FOREST FIRE DETECTION WITH SATELLITES FOR FIRE CONTROL
Yrjö Rauste
VTT Automation
Finland
Yrjo.Rauste@vtt.fi
Commision VII, Working Group 3
KEY WORDS: Forestry, Automation, Temperature, Real-time, Thermal, Wild-fire, AVHRR
ABSTRACT
A prototype software has been developed for automatic detection of forest fires using NOAA AVHRR (Advanced Very High
Resolution Radiometer) data. Fire detection is based on mid-infrared data (band 3 of AVHRR, 3.5 um). Bands 2 (near
infrared) and 4 (thermal infrared) are also used in the fire detection algorithm.
In an experiment in 1994, all six detected fires that were in areas where verification was possible were real fires. In 1995, at
least 17 out of 19 verified cases were high temperature targets. Three cases were industrial sites that can be excluded based on
their known location. This shows that satellite based detection of forest fires has potential for fire control purposes provided
that the supply of mid-infrared satellite data in day time is frequent.
1 INTRODUCTION
Satellite based wild fire detection has proven a viable tool
in collecting statistical information on the emission of green-
house gases in the past (see e.g. Setzer and Pereira, 1991).
The characteristics of the NOAA series of meteorological
satellites make it possible also to collect near real-time in-
formation in support of fire control activities (Flannigan and
Vonder Haar, 1986).
Forest fires is one of the most frequent type of natural haz-
ards in Finland. The Finnish Ministry of the Interior, VTT
Automation, and the Finnish Meteorological Institute started
a project to develop a prototype system for forest fire detec-
tion as a Finnish initiative in the context of the International
Decade for Natural Disaster Reduction (IDNDR) as declared
by the United Nations. The system is based on meteorologi-
cal satellite data. The real-time fire detection is done for fire
control purposes.
Many of the NOAA AVHRR -based fire detection studies con-
centrate on tropical areas (e.g. Kennedy et a/ 1994). In
tropical areas, surfaces with no vegetation such as deserts
or mountains usually warm so much in the afternoon hours
that the band-3 signal in AVHRR becomes saturated. In the
Boreal forest zone this phenomenon does not occur and there-
fore there is one possible error source less than in the tropical
areas or in the Mediterranean basin (e.g. lllera et al 1994 and
Gonzalez Alonso & Casanova Roque 1994).
Most fire detection algorithms make use of data in bands 3
and 4 of the AVHRR sensor. One fire detection algorithm —
based on the algorithm published by Kaufman et al (1990)
- is proposed by Kennedy et a/ (1994): (/) Bt: > 320 K,
(ii) Bts minus Bt4 7 15 K, (iij) Bt4 > 295 K, and (iv) ToA
reflectance in band 2 < 16 per cent where Bt; = brightness
temperature in band ? and ToA = Top of the atmosphere.
2 SATELLITE DATA SUPPLY
NOAA AVHRR data are received by the receiving station op-
erated by the Finnish Meteorological Institute. Of the five
wavelength bands present in the AVHRR sensor, three are
used in the fire detection system: band 2 (reflected infrared),
band 3 (middle infrared), and band 4 (thermal infrared). The
receiving station system extracts a sub-scene of 1024 lines by
584
1024 pixels from each satellite pass, which can be over 5000
lines by 2048 columns. The sub-scene is extracted in such
a way that its centre point coincides with the centre of the
defined monitoring area.
The extracted (3-band) sub-scene - together with the associ-
ated timing and calibration data — is then automatically sent
to the computer where the fire detection system runs via a
digital network. The fire detection system scans the direc-
tory of incoming image data at regular intervals. Every time
it notices a new data set in this directory it starts a processing
chain to find fires in the data set.
3 TRANSFORMATION OF NOAA AVHRR DATA
INTO FIRE INFORMATION
An overview of the derivation of fire information from NOAA
AVHRR data is shown in figure 1.
3.1 Pre-Processing of NOAA AVHRR Data
Location of a detected fire is of utmost importance if the fires
are detected for fire control activities. The NOAA AVHRR
data are therefore geo-coded before the fire detection. Geo-
coding is based on predicted orbit data.
Orbit telexes are received for satellites used once per day
(usually two to three days before the day of validity of the
telex). The orbit telexes are archived in the fire detection
system. Each time a new data set is fed into the system, the
most recent orbit telex of the satellite in question is used.
Many times the image reception from satellites is not per-
fect but the signal deteriorates during the tracking of the
satellite so that the data stream consists of random numbers
instead of data measured by the AVHRR sensor. These re-
ception errors are usually in blocks of one to ten consecutive
scan lines. The fire detection system checks the incoming
data set (by computing line averages for all three spectral
bands) and marks all lines where at least one of the three
line averages deviates from the scene average by more than a
pre-selected threshold. The detected reception-error lines are
then excluded from further processing.
Interpolation grids are used in image geo-coding. The geo
locations for a grid (approximately every 30th line and every
30th column) are computed based on the orbital data and
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
in
rey
low
are
