Geoscience Reference
In-Depth Information
the San Pedro watershed study area, including areas that were intermediate between classes. The
analyst performing the 1992 assessment also reviewed high-resolution color airborne video data
for comparison with the appearance of LC classes in the DOQQs. The video data were acquired
over the watershed in 1995 and vegetation in selected frames at 1:200 scale was identified to species
or species groups (Drake, 2000). Image “chips” were extracted from the DOQQs as an aid to LC
class recognition in the reference data (Maingi et al., 2002).
9.3.4.2
Sample Point Selection
Generation of sample points from LC maps relied on a window majority rule. A window kernel
of 3
3 pixels was moved across each cover class and resulted in selection of a sample point if a
majority of six of the nine pixels belonged to the same class. This ensured that points were extracted
from areas of relatively homogenous LC. A 180-
¥
¥
180-m DOQQ sample size was used to match
the 3
3 pixel map window and a map class was assigned and recorded for the DOQQ sample.
A total of 457 points were sampled to assess the 1992 map.
¥
9.3.5
Airborne Videography
Accuracy assessment of the 1997 LC map was performed using airborne color video data
encoded with GPS time and latitude and longitude coordinates. The video data were acquired on
May 2 through May 5, 1997, and were therefore nearly coincident with the June Landsat TM scene.
There were 11 h of continuously recorded videography of the San Pedro Watershed for the area
north of the U.S.-Mexico border, acquired at a flying height of 600 m above ground level. The
nadir-looking video camera used a motorized 15
zoom lens that was computer controlled to cycle
every 12 sec during acquisition, with a full-zoom view held for 3 sec. The swath width at wide
angle was about 750 m and was approximately 50 m at full zoom. At full zoom, the ground pixel
size was about 7.0 cm and the frame was approximately 1:200-scale when displayed on a 13-inch
monitor. Although the nominal accuracy of the encoded GPS coordinates was only 100 m, ground
sampling revealed that average positional accuracy was closer to 40 m (McClaran et al., 1999;
Drake, 2000). The video footage was acquired by flying north-south transects spaced 5 km apart
and the total flight coverage encompassed a distance of nearly 2000 km.
¥
9.3.5.1
Video and GIS Data Preparation
The encoded GPS time and geographic coordinate data were extracted from the video into a
spreadsheet for each flight line. Coordinate data from the spreadsheets were used to create GIS
point coverages of frames from each flight line. Individual frames of the video data were identified
during viewing by a time display showing hours, minutes, and seconds, in addition to a counter
that numbered the 30 frames recorded per second. The time display information was included as
an attribute to the GIS point coverages, which were inspected for erroneous coordinate or time
data indicated by points that fell off the flight lines or were out of time sequence; such points
were deleted.
9.3.5.2
Video Sample Point Selection
To minimize the likelihood of video sample points falling on boundaries between cover classes,
selection of random sample points along the video flight lines was restricted to relatively homo-
geneous areas within classes. This was accomplished by applying a 3
3 diversity or variety filter
to the 1997 map, which replaced the center pixel in a moving window by the number of different
data file values (cover classes) present in the window. Pixels assigned the value of one therefore
¥
