'rance
NG
actors
, Such
uman
>, the
raphic
om a
maps
ent of
s will
areas
loped
ent of
GIS-
ls are
rea at
meter
ool et
1989).
s with
maps
5).
s the
und a
| and
istics,
s and
everal
emote
rough
'orked
azar,
iences
OAA-
inally,
entary
ularly
ve to
anopy
The criteria followed for the combination of
these variables may be summarised in the
following alternatives: (i) Use of qualitative
criteria for assigning danger values to the
cross-relationships of the different variables
(Yool et al., 1985); (ii) Adaptation of standard
danger indices, such as the US National Fire
Danger Rating System or some modules of
BEHAVE (Woods and Gossette, 1992:
Chuvieco and Salas, 1996); (iii) Development
of new danger models, based upon the selective
weighting of the danger variables (Chuvieco
and Congalton, 1989), and (iv) Creation of
locally-oriented models, where danger weights
for each variable are obtained from multiple
regressions computed for that particular area
(Chou, 1992).
4. FIRE DETECTION
Fire detection through remote sensing has been
based on middle infrared data analysis.
Considering that forest fires temperatures
commonly range from 500 to 1,000 K
(Robinson, 1991), the most suitable band for
fire detection is located between 5.8 and 2.9
um according to Wien's displacement law
(middle infrared region of the spectrum), while
the thermal infrared region presents the peak
of emittance at common Earth temperatures
(around 300 K). As a consequence, the middle
infrared bands are more sensitive than thermal
infrared bands to detect and monitor active
fires.
Fire detection from space is obviously very
much dependent on temporal resolution. The
Earth resources satellites (such as Landsat or
SPOT) do not provide enough. temporal
frequency for fire detection. On the contrary,
meteorological satellites have proven to be very
useful for these purposes. NOAA-AVHRR
images are suitable for fire detection and
mapping because of their adequate coverage
cycle (12 hours) and good spectral resolution,
which also includes middle infrared bands.
47
The use of AVHRR images for fire detection
has been successfully tested in several studies,
both at regional and global scale, specially over
remote areas where traditional methods are
very costly. In Canada, high accuracy for
detecting large-scale forest fires was found in a
pilot study conducted in Alberta (Flannigan and
Vonder Haar, 1986). Accuracy for small fires
was limited, 10 to 12 %, although better results
were reported if only cloud-free areas were
taken into account (up to 87 %). Similar
studies have been carried out in Tropical forest
(Malingreau, 1990; Langaas, 1992; Kennedy et
al., 1994; Setzer and Pereira, 1991). Some
experiences are also available over
Mediterranean forest (Chuvieco and Martin,
1994b).
In spite of the potential interest in the use of
AVHRR channel 3 data for fire detection, these
images present several difficulties related to the
low thermal sensitivity of this channel, which
is saturated at 320 K. As a result, fire spots can
be easily confused with agriculture burns or
even overheat bare soils, which frequently
reach these temperatures during the summer at
the afternoon satellite pass (Belward, 1991).
Discrimination from agricultural fires could be
partially achieved by choosing evening or night
images, because this type of burning tends to
be done during daylight periods (Malingreau,
1990). Monitoring the temporal dynamism of
the target surfaces also provides good
classification of fire pixels (Lee and Tang,
1990). In any case, sensors with higher thermal
sensitivity are desirable. In the Brazilian
Amazonia, airborne experimental fire detection
scanners with saturation levels up to 900 K
have been successfully. tested (Riggan et al.,
1993). The future Moderate Resolution
Imaging Spectroradiometer (MODIS) will
include a middle infrared channel that saturates
at 500 K, which will notably increase the
potential of satellite fire detection systems.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B6. Vienna 1996
