(9) The valley bottom soils (Fringing forests) are among the most poorly classified units.
Although the Benue and its alluvial landforms were recognised and correctly classified by the
computer, the upland fringing forests were either not correctly classified or were ignored. This
may be due to the fact of their narrow expanse, making it possibie for the surrounding areas to
mask the reflectance of their vegetation. If they were not up to 57 m or 79 m across, as was
generally the case, there would not be many pixels through which they could show a domineering
reflectance. Kristof et al. (1977) detected this problem with such narrow features. Maximum
Likelihood classification is the algorithm that showed the distribution of the fringing forests most
clearly.
Quantitative Comparison
Hinzel et al (1980) considered various methods of quantifying the amount of soils in each
sprectral class. The methods of counting pixels and of drawing boundaries on pixel maps were
discouraged because of inherent problems associated with these methods. They tried the method
of dot count of soils occurring in spectral classes by overlaying a dot grid on an acetate paper
tracing of pixel class over a GT map. They seem to have more confidence in this method probab-
ly because some pixels predominantly define some soil series, e.g. poorly drained soils vs well
drained soils on which basis groups of soil series can be predicted. The poorly drained soils, for
example, do have lower reflectance than well drained soils, But, as Weismiller et al. (1877)
pointed out, there is no one-to-one correlation between spectral classification and soil class
whereas a soil series may be represented by one or more spectral classes.
Previous machine processing of MSS Data for soil mapping (Hinzel et al, 1980; Condit, 1970;
Weismiller et al, 1977; Kirschner et al, 1978) have been based on the reflectance of the exposed
visible surface soil. But it has been shown here and by previous workers that different soils do
have similar reflectance responses and the same soil may have different reflectance properties.
The reflectance responses recorded here by the scanner are not of pure soil but are either of a
mixture of soil and vegetation reflectance or of pure vegetation, further complicating the prob-
lem. These facts will affect any quantitative comparison of digitally and conventionally produced
maps as shown by table 1.
The following figures of the table 1 may be presented as a rough estimate of the result of
ma chine processing to duplicate within a few hours the laborious work of soil mapping.
The most obvious thing about these data is the close similarity between the figures tor the
two pixel maps. This is to be expected because of the similarity of the parameters of classifica-
tion % the spectral responses of the land covertype.
The deviations of the figures of the pixel maps from the physiograp hic-soil map are similar.
Broadly speaking, there are two main types of deviations: (a) ' — '' where the pixel map percen-
tages are higher than in physiographic-soil map and (b) '' * " where the physiographic-soil map
percentages are higher.
" — '"' The most obvious reasons for the higher percentages of the pixel map units is the
similarity of reflectance characteristics. With most of this being of soils having good drainage,
their reflectance characteristics are similar to those of the well drained soils of the interfluvial
slopes, which by the same token, suffered the greatest corresponding decreases, notably the
Makurdi undulating slopes and the Keana slopes.
T)
-- f^ -— e e p r3 À
-— — a
Ny OY "rN fx