The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B7. Beijing 2008 
k 
4 
Figure 4-4 Local fire points on August 11 th 
Figure 4-5 is scatter plot of brightness temperature of band 22 
and band 23, the abscissa is band 22; Figure 4-6 is scatter plot 
of the brightness temperature of band 20 and band 21, and the 
abscissa is band 20. We can clearly see abnormal points from 
the plot. Abnormal points are far away from points group. 
These abnormal points are most likely due to the saturation of 
bands 20, 22. The uncertainty data is in band 22 which can be 
got from HDF document. Many researchers use band 21 only 
because of its unsaturation, and abnormal data in band 20 and 
band 22 was often discarded. Our experiment shows that it is an 
efficient way to use these "abnormal" data combining with band 
7 to detect high temperature fires. 
Figure 4-5 Scatter plot of bands 22, 23 
Figure 4-6 Scatter plot of bands 20, 21 
4.2.2 Non-high Temperature Fires Judged with 
Comprehensive Threshold Criterion 
Non-high temperature fires include three major types: i) before 
the fire occurred with the temperature lower than ignition 
temperature; ii) after fire occurred, the temperature had fallen, 
but higher than normal temperature; iii) The fire point is at high 
temperature, but the firing area is smaller than that of a pixel. 
There are many difficulties and uncertainties in non-high 
temperature fire monitoring, particularly the third category of 
non-high temperature fires. Therefore, many researchers define 
the fire point that burning area is larger or equal to one pixel 
size. With non-high temperature fires complex situations 
considered, a comprehensive threshold criterion is used in our 
work. We combine visible, near-infrared, mid-infrared and 
thermal infrared bands together to get an integrated application. 
First, visible and near infrared bands (MODIS bands 1, 2) are 
used to generate NDVI. NDVI has two roles: i) The fire area 
where NDVI is less than 0.3 generally reveals combustibles; ii) 
Generally, there are unbumed vegetations with higher NDVI 
around the fire place. The above two are used together to judge 
forest fire points. At the same time, according to the 
characteristics of the forest fire, we use the differences of 
brightness temperatures between band 21 and band 31 to judge 
fires. The comprehensive threshold is: 
NDVI < 0.3 
BT 2X -BT M > 15 ( 4_1 ) 
' BT 2X >315 
BT 2X > aver(BT 3X ) 
Where BT 2 i is the brightness temperature of band 21, BT 3 i is 
the brightness temperature of band 31, aver (BT 3t ) is the 
average non-cloudy pixel’s brightness temperatures of band 31. 
Suspected fire points are generated with the comprehensive 
threshold, a second judgment is followed which purpose is to 
identify whether the background pixels are forest. It can be 
achieved with NDVI. Figure 4-7 is the result of using 
comprehensive threshold on August 11 th , 2006. 
Figure 4-7 The suspected fires points generated by threshold 
criterion on NDVI of August 11 th , 2006 (red areas were 
suspected fires, high temperature points in low NDVI were 
suspected as urban heat islands) 
In Figure 4-7, the further judgment is based on background 
pixels. If we can integrate land-use data, the accuracy will be 
improved. We can exclude most places with high temperature 
result from urban heat island. Other high temperature points 
may be fire points or ground facilities (further judgments with 
ground truth are needed). 
In order to make a further test, we extracted and analysed the 
data on August 8 th and August 30 th , 2006. Figure 4-8, figure 4-9, 
figure 4-10 and figure 4-11 were the results. 
425