15°30’E 
Lake Mai-Ndcmbe is located in the north-central 
part of SITE TWO where, according to White (1983), 
the vegetation to the east of the lake and in the 
northeast reaches of the study area is considered 
Guineo-Congolian swamp forest (including riparian 
forest). Lake Mai-Ndcmbe drains south into the 
Fimi River near the town of Kutu while along the 
southern edge of SITE TWO is the Kasai River. 
White described the vegetation along the western 
and southern edges of the lake as well as along 
these major rivers as drier Guineo-Congolian rain 
forest. A mosaic of Guineo-Congolian rain forest 
with secondary grasslands is found beyond the 
river corridors. 
Vegetation types identified by White (1983) for 
SITE THREE include: Guineo-Congolian rain forest 
(in the southem/southeastem part of the study 
rectangle); Guineo-Congolian rain forest: drier 
types (isolated patches in the central part of 
SITE THREE), and a mosaic of secondary grassland 
and Guineo-Congolian rain forest for most of the 
northern two-thirds of SITE THREE. The area is 
drained to the west by the River Nepoko (southern 
sections of SITE THREE), by the River Bcmokandi in 
the central section, and by the River Uele in the 
north. Relatively large settlements within this 
site include: Dungu at the northern edge (on the 
River Uele) and Mungbere. 
OBJECTIVES 
In this investigation, proven image processing 
methods for determining land cover and vegetation 
vigor were applied to Landsat MSS data to assess 
land cover characteristics from spectral 
signatures and changes in land cover for an area 
in equatorial Africa. The data provided herein 
are destined to serve as a link between more 
macroscale land cover data collection (e.g. AVHRR) 
and field collected site data. Availability of 
field data was highly limited, therefore, results 
generated from the interpretation of the spectral 
data are presented as a series of hypotheses for 
future field verification. These hypotheses are 
based on spectral analysis of Landsat MSS data. 
Once verified in the field, the actual land cover 
data will serve as one test of the quality of the 
estimates provided by FAO/UNEP (1981). These 
multitemporal assessments of land cover may be 
used as a valuable source of quantitative data for 
verification of models of deforestation in Africa. 
Although all three sites have been analyzed, the 
focus of this paper is on SITE ONE where ground 
truth, although discontinuous, is more readily 
available. SITE ONE contains a majority of land 
cover features found collectively in the three 
study sites and the trend in land cover change is 
considered representative of the other sites. 
DATA CHARACTERISTICS 
Many detailed retrospective studies of the Earth's 
surface have used Landsat MSS data because 
comparable data date back to July, 1972. 
Relatively cloud-free data were acquired for the 
southwestern Central African Republic by Landsat 
1 an 28 January 1973 (Scene ID: 1189-08422; Path 
196, Row 57; with approximately 20 percent cloud 
cover). Comparative data were obtained by Landsat 
5 on 17 January 1987 (Scene ID: 51052-08340; Path 
183, Row 57; approximately 10 percent cloud 
cover). Interactive display and visual 
interpretation of the 1987 data revealed that, 
while the percent cloud cover was relatively low, 
the vast majority of the scene was contaminated by 
varying amounts of backscatter from smoke (most 
likely from the dry season grass fires occurring 
in July and August and again in December and 
January). Increased Rayleigh and Mie scattering, 
as a result of these atmospheric contaminants, 
provided cansiderabl e and varying amounts of noise 
in the reflectance signal from the surface cover 
types. To insure accurate assessment of 
vegetation change, sub-sites with minimal visual 
evidence of smoke/haze contamination were selected 
for detailed investigation in the Central African 
Republic study area: Sites 1-A through F (Figure 
2). 
Preprocessing 
The data from the two different Landsat sensors 
were transformed into similar units (Robinove, 
1982) thereby increasing the reliability of the 
temporal comparison of surface vegetative cover 
and mitigating minor radiometric differences 
(i.e., gain settings). Standard procedures were 
used to calculate radiance and exoatmospheric 
reflectance from the raw digital counts (Robinove, 
1982; Markham and Barker, 1986). Reflectance 
calculations were done using the solar irradiance 
spectrum recommended by the World Radiation Center 
(Iqbal, 1983). 
The next phase of data preprocessing involved 
image registration. First, using SITE ONE for 
example, the 1973 Landsat 1 data were deskewed; 
the 1987 data were deskewed prior to user 
purchase. The 1973 data were then resampled to 
georegister the 1973 data with the 1987 data; the 
57 m by 57 m pixel size of the Landsat 5 (1987) 
data was used. A bilinear interpolation algorithm 
was used and the transformation equations were 
developed from data files that contained pixel 
coordinates for distinctive landscape features 
that could be identified in both the 1973 and the 
1987 data sets; the RMS error was 0.5 pixels.