355 
(Ul), mixed cultivated and fallow land (MF1), and mixed cultivated and 
pasture (CPI) were 27.2% 58.0% and 3.5%, whereas the errors for omission 
were 25.4%, 16.0% and 25.7%. These errors were very high in all three 
cases and indicate the poor discrimination of these areas from other 
classes. 
From these results it is concluded that the inability to discriminate 
among certain classes is due to the spectral similarity between those 
categories. Other spatially-oriented data are required to augment and 
enhance the classification process (Civco, 1987). Also, the limited 
range of the SPOT HRV sensor (green, red and near infrared) may not 
permit adequate spectral discrimination of the classes being mapped. 
Discrimination may be better using other regions of the electromagnetic 
spectrum. The classification shows that the percentage accuracy 
decreases as the level of detail is increased. The more spectrally 
heterogeneous areas also reduce the accuracy percentages of the 
classifications. 
CONCLUSIONS AND RECOMMENDATIONS 
1. Digital interpretations of SPOT 20m resolution imagery proved to be 
effective in determining detailed assessment of land use classes, 
soils, and surface hydrology for a selected area in the semi-arid 
region of northeastern Brazil. 
2. The limited spectral range of the SPOT HRV sensor did not appear to 
permit adequate discrimination of some land use and land cover 
categories. 
3. Classification became less accurate as the level of detail was 
increased. The relatively low accuracies at Level II were attributed 
to cross-confusion among related categories. 
4. Accuracy assessments of the digital classifications showed that some 
categories, such as water, forest and alluvial land, were identified 
more accurately than other categories. Other classes may be 
distinguished, but with a lower degree of accuracy. 
5. To obtain more reliable information regarding the land use and land 
cover classes, SPOT images from several seasons are necessary. 
6. In terms of operational reliability, the per-pixel maximum likelihood 
classification of SPOT image data offers the most satisfactory 
results. The more accurately identified categories may be used as a 
framework for the addition of residual classes through a more 
conventional approach (photointerpretation). 
7. Overall classification accuracy may be misleading if considered by 
itself. Examining errors of commission and omission in conjunction 
with the overall classification accuracy permits full understanding 
of the maximum likelihood classification performance.