EM 1110-2-2907
1 October 2003
b. Description of Methods. Originally, this study intended to evaluate roof damage
caused by an actual emergency. In the absence of such an emergency, alternate imagery was
collected over a housing development under construction in Lakeland, Florida, located 30
miles (48 km) northeast of Tampa, Florida. The different phases of housing construction
provided an analog to roof damage during an event such as strong winds or a hurricane. The
different structural states of both residential and commercial roofs included exposed rafters,
exposed plywood, and plywood covered by tarpaper or shingles.
neighboring test areas in Lakeland, Florida. Roof conditions at individual buildings were
evaluated and geo-referenced. After the first flight, an assessment of the ground sampling
distance (GSD) and sensor data determined that a finer resolution would be required to ade-
quately examine roof condition. Two additional flights were then acquired, resulting in a
collection of data gathered at resolutions of 3, 2, and 1 ft (91.4-, 61-, and 30.5- cm respec-
tively), and 8-in (20.3 cm). Landscaping features, such as tree type and leaf on/off state,
were also documented with digital photos. This information was later used to establish the
feasibility in mapping vegetation using the Emerge system.
d. Sensor Data Acquisition. The two test sites, occupying 8 square miles (~21 km2),
were surveyed at several resolutions using Emerge imagery (see
Emerge System). Multiple resolutions were collected over a 2-month period. As a result, a
one-to-one comparison of the effect of resolution on image analysis was difficult, as house
construction in some areas was completed during the 2-month interval. The volume of data
collected was equivalent to that required for a 60 square mile (~155 km2) area, with ap-
proximately 25% image overlap (at a single resolution). This volume of data totaled 5 giga-
bits.
at a 1-ft and 8-in. (30.5 and 20.3 cm) resolution. At this resolution, plywood can be distin-
guished from other construction materials and individual rafters can be observed. Tarpaper
was not distinguishable from shingles owing to their spectral similarities.
(1) Despite the functionality of the 1-ft and 8-in (30.5 and 20.3 cm). resolutions, in
places with bright spectral response, saturation on the high end of the intensity scale low-
ered the resolvability of rafters relative to the flooring material. This was the result of a high
gain set for radiation detection within the sensor. Over-saturation lowers the contrast be-
tween rafters and the flooring, making it difficult to fully evaluate the condition of the roof.
Lowering radiation saturation requires collecting data during low to medium sun angle. This
may, however, delay data acquisition.
(2) Sun angle controls image contrast in two ways. First, a low sun angle may in-
crease shadowing, leading to a loss in target radiation data. Secondly, a high sun angle may
over-saturate the sensor. Both extremes were shown to lower contrast in this study, making
roof analysis difficult.
(3) A scatter plot breakdown of band 1 relative to band 2 was performed to evaluate
the possibility of automating an analysis that would delineate intact roofs and damaged
6-6
Previous Page
