Geoscience Reference
In-Depth Information
Galiuro Mts.
Upper San Pedro Watershed
Winchester Mts.
Redington
N
W
E
Rincon Mts.
Little Dragoon Mts.
S
Benson
St.David
Dragoon Mts.
Whetstone Mts.
Flagstaff
Tombstone
ARIZONA
B
a
b
o
co
m
a
ri
R
i
v
er
Phoenix
Charleston
Ft Huachuca
Sierra Vista
Hereford
Mule Mts.
Upper
San Pedro
Watershed
Tucson
Canelo Hills
Huachuca Mts.
United States
Mexico
Sierra San Jose
10
0
10
20
30 Kilometers
Sierra Los Ajos
Cananea
Sierra Mariquita
Figure 9.1
Location of the upper San Pedro River watershed study area with shaded relief map.
the Sonoran and Chihuahuan deserts, and topography, climate, and vegetation vary substantially
across the watershed. Elevation ranges from 900 to 2900 m and annual rainfall ranges from 300
to 750 mm. Biome types include desertscrub, grasslands, oak woodland-savannah, mesquite wood-
land, riparian forest, and conifer forest, with limited areas of irrigated agriculture. Urban areas,
including several small towns and the rapidly growing U.S. city of Sierra Vista, are fringed by low-
density development that also occurs far from population centers. Numerous geospatial data sets
covering the upper San Pedro watershed can be viewed and downloaded at the U.S. Environmental
Protection Agency San Pedro Web site (USEPA, 2000).
9.2.2
Reference Data Sources for Accuracy Assessment
Aerial photography has long served in the creation of LC maps, both as a mapping base and
more recently as a source of higher-resolution reference data for comparison with maps produced
by classification of satellite imagery. Coverage for the conterminous U.S. at a scale of 1:40,000 is
available through the National Aerial Photography Program (NAPP) and is scheduled for update
on a 10-year, repeating cycle. Digital orthophoto quarter quadrangles (DOQQs) are produced from
the 1:40,000-scale NAPP or equivalent high-altitude aerial photography that has been orthorectified
using digital elevation models (DEMs) and ground control points of known location. A DOQQ
image pixel represents 1 m
on the ground, permitting detection of landscape features as small as
approximately 2 m in diameter. However, the image analyst may need site visits and/or supple-
mentary higher-resolution images to visually calibrate for DOQQ-based LC interpretation.
Marsh et al. (1994) described the utility of airborne video data as a cost-effective means to
acquire significant numbers of reference data samples for classification accuracy assessment. In
that study, very similar classification accuracies were derived from airborne video reference data
2
