sub-site, 21,316 hectares in size, includes the
settlement of Berberati. The change from CLOSED
FOREST to OPEN FOREST (twelve percent) and from
OPEN FOREST to CLOSED FOREST (three percent) was
identified. In both instances, the geographic
location of these changes do not occur as
contiguous patches or along boundaries between the
two classes. This lack of a distinct spatial
pattern suggests that the change identified may
represent possible variations in the amount of
understory vegetation detected through the forest
canopy.
An estimated 48 percent change from NONFOREST to
OPEN FOREST cover was identified and the change
from OPEN FOREST to NONFOREST was approximately
one percent. These areas of change have a
patch-like nature (contiguous parcels of land
cover), which suggests that both types of change
refer to parcels of land involved in crop
rotation.
Change in Land Cover for SITE 1-C
Of all six sub-sites, Site 1-C has the largest
proportion of area classified as NONFOREST on both
dates (approximately 90 percent). Within this
14.6 kilometer square area, the most noticeable
change was in the CLOSED FOREST class ; the closed
canopy forests on uplands in the eastern half of
this sub-site were virtually eliminated by 1987.
In general, only lowland forests, located along
the water courses, remained relatively unaltered
over time.
Multi-date NDVI Differences
Assessment of differences in NDVI between 1973 and
1987 indicates that, in general, NDVI values were
higher in 1973. This difference is observable in
the statistical summary of the entire study area
and for each of the sub-sites (Table 5). Lower
NDVI values suggest that conditions are less green
in 1987, which could result from either poorer
quality Landsat MSS data from 1987 or because of
the reduction in precipitation received in 1987
over that received in 1973.
Table 5. NDVI average statistics and differences
for the six sub-sites and the entire SITE ONE.
Sub-site
1973 NDVI
1987 NDVI
NDVI Difference
1-A
.214
.168
.045
1-B
.122
.096
.025
1-C
.039
-.052
.091
1-D
.133
.053
.080
1-E
.097
-.014
.111
1-F
.185
.057
.128
Whole Area
.117
.042
.075
Visual analysis of the geographic patterns of
variation in NDVI as compared with the false color
image displays was used to confirm the fact that
the NDVI data had the information content desired.
In general, this visual analysis confirmed that
NDVI values were highest in the closed rain forest
(Site 1-A) and lower in secondary grassland areas
(Sites 1-C and 1-E) with the lowest NDVI recorded
for bum sites (Table 5).
Analysis of each of the sub-sites suggests that
site specific differences in the direction and
magnitude of landscape change can be identified.
Largest NDVI differences (towards less greenness)
occurred in Sites 1-C, 1-D, 1-E and 1-F. These
sub-sites exhibit spatial patterns that indicate
extensive bum patches with different greenness
levels for the two time periods. In several
locales, an area was more green in 1987 whereas
the other patches show decreases in greenness
indicative of lower greenness levels in 1987.
Combined analysis and interpretation of the NDVI
difference statistical data and the change results
derived frcm the two different classification
approaches warrants further investigation.
Greatest environmental change toward more open
conditions occurs in Site 1-F (Carnot), which also
has the greatest NDVI difference. Site 1-B has a
considerable amount of landscape change, but the
net effect is toward more green cover; this
Sub-site has the smallest NDVI difference. Site
1-A exhibited little change over the fourteen year
period of assessment and also has a relatively low
NDVI difference. These results demonstrate a
correspondence or functional equivalence of the
classification and NDVI difference methods in
assessing environmental changes in moist tropical
forest and grassland environments of equatorial
Africa.
CONCLUSIONS
In this study, data availability restricted the
assessment of environmental variations at SITE ONE
to a time scale of fourteen years. The general
conclusion frcm this research is that
deforestation is not occurring at a high rate in
this region of Africa yet environmental stresses
in these areas warrant confirmation from other
kinds of data sources. Ground-based appraisals
of environmental conditions would help corroborate
this satellite-based interpretation. If dry
season burning had occurred just prior to
satellite data collection in 1987 and was yet to
occur at the time of overpass in 1973, then seme
of the landscape changes that were interpreted
from use of the two satellite data sets may
represent seasonal differences rather than
temporal trends. Interestingly, the 1973 data
were obtained on January 28; approximately one and
one- half weeks later into the dry season than the
1987 data.
The lack of a one-to-one correspondence between
spectral classes and land cover types may be
attributed to changes in vegetative vigor rather
than actual land cover change. Another possible
explanation may point to a problem with the
methods chosen for analysis.
Visual analysis and interpretation of the spatial
patterns in vegetative cover apparent in the
Landsat MSS data corroborate the statistical data
provided by the different change detection methods
(Mausel, et al., 1990). Visual interpretation is
possible because the thin layer of smoke that
covers a majority of the image does not obscure
the underlying vegetative variations.
While a review of the relevant literature on
forest change detection indicates that other
methods have been used successfully, the methods
used in this study are comparable and involve the
use of data frcm the same spectral bandwidths.
The following conclusions may be derived from this
investigation of three sites in equatorial Africa;
contingent upon more field verification. The
closed rain forest has undergone very little
modification in locales away from populated towns
and roadways, while forest degradation appears to
252
be occurring near populated areas and along the
