Schetselaar, Ernst 
4 DISCUSSION 
The classification on the basis of the TM channels shows a disproportionate performance in the prediction of basement 
and cover (85 % versus 52 %). This is largely the result of internal lithological variations in the cover that are 
misclassified as basement. These misclassified pixels are clearly reflected in the structural patterns that show-up in the 
classified patterns within cover (Figure 5 B). The classification of the magnetic data suggest a greater continuity of units 
in comparison to the TM classification and the geological map, possibly suggesting rapid transitions from unit to 
another in the near-subsurface. 
The TM classification fails to identify the mushroom interference patterns on the SW quadrant of the area, only 
showing thin slivers of cover. This could be a result of a widening of internal class spectral signatures. This would 
result in spectral overlap of basement and cover rocks, which could cause either class to be misrepresented. A 
geological facies change in the supracrustal suite or more paragneiss components in basement rocks could account for 
this discrepancy. 
An interesting observation that can be made form the results is that the classifications on the basis of magnetic data and 
TM data yielded similar coincidence percentages with the geological map, whereas their combined use only slightly 
improved the coincidence percentage. It essentially demonstrates that both data sets complement one another to only 
limited extent. This result would favour a classification approach where both data sets are used independently. 
Discrepancies with the geology map can than be tested against classification from independent remote sensing methods, 
reinforcing the significance of conflicting patterns. If, for example, both suggest a conflict with the geology map in an 
area between traverse lines, such an area would warrant attention in follow-up field studies. An additional advantage of 
this approach is that the magnetic data can be used to predict the basement-cover relationships underneath overburden. 
5 CONCLUSIONS 
* Combined data from Potential Fields Magnetic data from standard regional airborne surveys and TM multispectral 
satellite sensors can be used to give reasonable predictions of general bedrock classes. In this area 76 96 accuracy can be 
expected for supracrustal verses othogneiss terrain classification. 
e The TM data shows spectral contrast related to compositional variations between lithologies within the two structural 
levels that can be used to predict the basement and cover relationships. Misclassification of basement and cover in the 
TM data is likely due to spectral contrast both within the cover stratigraphic sequence, as well as lateral variations 
Within individual units. 
e The apparent magnetic susceptibility map provides a representation of surface and near subsurface geological 
patterns. Discrepant patterns with the geological map over several flight line intervals may be interpreted as areas where 
the basement-cover relationships occur at shallow depth. Although such an hypothesis requires further testing, such 
areas are potentially targets for magnetic inversion. In combination with structural observations from the surface this 
may prove useful to extend the mapping of the basement-cover relationships in the subsurface. 
e Classification on the basis of seven TM band and a single magnetic susceptibility channel yield about similar 
performance. This result support classification approaches for the region where the data sets are separately classified. 
Areas where discrepancies with geological map patterns appear in both classifications, provide interesting target areas 
for map refinement. 
e Pre-field, concurrent and post-field classification studies of this nature could be helpful during the planning stages of 
major field investigations, as an aid in targeting domains for detailed analysis and in extending well mapped areas 
beyond the training set. 
e Further studies are required to determine the minimal ground truth training set that will be able to give useful 
predications. 
e Further study is required to determine which additional discriminatory data layers, such as from advanced sensor 
technologies, will be able to enhance the predictive capacity of this method in the Arctic. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1331