Two Dimensional Forest Fire Detection Method by
Using NOAA AVHRR Images
Jun-ichi KUDOH, Kazuma HOSOI
Center for Northeast Asian Studies, Tohoku University,
Sendai, Japan
kudoh@cneas.tohoku.ac.jp
Abstract— We have studied Forest Fire Detection method
with NOAA AVHRR images. The 14 years forest fires points
were accumulate in a three dimensional histogram in Far East
Russia region. The combinations of channels were 1, 3 and 5.
The results shows almost all the channel 3 values were
saturated. So, two dimensional histogram composed on
channel 1 and 5 is obtained, which provides reliability fire
detection corresponding to occurrence number of the
histogram. This makes some looped area in the histogram. The
valuation shows that the area over 5 of the occurrence number
in the histogram detected the fire for another region images
without errors.
Keywords-Forest Fire Detection NOAA AVHRR
I. INTRODUCTION
To detect a forest fire with reliability, we have
studied a high accuracy method with NOAA AVHRR images.
The mainly problems of a fire detection error is based on
simple threshold method using temperature channel images.
The majority method of forest fire detection is composed on
temperature level and temperature difference of thermal
channel, for example, channel 3 and channel 4. The most
important thing of this method is to set these thresholds. In
Siberian area, fired areas were reported from 2 to 4 times over
estimate by several researches [1].
We have already studied a three dimensional
histogram method to classify NOAA AVHRR images
categories [2]. The 14 years forest fires points were
accumulate in this histogram in Far East Russia region. The
combinations of channels were 1, 3 and 5. The total scenes
including fire points are 681, which is every April to
September from 1989 to 2002.
The results shows almost all the channel 3 values
were saturated. So, two dimensional histogram composed on
channel 1 and 5 is obtained, which provides reliability fire
detection corresponding to occurrence number of the
histogram. This makes some looped area in the histogram. The
valuation shows that the area over 5 of the occurrence number
in the histogram detected the fire for another region images
without errors.
II.
FIRE DETECTION METHOD
The most useful method to detect a fire by using
NOAA AVHRR images is threshold method. For example,
Flasse [3] reported the following algorism:
• Ch3 ҆
҆҆
҆ 311K
• Ch3 – Ch4 ҆
҆҆
҆ 8K
This algorism is based on temperature difference of
the object. But it often makes errors for non fire areas as fired.
This is one of the main reasons of the several times difference
for fired area reported by researchers. We confirmed the
detected area by human eyes, and corrected the area Fire or No
fire as the features, smoke, higher temperature around it, low
NDVI values after days, and so on. The results are accumulated
by a three dimensional histogram. The target area is Far East
Russia, where is about square 1100km centered Khabarovsk
city. The fire monitoring period is 14 years (1989-2002). The
correct fired points are stored in a three dimensional histogram
shown in Fig.1.
Fig.1 Accumulated the corrected fired points in a three
dimensional histogram.
0-7803-7930-6/$17.00 (C) 2003 IEEE
0-7803-7929-2/03/$17.00 (C) 2003 IEEE 2494
