geocoded satellite images to rectify several hundred aerial
photographs; 2) establish efficient photo-interpretation and
feature encoding procedures compatible with a vegetation
classification system that takes into account vegetation species,
human impacts and hurricane damage; and 3) integrate GPS
surveys with attribute recording and digital image processing
on a laptop computer to facilitate the real time collection of
ground truth information by helicopter surveys.
2.1 South Florida GPS Surveys and Satellite Image Mosaic
To ensure the registration of vegetation, transportation and
hydrographic features in a GIS database, it was first necessary
to establish a network of ground control points (GCPs)
adequate for rectifying eight SPOT panchromatic images (10-m
resolution) of the study area. Past experience has shown that
because of their excellent internal geometry, only 4 to 6 GCPs
per image are required to geocode a SPOT satellite image to a
planimetric accuracy of + 0.5 to + 1.0 pixel - or + 5 to + 10 m
on the ground. Such accuracies are compatible with U.S.
National Map Accuracy Standards for 1:24,000 scale maps and
were considered acceptable for densifying the horizontal
control points needed to rectify the aerial photographs from
which the database would be constructed.
The survey to establish GCPs was influenced by the existing
road network and by the availability of 1:24,000 scale
topographic line maps of recent vintage for the lands adjoining
the Parks. Thus, it was decided to use the existing maps for
perimeter control and to conduct the GPS survey along the few
roads through the middle of the study area to provide the
interior control necessary to rectify the satellite images.
Control points included the intersections of roads and of roads
and canals, both of which are easily located on the SPOT 10-m
panchromatic images. In addition, the survey included eight
monumented National Geodetic Survey (NGS) and Florida
GPS (FLGPS) base points of Order B (1:1,000,000) accuracy
(or better) which were used to adjust the network. Based on
checks conducted in the post-processing adjustment of the GPS
observations, the accuracy to which the UTM coordinates of
the 23 points were established as referenced to NAD 83 was
about + 0.05 m (RMSE,y).
The SPOT image mosaic was created from individual SPOT
scenes using the Desktop Mapping System (DMS)'M software
package. Each digital SPOT image (or tile) was rectified to the
UTM coordinate system using 6 to 15 GCPs per image and
then placed at its correct location (according to the UTM
coordinates of the upper left corner pixel) within a coordinate
box for the entire study area. In order to facilitate its use, a
UTM grid (NAD 83) with 5,000 m spacing was registered to
the mosaic. The planimetric accuracy of the geocoded mosaic
was evaluated at 29 withheld control points (check points) and
found to be better than + 1.5 pixels. The mosaic occupies
slightly more than 200 Mbytes of disk space.
2.2 Interpretation of CIR Aerial Photographs
Development of an accurate, detailed vegetation database for
the Parks requires the use of remotely sensed data of sufficient
resolution to allow the identification of plant species/asso-
932
ciations, the delineation of one-hectare or larger vegetation
plots and classification accuracies of 90 percent of better.
These requirements precluded the use of Landsat TM (28.5 m)
or SPOT (20 m) multispectral imagery for the construction of
the vegetation database. Instead, the use of aerial photographs
was deemed essential for the project. Color infrared (CIR)
aerial photographs of 1:40,000 scale acquired as part of the
U.S. Geological Survey (USGS) National Aerial Photography
Program (NAPP) in 1994 and 1995 were purchased for the
entire South Florida study area. In order to facilitate the
interpretation of these photographs and the construction of
vegetation coverages in digital format, vegetation classes were
delineated directly on CIR paper print enlargements (4x)
produced from the CIR film transparencies. Point features
common to both the SPOT images and the analog air photo
enlargements were annotated, numbered and their UTM
coordinates determined from the geocoded SPOT images to
establish a GCP file for each CIR photograph.
The GCPs transferred to the photographs were then digitized,
along with the vegetation polygons and other point and line
features. By digitizing the GCPs first, a set of photo recti-
fication coefficients were generated that allowed x,y digitizer
coordinates for the vegetation polygons to be transformed to
UTM map coordinates (Easting, Northing). This procedure
permitted a segregation of tasks (interpretation, digitizing,
editing) and greatly facilitated the development of the veg-
etation database. To further improve digitizing efficiency in
areas of complex vegetation patterns, clear plastic overlays with
annotated vegetation boundaries were scanned at 65 um and
accessed by the VTRAK® software package (Laser-Scan, Inc.)
for automatic vectorization and input to ARC/INFO®. Tests
conducted to assess the accuracy to which the features were
digitized yielded RMSE,, values of between + 5 and + 10 m.
2.3 Field Verification with GPS Assisted Helicopter Surveys
and Vegetation Classification
A new Everglades Vegetation Classification System was
developed in conjunction with personnel from both the NPS
and South Florida Water Management District (SFWMD)
because existing vegetation classification schemes did not
contain the level of detail required for this vegetation mapping
project. This classification system includes hierarchical veg-
etation classes at the individual species or species association
level that can be identified from the CIR aerial photographs. In
addition to the floristic characterization of the naturally
occurring Everglades plant communities, the classification
system includes categories of human impacts and episodic
disturbances that influence vegetation species distributions.
For example, invasive exotic plants, indicators of human
influence such as agriculture or off-road vehicle (ORV) trails
and three hurricane damage classes are added as modifiers to
the vegetation class.
Fieldwork to verify the image interpretation was greatly
facilitated by the use of NPS Bell Jet Ranger 206 helicopters
that are available for ground truth collection. Since helicopter
flight time is expensive (— $ 650.00/hr), a procedure involving
use of the SPOT image mosaic and the latest technology in
laptop computers, GPS receivers and image processing/posi-
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996