instances. Confusion between
vegetation types (i.e. forest, grass, wet vegetation) accounted for 46% of the error in
the nonagriculture classes. This illustrates the necessity for separately classifying
agriculture from non agriculture to reduce errors between these classes.
Although the above method for land cover classification produced an accurate
classification by remote sensing standards, several errors were evident that would
likely affect the accuracy of later analyses. Most misclassifications were traced
back to the preprocessing of the data when the raw satellite data was clipped by
CAUV status. Although CAUV was a consistent method of defining agriculture
across the three counties in the study area (as opposed to zoning status), it did not
guarantee that all agriculture parcels were included in the clip. Consequently, the
most obvious errors were the misclassification of agricultural fields as non-
agricultural bare soil. Any large areas classified as bare soil were examined against
the DOQ and misclassifications were reclassified. Errors in the other classes were
not as frequent, but when noticed, these were reclassified as well. As reclassification
was conducted post-accuracy assessment using high resolution the DOQs the final
classification was more accurate than initial accuracy assessments suggest.
