A Burn Scar Mapping Algorithm for MODIS Direct Broadcast Data

1
A Burn Scar Mapping Algorithm for MODIS Direct
Broadcast Data

• J. Meghan Salmon, W. M. Hao, and B. Nordgren
• USDA Forest Service, Rocky Mountain Research
  Station, Fire Sciences Laboratory, Missoula,
  Montana

Introduction At the Missoula Fire Sciences
Laboratory, a MODIS DB station is in place to
demonstrate the potential for monitoring forest
fires in near-real-time and predicting their
effects on air quality downstream. As building
blocks of this system, we extract active fire
locations (MOD14-4.3.2) and burned area
detections (Li et. al. 2004). The former are
visible on our website, mapped over a
course-scale resampling of each pass for the day
(www.firelab.org). Li, Rong-Rong, Y.
Kaufman, W. M. Hao, J. M. Salmon, and B.-C. Gao.
June 2004. A Technique for Detecting Burn Scars
Using MODIS Data. IEEE Transactions on Geoscience
and Remote Sensing, V. 42 N. 6, pp. 1300-1308.
Algorithm Description The burnscar algorithm has
multiple layers. The first is a spectral test for
the signature characteristic of a recently-burned
area. For this, we use the algorithm outlined in
Li et al 2004, but we have found it to have a
prohibitively high false alarm rate. We have
therefore added a subsequent tests. The first is
a convolution filter to remove small clusters of
detections (less than 6 in a 5x5 pixel box).
Next, we filter on the minimum distance to an
active fire detected in the past ten days
(eliminate when dmin gt 5km). Finally, the active
fire and burnscar detections are aggregated
together to form fire perimeters for each burning
event. These perimeters grow with the addition of
detections, merge together when appropriate, and
become inactive after 10 days pass without any
active fire detections.


The key element in implementing this burnscar
mapping method was storing all fire information
(hotspots, potential burnscar detections, and
fire perimeters) as geographic objects in a
PostGIS database (http//postgis.refractions.net/)
. This allows the spatial and temporal searches
required for generating time-series of fire
perimeters.
Results We compared the algorithm results to
infrared mapped fire perimeters from the I-90
Complex incident near Missoula, Montana August
4-21. In some cases, the MODIS 250m corrected
reflectance product is shown in the background,
and in others a topographic map is shown instead.
When available, the fire perimeter from the
evenings IR flight is shown in orange.
Conclusions Monitoring the algorithms
performance during the 2005 fire season, and
comparing its results to data from fires in Idaho
and Western Montana has led us to conclude that
it effectively maps fire growth in near-real
time. A complete validation is still necessary,
and planned for winter 2006. The above
perimeters, when combined with a fuel type map
and FOFEM (First Order Fire Effects Model)
software, allowed us to generate daily emissions
estimates from the I-90 Complex, shown at left.