Salami, Ayobami T. 
  
  
2. STUDY AREA 
The study was carried out in Ede Local Government Area of Southern Nigeria. The area has two rainfall 
maxima and the period of raining season varies between seven and nine months. The area was mapped as 
lowland rainforest by Keay (1959). However, only patches of this natural vegetation are now found in the 
area, as a result of the practice of shifting cultivation and peasant cash crop economy (Ekanade, 1985). 
The most extensive soils in the area are Egbeda, Olorunda, Iwo, Ibadan and Balogun series (Smyth and 
Montgomery, 1962). Adejuwon (1971) opined that most of these soils could be considered as good soils 
with regard to the performance of biological communities on them. 
3. METHODOLGY 
Landsat TM scene 6967 x 5965 with Worldwide Reference Number (WRN) 190/055 of 1991 was used for 
this study. A window with an areal coverage of 100km^ was extracted from the scene. This contains 334 x 
334 pixels. The coordinates of the first corner of the window are defined by X=1419 and Y=1273. The 
window was subjected to digital image processing using MULTISCOPE Software Package. Channels 4, 5 
and 2 were chosen out of the seven channels available on Landsat TM. This selection was based on the fact 
that these channels are more relevant for vegetation studies. Channel 4 was loaded on the red plane, 
channel 5 was put on the green plane while channel 2 was loaded on the blue plane to obtain a colour 
composite. 
Training parcels were chosen from ground control points and they were eventually assigned to six classes. 
Some of the training parcels were reassigned by the automated procedure. The parcels reassigned are those 
for which more than 50 % of their pixels were wrongly classified during the preliminary grouping (CNES, 
1988). 
The extent to which each of the vegetal classes differs from one another was computed prior to final 
classification. The result of this exercise indicates what to expect in the final mapping. The output of this 
procedure is shown by divergence matrix. The latter is the exact opposite of a correlation matrix although, 
each relationship is expressed by a value ranging from 0 to 1. A value of 0 means that there is total 
correlation and absolutely no (or zero) separability while a value of 1 means absolutely no (or zero) 
correlation and total (100 %) separability between the two classes concerned. In other words, the higher the 
value, the lower the correlation and the higher the separability and vice-versa. 
The final mapping was done with 95 % level of confidence under the Gaussian hypothesis, using the 
maximum probability algorithm. In this case, we accept the normality hypothesis and consider each Ci 
class, averages Mi and covariance matrices Qi. The multivariate normal statistical theory defines the 
probability that an observation X will occur, given that it belongs to class K, as the following function 
(Tatsuoka, 1971): 
0.00 = 20) POLS 2 Xe 2 0 (XU) enon: eq.(1) 
Where: 
Q«(X; = Probability density value associated with observation vector X; as evaluated for 
class K. 
X; = Vector of measurements on p variables associated with the ith object or observation; i=1, 2, ...,N 
P = Number of measurement variables used to characterise each object or observation. 
Uy = Parametric mean vector associated with the Kth class. 
[x = Parametric p by p dispersion (variance-covariance) matrix associated with the kth class. 
X = p-dimensional vector. 
  
1302 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.