Salami, Ayobami T. 
Class purity rating (Table 4) was derived from confusion matrix (Table 3). The former is obtained by 
dividing the number of pixels well classified by the total number of pixels in the class. Table 4 shows that 
among the vegetation types in the area, arable cropland/fallow complex and burnt patches, are the most 
efficiently studied class using digital image processing procedure employed. The former was classified 
with 91.3% precision while the latter was classified with 90.7% precision. This is due to the fact that arable 
cropland/fallow complex consists of young luxuriant vegetation with distinctive bright red signature while 
burnt patches have distinctive black signature on the original imagery prior to classification. 
Table 4: Class Purity Rating of Cover Types 
Class Class Content % of Pixels % of overlapping Purity Ratio Purity Rating 
No. Well Classified Pixels 
1. Burnt Patches 90.70 621 0.907 DM 
2. Arable Cropland/ 91.30 7.39 0.913 j* 
Fallow Complex 
3. | Exposed Land/ 83.17 12.02 0.832 4 
Settlement 
4. Mixed Grassland 81.83 13.75 0.818 i 
5 Tree Crop/Mature 83.42 15.36 0.834 3% 
Secondary Regrowth 
6. Young Secondary 54.01 40.39 0.540 6^ 
Regrowth 
Young secondary regrowth is the most difficult vegetation unit to classify. It was classified with 54.0196 
precision. This is probably due to the fact that the field survey showed that it constitutes the 
undergrowth/understorey of mature secondary regrowth in the area. In fact, in a generalised procedure, it 
could be merged with mature secondary regrowth to form a vegetation complex. This is what Table 2, 
which shows that young secondary regrowth, has only 27.9396 separability from tree crop/mature 
secondary regrowth implies. An attempt was however made to distinguish the former from the latter and 
hence, its poor purity rating (54.0196). This type of problem has been noted to be common as a result of the 
fact that automated procedure represents transitions as boundary lines (Martin, 1959) and as such, 
interpretation of satellite imageries whether by analogue or digital methods is not a precise science 
(Nahani,1990). This made Allan (1989) to conclude that beguiling elegant cartography can conceal high 
levels of error while Hobbs (1990) argued that accurate representation of terrestrial vegetation in Earth 
system models has been a continuing challenge. Given the nature of the landscape in Nigeria, it is 
necessary to amalgamate like-classes during the digital image processing in order to enhance the level of 
classification accuracy. This however, means that only broad rather than specific classes could be generated 
with a high level of reliability. 
4.3 Automated Mapping of Vegetal Cover 
Figure 1 shows the result of the final mapping. The summary of its statistics (Table 5) shows that only 
43.96% of the area could be said to be under forest cover. This is the total area occupied by tree 
cropland/mature secondary regrowth (17.14%) and young secondary regrowth (26.82). On the other hand, 
33.63% of the area was occupied by mixed grassland. This situation implies that although the climatic 
climax of the area is Lowland Humid Tropical Rainforest (Keay, 1959), there is now quite a keen 
competition between forest and savanna elements for ecological space in the area. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1305