eld observations
| areas do not
gion and such
/ problematic for
vo of the twelve
n remnant trees
irmed from plot
iably distinguish
and mangrove,
jle for the latter
n the different
r priority for the
on of important
salt marsh and
ould help both
and ecotourism
rts.
most troubling
ion concerned
rops of manioc
arly banana. A
manioc is often
small = plots.
its classification
om mixed pixel
\s a subsistence
uld usually be
(ear other crops
wy not seriously
indications of
lassifying banana
t is understand-
latively mature
jven their high
ht and cover.
is an important
f the current
Janana is an
p in the region,
ated in large
ear the ends of
r proximity to
Is may help in
of conservation
and remoteness
require further
future reclassi-
S.
es and Maps
rages for each
cover type must still be combined into a single
unified coverage. This will require some cleanup
steps. The photointerpretation process digitized each
cover type as a separate coverage and left dense
forest as the unclassified default condition. Slivers
and overlap between the different photointerpreted
cover types can occur and will need to be resolved.
Rasterization of the final coverage will also need to
be performed to facilitate data sharing and
environmental modeling.
The first graphical cover type map produced
from the project has proven extremely informative
about land cover and characteristics in the region.
The combination of false color composite imagery
overlain with pattern fills showing mangrove,
restinga, humid soils, and planalto allows ready
identification of interpreted cover types with
minimal obscuring of the original imagery. Only
beaches and the scattered exposed soil polygons are
shown in solid fill, permitting most of the false color
and topography to show through. Orientation is
provided through localization maps, roads, and
annotated place names.
A second map showing false color
composite for the broader region provides a useful
broader adjunct to the classified image. However, for
both of these maps their large size and use of full
color limits their ease of production and duplication.
Simpler versions are planned for broader
distribution.
5. DISCUSSION
As a technology transfer project built around
a case study of ecotourism and remote sensing, the
project’s success can be judged in two ways: the
utility of the resulting land cover classification and
GIS database, and the improved skills and local
capacities resulting from the effort. These are
evaluated following discussion of some ways the
work might have been improved.
In hindsight we can suggest several ways the
study might have been done better. Scheduling the
field season to match the drier Brazilian winter
would have been one such move, perhaps with two
concurrent crews and GPS receivers. More field
plots would certainly have helped deal with the large
diversity within the region.
A more interleaved sequence of field and lab work
seems preferable to our sequential approach. A
preliminary analysis of the satellite image could be
done first, followed by establishment of field plots,
then supervised classification, then followup ground
truthing, and lastly critique by local experts.
77
Better on screen digitization procedures
would have minimized gaps and overlap between the
cover type polygons, although we are optimistic that
improved classification methods may reduce the
need for photointerpretation.
In this next round of classification we plan to
sequentially classify groups, mask them out, then
classify the remainder. With the recently available
coverages from IPARDES we can now test whether
stratifying the image based on geomorphic properties
aids classification accuracy. Anticipated upgrades to
the Imagine software may also reduce some of the
classification difficulties we encountered.
A high priority for future work should be
development of a DEM. It would help remove
differential illumination problems with mountain
shadows as well as contribute more directly to a wide
range of environmental models and analyses.
With the exception of banana the coverage
succeeded in separating those places where human
disturbance has been recent and severe from those
where natural vegetation occurs largely intact. Given
the relatively rapid changes that land uses in the
region can undergo, such a distinction is extremely
useful, particularly if it can be updated periodically
to provide feedback on the effectiveness of forest
protection policies. This mapping was accomplished
with a spatial accuracy consistent with the 30 m
pixels of the original imagery and the spatial scale of
the desired features.
The spatial land cover data from this project
are being made available to other interested parties,
and can now be analyzed together with other spatial
data, allowing more timely, objective, and complex
investigations of management alternatives than have
been previously possible.
The potential benefits of the mapping effort
were demonstrated during Idrisi training near the
conclusion of the project. Early versions of the
coverage were used along with several from
IPARDES in exercises demonstrating trail planning,
conservation zoning, and deforestation assessment.
The technology transfer objectives of the
project favored development of local capabilities
over reliance on more experienced facilities available
elsewhere in Brazil or the United States. Building
local expertise in remote sensing of natural resources
and application of GIS offered long term benefits far
beyond the direct benefits of the database itself,
although it did differentiate the effort from more
operational projects by requiring more flexible
planning and greater attention to communication
needs.
Most of the individuals in the project are