Geoscience Reference
In-Depth Information
Table 9.3
Error Matrix Comparing Aerial Photo Interpretation and 1973 Digital Land-Cover
Classification, with Producer's and User's Accuracy by Class
1973
Land-Cover
Class
Reference (Aerial Photo Interpretation Class)
1
2
3
4
5
6
7
8
9
10
Grand total
1
19
1
0
0
0
0
0
0
0
0
20
2
1
33
0
3
0
0
0
0
0
0
37
3
0
1
16
1
0
2
0
0
0
0
20
4
0
0
13
92
21
0
0
0
1
1
128
5
0
0
14
11
96
0
0
0
0
1
122
6
0
0
3
0
2
15
0
0
0
0
20
7
0
0
3
0
7
1
10
0
0
1
22
8
0
0
0
2
5
0
0
13
0
0
20
9
0
0
4
3
6
0
1
0
3
3
20
10
0
0
0
2
15
0
0
0
1
2
20
Grand Total
20
35
53
114
152
18
11
13
5
8
429
Producer's
Accuracy
(%)
User's
Accuracy
(%)
1973 Map
Total
Photointerpreter
Total
Number
Correct
Land-Cover Class
1. Forest
20
20
19
95
95
2. Woodland Oak
37
35
33
94
89
3. Woodland Mesquite
20
53
16
30
80
4. Grassland
128
114
92
81
72
5. Desertscrub
122
152
96
63
79
6. Riparian Forest
20
18
15
83
75
7. Agriculture
22
11
10
91
45
8. Urban
20
13
13
100
65
9. Water
20
5
3
60
15
10. Barren
20
8
2
25
10
Total
429
429
299
Note:
Overall accuracy = 70%; Tau = 0.66; Cohen's Kappa (Khat) = 0.62; standard error = 0.027.
and barren classes, which showed substantially less than average accuracies in both years. The
water class in 1973 had very low accuracies of 25% (producer's) and 10% (user's) and could not
be assessed for 1986.
9.4.2
Digital Orthophoto Quadrangle Method
Accuracy assessment results are summarized in Table 9.5. Overall accuracy was about 75%,
with Khat of 0.70 and Tau of 0.72. The producer's accuracy was 100% for four classes (forest,
urban, water, and barren), indicating that all pixels examined in the DOQQs for these classes were
correctly labeled in the 1992 map. The user's accuracy was also high for forest and water classes
but was substantially less for urban and barren classes at 44 and 55%, respectively. Accuracies of
mesquite woodland and grassland classes were lower than those for other classes.
9.4.3
Airborne Videography Method
Overall 1997 map accuracy was 72%, with Khat of 65% and Tau of 68% (Table 9.6). A detailed
examination of results by cover class shows substantial variability in classification accuracy, with
producer's accuracies ranging from 54 to 100% and user's accuracies from 13 to 100%. For most
classes the two measures were roughly comparable and fell within the range of 60 to 90%.
Exceptions were the mesquite woodland class with accuracies around 50% and agriculture and
barren classes with relatively high producer's accuracies (71 to 100%) but lower user's accuracies
(13 to 21%).