linking economic benefits for residents to successful 
conservation of the natural resources on which such 
tourism relies. The International Forestry program of 
the USDA Forest Service provided a technology 
transfer grant to assess its feasibility for Guaraque- 
caba, investigating potential economic demand, local 
economic impacts, and available ecologic resources 
and risks (Cubbage et al., in press). This paper 
describes the land cover mapping effort designed to 
assist ecotourism and conservation planning in the 
region. 
High priority was placed on collaboration 
and capacity building in Brazil. The Brazilian side of 
the work was anchored by SPVS, a conservation 
organization with considerable experience in the 
region. Image classification was conducted at CIEG. 
Forest Service employees worked closely with the 
Brazilian organizations, providing coordination, 
training, and overall project direction. Other major 
participants included the state planning ministry 
IPARDES and the Brazilian federal park service 
(IBAMA). 
2. OBJECTIVES 
Maps of current land cover were needed to 
assist ecotourism planning by: 
€ locating examples of ecosystems potentially 
attractive to ecotourism, 
€ facilitating conservation “zoning” to protect areas 
vulnerable to disturbance, and 
® monitoring successes and failures of forest 
protection efforts. 
An additional objective involved exploring 
the relations between accessibility and deforestation 
in the region. Encircling mountains and poor roads 
have limited access in the past and prompted 
proposals for construction of paved roads into the 
region. Quantifying the present relation between 
accessibility and deforestation would allow us to 
simulate the likely effects of improved road access 
on deforestation risk. 
Initial interviews found that the most recent 
information on land cover in the region was based on 
orthophotos taken in the early 1980's. These had 
been manually interpreted and were available only in 
hardcopy form. Such maps predated many important 
milestones for the region: reserve status for the 
region, introduction of water buffalo ranching, and 
prohibitions against tree cutting. Along with such 
gradual processes such as shifting agriculture, land 
tenure consolidation and road network expansion, 
these changes suggested that existing maps of land 
cover were sadly out of date. 
72 
Efforts were already underway at IPARDES 
to digitize hardcopy maps of soils, hydrology, 
geographic province, and other themes into a 
geographic information system (GIS). Although 
these hardcopy maps were developed at the same 
time as the orthophotos, we felt that these physical 
characteristics were generally less changeable over 
time than land cover, and thus sufficiently current. 
What was needed was a cost effective method of 
acquiring current information on land cover, which 
could be repeated periodically, and which produced 
data which could be readily incorporated into the 
developing GIS for use by IPARDES, SPVS, and 
other interested parties. 
3. METHODS 
Our plan was to map land cover in the region 
to approximately 1:150,000 scale using supervised 
classification of a Landsat TM image. National map 
accuracy standards would be followed, and 
collection of data at training sites in the field would 
support the supervised classification and allow 
measurement of classification error. Land cover data 
would be supplemented by digitized roads and 
tourist points and the coverages made available to 
interested parties in  Arc/Info (Environmental 
Systems Research Institute, Inc.) vector and ERDAS 
Imagine (ERDAS, Inc.) raster formats. 
While coverages would be created using 
workstation versions of these GIS programs we 
anticipated that a more modest DOS-based GIS 
program, Idrisi (Clark University), would be 
employed for most subsequent analyses. Copies of 
Idrisi were provided along with training to both 
SPVS and IPARDES for this purpose. 
Landsat TM imagery offered several 
advantages for this project, including its powerful 
spectral offerings, reasonable resolution, ready 
availability over time, and affordable pricing. Some 
recent studies have shown success at mapping 
vegetation using Landsat TM in similar ecoregions 
(Pope et al., 1994; Kachhwaha, 1993; King, 1994; 
Curran and Foody, 1994; Rey-Benayas and Pope, 
1995) but at the beginning of this study the 
applicability of automated classification in this 
diverse and mountainous region was still untested. 
Images were obtained in April 1994 from the 
National Institute for Space Research (INPE), 
Brazil’s space agency. Although our field survey was 
scheduled for early 1994, to avoid heavy cloud cover 
we selected an image from September 1993, four 
months earlier. We also purchased a second image 
from June 1986 to allow change detection analysis 
(not yet atte 
for our anal 
We 
methods bec 
unsupervise 
methods: 
© encourag 
literature 
e facilitate 
compatib 
e ensured 
integral r 
3.1 GPS an 
Glo 
collected gi 
work, whic 
February 1 
needed: 
€ control | 
georefere 
e training 
classifica 
€ road an 
productic 
accessibi 
Col 
tourist attra 
channel Pa 
GPS receiv 
to 1 megab: 
objects into 
the United € 
the data log 
used for th 
point locati 
rented for tl 
Janeiro. 
All 
after downlo 
from a surv 
Curitiba esp 
Twi 
and Februa 
for the sum 
data. Their 
additional e 
GPS receive 
Twi 
Forest in M 
joined the te 
establish GP 
their use. Th