1. SCIENCE AND TECHNOLOGY SUMMARY
Accurately measuring the heat energy released from a wildland fire
is essential to determining biological, ecological and meteorological effects.
The heat release has been estimated using visual inspections of the post-burn
area (NWCG (1994)), measurement of peak soil temperatures, examination of
post-fire soil conditions, measurement of fuel consumption, single overhead
infrared images of the fire (captured with satellite and/or airborne image
sensors) and other methods both biological and physical. All of these methods suffer from a
number of shortcomings – requirements for an expert observer, sparse
spatial and temporal sampling, and a lack of real time in-fire calibration
(ground truth) to determine a reliable, physical number for total energy
release.
Of the methods used to determine heat release, only airborne infrared
(IR) remote sensing can provide the necessary spatial and temporal resolution
and synoptic view to be able to quantify wildland fire heat release. Such infrared methods have been used for
many years in both analytic and tactical frameworks (Wilson (1971)). Riggans (2004) has described a three-step
method for measurement of energy release from wildland fires using the
following method:
1)
Infer
effective fire pixel temperature using two-band IR imagery
2)
Estimate the
emissivity/area product from the measured sensor-reaching radiance using the inferred
temperature from 1)
3)
Calculate
the wavelength integrated radiant flux density from 1) and 2) above using the wavelength-integrated Boltzmamn
equation:
Where P is the total flux density (W/m2) e is the emissivity of the target area, s is the Stefan-Boltzman constant, A is the
fire or pixel area and T is the temperature of the fire as inferred from 1).
A major problem with this method arises in step 1, where an
estimate of the fire temperature is made assuming that a black body spectral energy
distribution reaches the sensor.
We expect atmospheric absorption from water vapor, carbon dioxide and
other combustion products and emission from combustion products to affect the
sensor reaching radiance and thus the inferred temperature. Since the flux calculated in step 3 is a sensitive function
of temperature, small errors in temperature measurement caused by distorted
spectra can produce large errors in predicted flux. Although the inferred temperatures from these prior
experiments agreed fairly well with laboratory measurements of fire
temperatures, without more detailed spectral characterization it is difficult
to place error bounds on the thermal flux calculated. We thus sought to develop
a better method to measure remotely sensed energy release that was independent of
temperature measurements (or even the assumption of an equilibrium temperature)
or spectral modeling.
Heat release maps of a prescribed burn in SE Ohio were created
from images captured by a multi-band infrared camera constructed by the Rochester
Institute of Technology Wildfire Airborne Sensor Program (WASP). (see Figure 1)
The entire fire event (~3 hrs duration) was captured in multiple frames with
typical sampling interval of 5 -10 min.
The infrared flux calculated from these frames was time-integrated,
producing a map of total heat release, and calibrated using several
ground-based infrared sensors in the fireground that measured thermal flux
emitted from the surface. As the particular sampling interval (~ 300 seconds) gives
a discontinuous description of fire behavior with insufficient time resolution
to observe dynamic fire effects and cannot provide exact heat output for all
points within the burned area, we have developed an extrapolation method that
obtains an upper limit and lower limit to the released thermal energy based on
the available imagery, temporal sampling rates and observed thermal decay
rates. The basic method for
construction of a heat map is as follows;
1)
Capture a
time series of airborne IR images (using the WASP imager).
2)
Simultaneously
capture time-resolved ground calibration data and burn plot samples (pre- and
post- fire).
3)
Use the
ground calibration data to absolutely calibrate the image and remove the
effects of atmospheric propagation. This method is similar to a ground truth
based calibration method described in Schott (1995).
4)
Geo- and
ortho-rectify images and create single-frame mosaics for each time interval.
5)
For each
pixel in the mosaic image, obtain a cooling rate for that pixel by observing
the flux difference between successive frames after the passage of a fire
front.
6)
Time
integrate the images from 4) to produce the total energy release map:
where L = total energy leaving a pixel, Dn
= digital data value from the image, Dt = time interval between frames, C = calibration derived from
ground sensors, f = frame number, n = total number of frames. This process is repeated for all
pixels. Since there are a limited
number of ground sensors, we use the calibration value C derived from the
ground sensor closest to the pixel in the summation.
This gives the lower bound for the energy emitted from
the fire for each pixel.
7)
For each frame in the
time sequence image, extrapolate the energy flux back in time using the cooling
rate calculated in 5).
8)
Time integrate the
images from 7) using the technique described in 6). This is the upper bound for the heat release from the fire.
We compare the heat release map produced with this method to more
than 40 ground-sample stations in an effort to predict fuel consumption and
other fire effects parameters. There
was excellent agreement between fuel burn up (weight of fuel in a sample plot
before and after passage of the fire front) and the energy release map
generated by this method. For the
case of no energy released (fuel unburned), Table 1 shows the agreement for 9
sample plots between our energy release map and ground sampling. Figure 2 shows
an expanded image detail of two of the 9 unburned sample plots. As this method uses ground sampling to
calibrate the overhead imagery, the upwelled flux from the scene, which could
include significant energy released by hot gases above the fire, is
ignored. This method therefore is
best suited to smaller fires with low energy contained in the smoke and gas
plume. Future experiments will
include an upwelled fluence sensor to measure the upwelled radiance and eliminate
this shortcoming.
Figure 1 - A thermal flux map for a prescribed fire in Vinton
Furnace, Ohio. This map was
prepared from 11 time-sequenced frames using the technique described in the
text.
|
Sample Plot Name |
Energy release from image |
|
S1 |
0 (unburned) |
|
S2 |
0 (unburned) |
|
N5 |
0 (unburned) |
|
N6 |
0 (unburned) |
|
N9 |
0 (unburned) |
|
N10 |
0 (unburned) |
|
N12 |
0 (unburned) |
|
R1 |
0 (unburned) |
|
T15 |
0 (unburned) |
Table 1 - Comparison of heat release as
calculated by the method described in the test and ground truth. All these samples were unburned as
determined by ground truth physical examinations.
Figure 2 - Sub-images from the energy
release map showing unburned areas around two ground-sampled plots. These areas had no fuel consumption and
no visible fire effects during post-fire ground truth examinations.
2. CONCLUSIONS
We have developed a method of measuring radiant energy release
from wildland fires that is independent of atmospheric conditions, assumptions
about equilibrium and thermal properties of the fire, and models relating
temperature to total energy flux. This method gives a lower and upper bound for
fire energy release. We are extending this method to include energy released in
the upwelled column above each pixel, more ground calibration stations and better
modeling of the energy decay rate after the passage of a fire front. Excellent agreement has been obtained
with conventional field survey methods for 40 plots in a 40 hectare in a
prescribed fire.
This work has been performed under NASA contract NAG13-02051, and
with support from the National Fire Plan, whose support is greatly appreciated.
3. BIBLIOGRAPHY
Kremens,
R.; Faulring, J.; Gallagher, A.; Seema, A.; Vodacek, A. 2003a. Autonomous
field-
deployable wildland fire sensors. Int. J. of Wildland Fire. (12):
237-244
Kremens,
R.; Faulring, J.; Hardy, C. 2003b. Measurement of the time-temperature and
emissivity history of the burn scar for remote sensing applications. 2nd
International Wildland Fire Ecology and Fire Management Congress, 16-20 November,
Orlando FL. American Meteorological Society
National Wildfire Coordinating Group (NWCG), 1994: Fire Effects
Guide. NFES 2394.
Riggans, P.J., Tissel, R.G., Lockwood, R. N., Brass, J.A.,
Perreira, J.A.R., Miranda, H.S., Miranda, A.C. 2004: Remote measurement of
energy and carbon flux from wildfires in Brazil. Eco. Appl. 14(3) 855-872.
Schott, J.A. (1995) Remote Sensing, An Image Chain Approach.
Springer Publishing.
Wilson, R.A., Hirsch, S.N., Madden F.H., Losensky, B.J., 1971:
Airborne infrared forest fire detection system: final report. USDA Forest
Research Paper, INT-93.
4. THE AUTHOR
Robert
Kremens has been employed by the Rochester Institute of Technology (RIT) at the
Center for Imaging Science since 2000. At RIT he has specialized in physical
measurements of wildland fires for remote sensing and fire behavior purposes
and in constructing new airborne and ground based systems to monitor the
environment and wildland fire.
Dr. Kremens received his B.S. (1975) in physics form the Cooper Union,
M.S. and Ph.D. in physics from New York University, (1981) and a M.S. in
environmental studies from the University of Rochester (2000). Before RIT, Dr. Kremens worked at the
University of Rochester on the worldÕs largest pulsed laser, building and
analyzing nuclear and laser diagnostics on an inertial fusion experiment, at
the United States Army Ballistics Research Laboratory in Aberdeen MD, and at
several industrial positions where he designed and constructed high speed data
acquisition and imaging systems.