Fire Detection using
Geostationary and Polar
Orbiting Satellites
Dr. Bernadette Connell
CIRA/CSU/RAMMT
Dr. Vilma Castro
UCR/RMTC
March 2005
Objectives
• Background
• Environmental and weather conditions
conducive to fires
• Satellite fire detection techniques for hot
spots
• Examples
• Lab exercise
Monitoring Fire Activity
Why?
• To detect and monitor wildfires in real-time for
response and mitigation.
– Are the fires posing danger to population centers
or economic resources?
• To determine trends in fire activity from year
to year.
– Are they the result of agriculture burning and
deforestation?
– Are they the result of a buildup of fuels?
– Are they affected by drought?
• To determine the extent of smoke transport
• To determine the effect of burning on the
environment.
United States - Fire Weather Activities
• Various FIRE DANGER RATING systems have
been developed to express fire hazard.
They incorporate some of these basic questions:
• Are the “fuels” dry enough to burn?
• Is the current or forecast weather conducive to
starting fires and sustaining them?
– Is it dry, windy?
– Is the atmosphere stable or unstable?
– Will there be lightning with very little rain?
United States - Fire Weather Activities
• To address the condition of fuels:
– Long term monitoring for drought (satellite)
– Monitoring of vegetation health and accumulation of dead
vegetation (fuels) (satellite and ground)
• To address weather conditions:
– Outlooks for precipitation and temperature
(climatology/model prognosis)
• Information Sources:
– Climate Prediction Center (CPC)
– USDA Forest Service
– NOAA/NESDIS/ORA
Real-time NWS Fire Weather Services
• Storm Prediction Center – issues 1 and 2
day fire outlooks
http://www.spc.noaa.gov/products/fire_wx
– maps
– text discussion
– hazard categores:
• critical areas – outlines
• extremely critical – hatched
• dry thunderstorm risk - scalloped
Real-time NWS Fire Weather Services
• Weather Forecast Offices – issues fire
weather forecasts/watches, smoke
forecasts, red flag warnings, spot
forecasts
• IMET – Incident METeorological
information for fire behavior forecasts, spot
forecasts, nowcasts
Real-time (non-routine) Products
• Fire Weather Watch; valid 24-48 hr
– 1-min sustained winds at 20 ft. > 15-25 kts
– Relative humidity < threshold (see following slide –
varies by region)
– Temperature >65-75°F
– Vegetation moisture <8-12%
• Red Flag Warning: valid 0-24 hr
– Same criteria as Fire Weather Watch (above)
• “Spot” Forecasts
– Forecasts for prescribed burns, rescues, wildfires in
progress
Threshold Relative Humidities for Red Flag Watches/Warnings
Haines Index
• This index is correlated with fire growth in plume
dominated fires
• Composed of two parts:
– stability: temperature difference between two atmospheric layers
near the surface
– moisture: temperature/dew point difference for that layer
• The index is adaptable for varying elevation regimes
• Index value estimates rate of spread:
2-3: Very Low Potential (Moist Stable Lower Atmosphere)
4: Low Potential
5: Moderate Potential
6: High Potential ( Dry Unstable Lower Atmosphere)
Calculating Haines Index
LOW ELEVATION
<2,000 FT
Stability Term (T950T850)
1… 3 C or less
2… 4 to 7 C
3… >= 8 C
Moisture Term
(T850-Td 850)
1… 5 C or less
2… 6 to 9 C
3… >= 10 C
MID ELEVATION
2,000-6,000 FT
Stability Term (T850T700)
1… 5 C or less
2… 6 to 10 C
3… >= 11 C
Moisture Term
(T850-Td 850)
1… 5 C or less
2… 6 to 12 C
3… >= 13 C
HIGH ELEVATION
>6,000 FT
Stability Term (T700T500)
1… 17 C or less
2… 18 to 21 C
3… >= 22 C
Moisture Term
(T700-Td 700)
1… 14 C or less
2… 15 to 20 C
3… >= 21 C
Sum of two terms = Haines Index
GOES Fire Detection - VISITview
2-very low
3-very low
4-low
5-moderate
6-high
water
U.S. Drought Monitor – Severity Classification
Category
Description
Fire Risk
Palmer
Drought
Index
D0
Abnormally
Dry
Above
average
-1.0 to
-1.9
D1
Moderate
Drought
High
-2.0 to
-2.9
D2
Severe
Drought
Very high
-3.0 to
-3.9
D3
Extreme
Drought
Extreme
-4.0 to
-4.9
D4
Exceptional
Drought
Exceptional
and
Widespread
GOES Fire Detection - VISITview
< -5.0
CPC Soil
Moisure
(percentiles)
21-30
11-20
6-10
3.5
0-2
Weekly
Streamflow
(percentiles)
% of
Normal
Precip
Standardized
Precipitation
Index
Satellite
Vegetation
Health Index
21-30
<75%
for 3
months
-0.5 to -0.7
36-45
11-20
<70%
for 3
months
-0.8 to -1.2
26-35
6-10
<65%
for 6
months
-1.3 to -1.5
16-25
3-5
<60%
for 6
months
-1.6 to -1.9
6-15
0-2
<65%
for 12
months
< -2.0
1-5
Vegetation
Health
• Showing vegetation
health for this year
compared with last year.
• Fire becomes a concern
when the vegetation is
stressed (values less than
50) and when drought and
other weather is of concern.
Loop of plume dominated fire
VIS 03246
IR2 03246
British Columbia
Alberta
Montana
Idaho
Washington
Oregon
Loop of wind driven fire
California
VIS
IR2
IR2 24hr
Mexico
Satellite Monitoring of FIRES
Geostationary or Polar Orbiting?
• Monitoring from both types of satellites utilize
visible, shortwave, and longwave infrared
channel observations.
• Geostationary Satellites (GOES)
– Coarser resolution (~4km)
– Good temporal resolution (every 15-30 min.)
which provides information on the diurnal timing
and spatial distribution of fires.
– Saturation brightness temperature: 338K (for GOES-8
and 12)
• Polar Orbiting Satellites (AVHRR)
– Finer resolution (~1km)
– Only 2 passes per day
– Saturation brightness temperature: 320 K
“Quick” RAMSDIS Products for
fire detection
NIGHT: Fog-Stratus Product
DAY:
Reflectivity Product
These products are made with images
from channels
3.9 and 10.7 µm
Characteristics of 3.9 micrometer channel that make
it suitable for “hot” spot detection
200
3
wavelength = 3.9 um
2
1
Radiance (mW/(m2.sr.cm-1)
4
Radiance (mW/(m2.sr.cm-1)
Radiance is not linear
with temperature
• A small change in
radiance at 300 K
at 3.9 um creates a
larger change in
temperature than at
10.7 um
(note the different
scales:
3.9 um from 0-4
10.7 um from 0-200
150
wavelength = 10.7 um
100
50
0
180
220 260 300
Temperature (K)
340
0
180
220 260 300
Temperature (K)
340
Sub pixel response
• Rλ = Rλ cloud * % area cloud + Rλ ground * % area ground
• Similarly for fires:
Rλ = Rλ fire * % area fire + Rλ ground * % area ground
GOES 3.9 um Channel Tutorial
NIGHT:
Fog-Stratus Product
Subtract temperature, pixel by pixel, of:
10.7mm - 3.9 mm images
As temperature is
warmer at 3.9 mm
The result is a
negative number
NIGHT:
Fog-Stratus Product
• The result is normalized by adding
150 to each pixel’s value
• Values correspond to a scale of 0.1
K per brightness unit
In a black and white color table, pixels with
fire appear darker than the background
NIGHT:
Fog-Stratus Product
Pixels with fire are 80 brightness units
darker than the background
Observations:
1 brightness unit
80 brightness units
= 0.1 Kelvin
=8K
Temperature difference
among pixels without fire: 3 K
4- 6 K Difference among pixels:
fire cannot be detected with certainty
DAY:
Reflectivity Product
• Channels involved: 3.9 and 10.7 microns
• Reflective component is subtracted from the
3.9 micron signal.
The temperature at 10.7 microns is used to
estimate the reflective component at 3.9
microns
• Fires appear as white spots
• Do not need to set thresholds
• Limited to daytime use
Reflectivity Product
Observations
• Products allow the identification of fires
smaller than a pixel
• Weaver et al. show that it is possible to
detect:
– 500K fires against a 300K background
– covering only 5 % of a 4 x 4 km pixel
Weaver, J.F., Purdom, J.F.W, and Schneider, T.L. 1995.
Observing forest fires with the GOES-8, 3.9 µm imaging
channel. Weather and Forecasting, 10, 803-808
Observations
• Can the visible channel be used to detect fire?
Yes. The smoke plume can be seen in the visible.
However: Fire must be well developed to create
a plume that can be detected in the visible.
Types of Fire Detection
Algorithms
• Fixed threshold techniques
– Rely on pre-set thresholds and consider a single pixel
at a time.
• Spatial analysis or contextual techniques
– Compute relative thresholds based on statistics
calculated from neighboring pixels.
Real-time products for Central America:
http://www.cira.colostate.edu/ramm/sica/main.html
Example of Fixed Threshold
Algorithm by Arino et al. (1993)
1. BT3.9 > 320 K (to identify probable fires)
2. BT3.9 – BT10.7 > 15 K
3. BT10.7 > 245 K (to prevent false alarms
due to reflective clouds)
GOES-83.93.9
micrometer
GOES-8
micron
GOES-8 3.9 micrometer
Blue areas represent pixels:
T3.9 >320K
GOES-8 Product: T3.9 – T10.7
Blue regions represent pixels with:
T3.9 – T10.7 > 15 K
Resulting Fire Threshold Product
Blue represents fire pixels
Problems
• Very warm, dry ground is detected as fire.
• Will not pick up night fires that are cooler
than 320 K
Example of Contextual Algorithm by
Justice et al. (1996)
1.
2.
BT3.9 > 316 K (to identify probable fires)
Estimate a background temperature with
surrounding ‘valid’ pixels:
A valid pixel
* Is not a cloud
* Is not a fire pixel
3. The window starts as a 3X3 pixel area and
expands to a 21X21 pixel grid until at least 25% of
the background pixels (or at least 3) are valid.
4. Let DT=MAX(2 std dev of BT3.9-BT10.7, 5 K)
FIRE pixel:
if BT3.9-BT10.7 > mean BT3.9-BT10.7 + DT
and BT10.7 > mean BT10.7 – std dev of BT10.7
Fire Justice Product
Blue pixels represent detected fires
Problems
• Does not adequately detect fire pixels in
regions of very warm and dry ground.
• May also need to implement a correction
for (horizontal) temperature changes in
mountainous regions.
• Will not pick up night fires that are cooler
than 316 K
GOES- 8 Reflectivity Product
Shot-noise filter applied to Reflectivity Product
Red pixels denote potential fires.
Experimental ABBA
• Automated Biomass Burning Algorithm
• Contextual Algorithm
• Developed at the Cooperative Institute for
Meteorological Satellite Studies (CIMSS) at
the University of Wisconsin in Madison.
• Initially ‘calibrated’ to Brazil Fires
http://cimss.ssec.wisc.edu/goes/burn/wfabba.html
Polar Orbiting Satellites
• The same detection algorithms presented
here can be applied to imagery from polar
orbiting satellites.
• For AVHRR, the 3.9 um sensor saturates
at ~ 323 K
(GOES-8 saturates at 338 K)
• We will view an example of GOES vs.
AVHRR imagery in the lab.
References/links
GOES Fire Detection – VISITview session
http://www.cira.colostate.edu/ramm/visit/detection.html
see reference/links at the bottom of their page
Fire Products for Central America
http://www.cira.colostate.edu/ramm/sica/main.html
Wildfire ABBA
http://cimss.ssec.wisc.edu/goes/burn/wfabba.html
CIRA GOES 3.9 um Channel Tutorial
http://www.cira.colostate.edu/ramm/goes39/cover.htm
Storm Prediction Center – 1 and 2 day fire outlooks
http://www.spc.noaa.gov/products/fire_wx
Drought Monitor - long term drought indicators for the US:
Drought Index, Crop Moisture Index, Standardized Precipitation Index, Percent of
Normal Rainfall, Daily Streamflow, Snowpack, Soil Moisture, Vegetation Health
http://drought.unl.edu/dm