Evaluation of New Sensors for Emergency Management
19
differences due to shadowing (Lunetta 1998). These expectations were supported
in the resulting scatter plots that were nondirectional bright-end clusters (denot-
ing damaged or under-construction roofs, shown in green) and dark-to-bright
linear clusters (denoting intact roofs, shown in red), respectively (see Figure 9).
The scatter plots suggest an automated approach to counting the damaged house
roofs, given that the locations of the buildings are known ahead of time and that
monochrome shingles are used on house roofs. The automated process would
require counting all the abnormal roof scatter plots within the damage swath
using a pre-existing GIS coverage of all roofs as a mask. This technique is not
expected to work in areas with roofs that have multicolored shingles where the
spatial color changes are larger than a pixel size.
Sun angle
The angle of the sun with respect to the local vertical makes a difference in the
detail seen in the imagery. As noted above, the effects of shadowing increase the
length of the linear scatter plots taken of intact roofs. The length is directly
related to an increase in visible detail due to scene contrast. Assuming that the
imagery is taken with a sensor that is pointing near nadir, if there is a medium
sun angle, the sensor will "see" both lighted and shadowed portions of objects at
discrete sites within an area. Our study has revealed that 2-ft, 1-ft, and 8-in. ima-
gery is useful for roof damage assessment and so should be taken with medium
sun angle when possible. The 3-ft imagery is primarily used as a location tool. It
is best taken with an overhead sun angle where maximum illumination of an
entire area is the goal.
Broad leaf/narrow leaf
Additional analyses were carried out on the EMERGE imagery of Lakeland,
Florida, to evaluate its capability for discriminating vegetation. At site 5 in area
2, we evaluated the discrimination between broadleaf and narrow-leaf (ever-
green) tree species. It was found that neither spectral nor spatial resolution was
sufficient to permit this distinction. However, the presence or absence of leaves
(leaf on/leaf off) was found to be visibly distinguishable, as the deciduous trees
were bare when the imagery was acquired in February and fully leafed out in the
March scenes. Palm trees were always visibly identifiable by their distinctive
fronds in the 8-in. GSD imagery.