hase. Be. 
generated 
nt by tele. 
There the 
the fax to 
located or 
redirected 
Detween 4 
4 000 hot 
real fires 
1 the area 
ning fires 
1 1995 in 
ari 
© CO rT? 
L0 DU Uo ovo roo © Wm 
ndicated 
mn V of 
from the 
ılts were 
wer was 
untered. 
es where 
jon Was 
analysis 
  
  
  
  
  
  
  
dn e 
Fires detected 199 
  
  
N € 
  
Figure 3: Fires detected in experiments in 1994 (left) and 1995 (right). 
of the performance of the false-alarm criteria can be found in 
(Rauste et al 1996). 
5 DISCUSSION 
Most of the hot spots identified as false alarms were in those 
areas of the input scene where the imaging geometry is close 
to specular reflection. If data from AVHRR or a similar sensor 
(a wide field of view, pointed to nadir) is used these areas 
cannot be avoided in the input data stream. The imaging 
geometry can be used (as done in this study) to find false 
alarms due to reflections from clouds or (possibly sub-pixel) 
water surfaces. If a satellite system is designed specifically 
for fire detection, this type of false alarms can be avoided 
for instance by tilting the sensor axis away from sun when 
sun-synchronous orbits are used. 
The experiments with the automatic fire detection system 
have shown that the total response time of the system (from 
the start of the fire to the delivery of alert message to the 
local fire authorities) is a critical factor when evaluating the 
feasibility of the system. The total response time can be 
enhanced 
e by enhanced satellites (with increased spatial resolution 
and mid-infrared dynamic range), 
* by increasing the number of satellites used, 
* by speeding up the feeding of satellite data into the 
System, and 
* by sending the alert messages from a fully automatic 
system directly to the local fire authorities. 
Development plans of the system include: 
* integration of an automatic GCP (ground control 
point) based refinement of the geo-coding, 
e elimination of known industrial sites as false alarms, 
* increase of the input sub-scenes, and 
587 
e sending of direct automatic alert-messages (with the 
fire spot indicated on a map) by telefax to regional 
dispatching centres. 
Satellite based fire detection is a feasible tool in the Boreal 
forest zone to augment fire surveillance by other means such 
as air surveillance. This requires that the supply of mid- 
infrared satellite data in day time during the summer season 
is frequent enough. 
6 ACKNOWLEDGEMENTS 
The author would like to thank the Finnish Ministry of the 
Interior for funding the work described in this paper. The 
author would also like to thank Harry Frelander (of the Min- 
istry), Kristiina Soini (of the Finnish Meteorological Insti- 
tute), and Väinô Kelhä (of the VTT Automation), who have 
contributed to the development of the fire detection system 
in the steering group of the project. 
The author would also like to thank the people who have con- 
tributed to the work in the three demonstration phases of the 
system, among them especially Einar-Arne Herland, Seppo 
Väätäinen, Ari Westerlund, Juha-Petri Kärnä, and Mikael 
Holm in VTT Automation, Timo Kuoremäki in Finnish Me- 
teorological Institute, Arto Ruokari and Markus Gronholm in 
Helsinki Regional Dispatching Centre, and Karin Vikstrom, 
Janne Koivukoski, and Tero Paasiluoto in the Finnish Min- 
istry of the Interior. 
The work of numerous individuals who contributed to the ver- 
ification of the fires (in Finland, Estonia, and Latvia) observed 
by the fire detection system is also acknowledged. 
7 REFERENCES 
Flannigan, M. and Vonder Haar, T. 1986. Forest fire mon- 
itoring using NOAA satellite AVHRR, Canadian Journal of 
Forest Research,16, p. 975-982. 
Gonzalez Alonso, F. and Casanova Roque, J. 1994. Applica- 
tion of NOAA-AVHRR images to the study of forest fires in 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996