207
■
In the Rhine River plain near Freiburg several fields were at water capacity. On
Pass 7, these fields produced a white specular return akin to that of settlements. In the
other passes, with larger incidence angles, such returns were absent from the imagery. In
mountainous areas distortions of terrain features (i.e. layover and foreshortening)
hampered location of the signal return on the maps as well as concealing (i.e. radar
shadow) settlements. Settlement detection was minimal in valleys parallel or
perpendicular to the look direction. The areal extent of settlements was frequently
difficult if not impossible to delineate, producing only a single bright return or a few
scattered points of high return.
Pass 1 had the largest incidence angle (52.3 degrees), and the lowest number of
commission errors (18)...but not the highest accuracy (62.7 percent) due to errors of
omission (277). Settlements did not appear as sharp spots of high return but, typically, as
mottled, medium to light gray textured areas. This characteristic allowed the areal
extent of the settlement to be observed in some cases, particularly as the settlement size
increased. Ford et.al. (1986) stated that as incidence angle increases there is increased
sensitivity to surface roughness and decreased sensitivity to topography. Although the
authors were alluding to small, detailed surface roughness parameters these three 5IR-B
data takes may indicate that a similar pattern exists for such man-made surfaces as
settlements.
Among the detection errors were road intersections, a hydro-electric power dam,
and some water/wetland areas. In the latter cases ground checking indicated that the high
return was originating from the banks of geometrically shaped reservoirs. In wooded and
forested areas of the Black Forest detection of very small (4-6 structures) settlements
was not possible; relief and dissected terrain inhibited detection. Settlements abutting
the edge of forest stands on the fore- and back slopes were also masked by the forest
return and omitted from the count on the image.
String settlements (i.e., houses aligned on one or both sides of a road or stream) were
not detected. In contrast to Pass 2, settlements in valleys parallel to the look direction
were visible, and there was a notable difference in visibility as a function of range. In this
case, settlements in the near range were less visible than those located in mid- and far
range. It is hypothesized that the latter settlements were more sensitive to surface
scatter and presented a rougher surface to the radar signal. However, no quantitative
evaluation was attempted to define the precise incidence angle/radar return relationship.
Mountainous terrain remained areas of low detectability with the visibility
increasing as settlement size increased. Accuracy was high here as there was little
chance for confusion (i.e. commission error)--a settlement was visible or it was not.
Concomitantly, the per cent of settlements visible was very low. This relationship is
discussed in the analysis of forest and mountainous areas later in the paper.