Geology Reference
In-Depth Information
Fire-Area Estimation
R emote sensing-based fire-area estimates are close approximations of the actual fire area. The reason why
they are only approximations is because the term fire area is not rigorously defined. In the field, the fire
area may mean the actual area showing visible flames and excessively high temperatures. In remote sensing,
however, the fire area implies the area showing a surface thermal anomaly. Thermally anomalous area have
pixels that exhibit temperatures higher than a preselected background temperature. The remote sensing
based area estimates are therefore a function of the pixel size (related to spatial resolution of the images) and
the user-defined threshold of the background temperature value.
Where and how to set a threshold to differentiate background areas and fire areas is a subject of debate and
continued research. Threshold selection has evolved from using a general value based on trial and error (Prakash
et al., 1997), to taking a statistical threshold set at a standard deviation of two (Prakash et al., 1999), to site
specific threshold using the shape of the slope of the histogram of the image subset for that site (Rosema et al.,
1999), to taking the first local minimum after the main maximum digital value during several iterations of a
moving window kernel (Kuenzer et al., 2007a; Figure 14.4.3). However, given the complex nature and
differences among fire sites, no one technique is known to work well in all situations to delineate fire areas
from remote sensing images.
A
B
Threshold
Threshold
2
σ
DN values
DN values
C
D
Threshold
Threshold
DN values
DN values
Figure 14.4.3. A conceptual figure showing histograms of a thermal infrared image or an image subset. The
Digital Number (DN) value range is shown on the x axis. On the y axis is the frequency or number of
occurrences of a specific DN value. Different criteria for selecting a threshold to delineate thermal anomalies
from background temperatures are depicted: (A) threshold is selected by trial-and-error and guided by field
data; (B) threshold is determined using a statistical parameter, in this case a standard deviation of 2; (C) the
location of the change in slope of the histogram is identified and projected on to the
-axis to determine
threshold; (D) the first minimum-value dip after a major DN value maxima is selected as the threshold. Figure
by Anupma Prakash and Rudiger Gens, this work.
x
Search WWH ::




Custom Search