206
seen in Figure 1, each of the data takes overlapped in one limited area, but each also
recorded relatively extensive areas of terrain (i.e.. Data Take 97.4--8484 sq. km.; Data
Take 93.2--4148 sq. km.; and Data Take 45.2--7600 sq. km.). A changing and varied
mixture of forest and field landscape patterns, topography, and numerous settlements in a
northern boreal forest environment comprised the study area for this investigation.
METHODOLOGY
The 1:500,000 master positive transparencies (first generation) of the imagery were
optically enlarged (2X to 8X) and all sites thought to be settlements were recorded on
transparent overlays. The overlays were then registered to maps at a scale of 1:200,000.
For purposes of this study, "settlements" were defined as all clusters of 4 or more
structures and dwellings indicated by individual structure symbols and/or designated by
placename on the 1:200,000 topographic maps.
Errors of omission and commission were tabulated and their locations noted on the
maps. In some instances topographic maps at scales of 1:100,000 and 1:50,000 were also
consulted for error source identification. The spatial patterns of errors and correct
identifications were examined by referring to and comparing the radar imagery, overlays
and maps. Field work was also conducted be traversing approximately 75 per cent of the
entire study area and noting site and causes of errors.
After preliminary analysis of all three data sets (but prior to error analysis) a second
interpretation was completed using only Data Takes 45.2 and 97.4. As these two sets
slightly overlapped and ran almost perpendicular to each other (see Figure 1) the purpose
was to compare the effect of incidence angle and look direction over similar terrain and
topographic conditions. In this instance, only those areas comprised of forested hills and
mountains with intervening meadow areas were included. That is, this simple landscape
pattern reduced the environmental variables in the data set. The simplified environment
allowed a comparison of the effect of look direction with regard to terrain slopes and an
examination of detectability under conditions where the high specular return of settlement
structures in open areas and clearings would provide high contrast to the uniform
medium-to-dark gray surrounding forest cover and fields.
ANALYSIS AND RESULTS
The results of the interpretation of the three SIR-B data takes are found in Table II.
It is apparent that Pass 2 (Take 93.2) with the smallest angle of incidence (15.7 degrees)
was the poorest image in terms of percent detection (44.9 %) and in absolute
accuracy--27.7% (i.e., the greatest number of commission errors). The steep incidence
angle of this pass produced a great many specular returns from non-settlement targets
including road intersections, some hill slopes, quarries, and open water/wetland areas.
These results were the exact opposite of that reported by Sieber and Hartl (1986).
However, the present study incorporated a much larger sample size and study area and
concentrated only on settlement detection.