Geoscience Reference
In-Depth Information
Table 9.2
Upper San Pedro Watershed Land-Cover
Classes: Absolute and Relative Areas;
Representative Values from 1997 Land-
Cover Classification
Proportion of
Total Area (%)
Land-Cover Class
Area (ha)
Grassland
263,475
36
Desertscrub
229,571
31
Woodland Mesquite
101,559
14
Woodland Oak
90,540
12
Urban
16,562
2
Agriculture
14,530
2
Riparian
9,217
1
Forest
7,193
1
Barren
6,814
1
Water
417
<0.1
Total
739,878
100
180-m homogeneous areas on the map. Background, clouds, and
cloud-shadowed pixels were excluded to prevent the selection of pixels that fell at the edge of the
map, within openings in clouds, or in cloud-shadowed areas where the adjacent cover classes were
not known.
Video flight line coverages were overlaid on the map of homogeneous cover, and a subset of
frames falling on homogeneous areas (
represented centers of 180-
¥
=
18,104). The map class under each subset frame was added as an attribute to the “candidate frames”
GIS point coverage for stratification purposes.
n
= 4,567) was drawn from all study area frames (
n
9.3.5.3
Random Frame Selection and Evaluation
Video sample points were drawn randomly from the homogeneous subset, stratified by map
class area, and were distributed throughout the Arizona portion of the study area. The water class
was excluded from analysis for lack of adequate reference data (
= 6) and was not presented in
the final error matrix. A surplus of approximately 15% over the calculated minimum number of
frames needed for each cover class was selected. The videography interpreter was provided with
spreadsheet records containing the videotape library identifier, latitude, and longitude for each
sample frame, along with GPS time for frame location on the tape. A cover class was assigned to
each sample point and recorded in the spreadsheet.
Although the accuracy of video frame interpretation was not assessed in this study, it is expected
to be very high. Drake (1996) reported that LC identification of similar airborne videography at
the more detailed biotic community level averaged 80% accuracy after only 3 h of interpreter
training. The interpreter for this study had substantial prior experience in both video frame inter-
pretation and ground sampling for videography accuracy assessment in this region.
n
9.4 RESULTS
9.4.1
Aerial Photography Method
Results of accuracy assessment are presented in Table 9.3 (1973) and Table 9.4 (1986). Overall
map accuracies were similar at 70% for 1973 and 68% for 1986. Khat and Tau statistics were also
similar at 0.62 and 0.59 (Khat), and 0.66 and 0.65 (Tau) for 1973 and 1986, respectively. The user's
and producer's accuracies were similar to overall accuracy for all except the mesquite woodland
