ler 
er Van Wie 
in utilizing 
onducting re- 
value of the 
vailability of 
has the great- 
1ysical, eco- 
s particularly 
iate programs 
sir current re- 
inces Landsat 
a test applica- 
id use in two 
ration with 
of the NASA/ 
sites within 
b) to provide 
base manage- 
hemes to the 
ibsequent land 
the Dajabon 
an Island of 
ie northwest 
er with Haiti. 
le varied. The 
o rolling hills 
1 of the pro- 
3,000 meters. 
maining forest 
vide charcoal 
re accelerated 
itian mountain 
ics along the 
rically recti- 
). Thus sub- 
image for the 
by 67 meter 
.ID's data base 
he resampled 
via training 
Dominican Republic/Williams, Stauffer, Bly and Van Wie 
areas and/or radiance cluster analysis. The validity of the preliminary classification scheme 
and map was assessed by the first author during a two week field trip in May of 1977. A de- 
tailed, statistical verification of the preliminary and subsequent final classification maps was 
virtually impossible due to the small-scale, heterogeneous nature of the land uses, coupled 
with the fact that nearly 3-1/2 years had elapsed since the November 1973 Landsat coverage. 
However, the delineation of broad land use categories such as wetlands, agriculture, forests, 
brushlands, and barren areas, suitable for a national-scale assessment, was very good. A few 
discrepancies in the preliminary classification map were noted in the field and appropriate re- 
visions in the training data were made. A final classification map representing the above noted 
broad land use categories was generated using a total of 17 spectral signatures (i.e., several 
spectral signatures were needed to properly map each of the major land use categories) (Fig. 
3.1b). The concentration of forested areas and brushlands (i.e., the darkest gray tones in Fig. 
3.1b) in the center lower-center portion of the province is readily apparent upon inspection 
of the final classification map. This corresponds to the increased topographic relief and for- 
est cover in the mountainous areas in the southern portion of the province. The lighter gray 
tones (i.e., the non-forest categories) scattered throughout these areas, as well as the predom- 
inance of lighter tones on the Haitian side of the border is evidence of the forest exploitation 
occurring in this part of the world. 
The final 67 meter resolution land cover map of the Dajabon Province was next summarized 
into 1 km by 1 km cells for input to AID's computer-based system. Each new 1 kilometer 
céll was an aggregation of approximately 220 of the 67 by 67 meter classification cells. Addi- 
tional experiments were also conducted to determine the affects of classification sub- 
sampling or aggregation techniques upon the validity of the resulting land cover maps. 
A 1 km? cell map depicting various percentages of forest canopy cover within the province 
was generated as a test output product (Fig. 3.1c). This type of aggregation reduces spatial 
noise and future processing costs, and more clearly illustrates the spatial distribution of the 
forest cover within the Dajabon Province. 
The sampling experiments indicated that at the province level a sampling of as little as 1 or 
2% of the classified Landsat cells would provide an adequate representation of land use. Thus, 
an approach of image sampling may be feasible for reducing processing costs in monitoring 
large-scale forest alteration on a national or worldwide basis. 
Further Information 
References 
Bly, B. G., and Stauffer, M. L., 1977. Intralab Interim Report III. Contract NAS 5- 
11999, Task Assignment 646. Computer Science Corporation/TR-77/6007. 37 p. 
Place, M. C., and Van Wie, P., 1977. AOIPS land cover data management system. 
93 p. 
Sellman, A. N., Bly, B. G., Stauffer, M. L., and Williams, D. L., 1977. Computer 
analysis and classification of Landsat data for land cover of selected test sites in 
the Dominican Republic. Intralab Draft Report to the Economic Research Service 
(USDA) and the U.S. Agency for International Development. 15 p.