S 
the 
to- 
ne 
ult 
for 
nd 
The Chinese problem is vast; in fact, the amount of coal lost to 
fires there can only be guessed at (The loss was estimated to be 
between 100 and 200 million tonnes in 1992 (Rozema et al., 
1995).) Particular problems in the case of the Chinese coal fires 
are the vast area over which the fires are spread and the 
remoteness of many of the areas in which they occur. This 
situation would seem to present a good opportunity for the use of 
remote sensing data for fire detection and monitoring. Obviously 
one would want to use data with as high as resolution as possible - 
probably airborne data - for this, but it would be impractical to 
gather such data for the whole of the fire-prone areas. For this 
reason, a hierarchical detection scheme is being tested, whereby 
thermal anomalies are first looked for in low resolution data and 
subsequently in data sets of higher and higher resolution. At each 
stage, the area investigated is smaller. 
Images from the thermal bands of the NOAA-AVHRR, ERS-1- 
ATSR and Landsat-TM sensors have been acquired together with 
thermal data gathered by an airborne multispectral scanner 
belonging to the ARSC. It is also hoped to acquire some 
RESURS-01 data later. 
1.2 Detection of Thermal Anomalies 
In order to test the detection methods, three small test areas, 
known to contain active fires were chosen for detailed study. 
These test areas lay in north-western China, close to the city of 
Urumgi, the capital of Xinjiang Autonomous Region. 
From a first look at the data available, it was obvious that the 
most clearly visible thermal anomalies were to be found in the 
airborne thermal scanner data. These data were therefore 
considered first, in contrast to the pattern of the intended final 
detection scheme. The first question to decide upon was how to 
define a temperature anomaly. Looking at the airborne data, there 
were clearly defined areas which would instinctively be 
considered as anomalously warm (see figure 2). It was decided to 
use this as a working definition; in fact, it was found that the 
pixels defined as being anomalous in this way closely matched 
the hottest 0.1% of pixels for all the airborne scenes. 
The next problem was how to determine which of the thermal 
anomalies really were caused by coal fires. From fieldwork 
carried out in 1994, it was known that there were four possible 
causes of temperature anomalies within the study areas. These 
were: coal fires, solar heating of the ground, abnormal geothermal 
fluxes and human activities. The effect of abnormal geothermal 
fluxes was known to be very small and could therefore be 
neglected. Heat sources such as power stations and steel works 
have well-known positions. That left solar heating of the ground 
as the most likely source of temperature anomalies that could be 
confused with coal fires. 
The positions of the thermal anomalies detected using the 
airborne thermal data were compared with digitised geological 
maps of the test areas. All thermal anomalies that lay within the 
coal-bearing rock strata were considered to be caused by coal 
fires. However, there were far fewer clearly visible thermal 
anomalies in the day-time images than in the night-time images - 
compare figure 3 with figure 2. The distribution of thermal 
anomalies in the night-time data corresponded closely to the 
positions of fires known from fieldwork. It was considered that it 
was the (uneven) solar heating of the test area that made the day- 
time data less suitable for detecting the coal fire thermal 
anomalies. 
The day-time TM imagery was also affected by solar heating. 
Figures 4 and 5 are day- and night-time band 6 images, 
respectively, of the Kelazha test area south-west of Urumqi. The 
723 
night-time image was acquired on 7th April 1995 and the day- 
time image on 14th September 1994. Two anomalously warm 
areas can be seen in the day-time image: the long white belt near 
the top and the curved feature near the centre. The first anomaly 
could not be detected in the night-time image and its appearance 
in the day-time scene was assumed to be due to solar heating. 
From its position and form, it was reasonable to think that the 
other anomaly corresponded to the fire already detected in the 
airborne imagery. Because of the lower spatial resolution of the 
TM data, there is less detail visible in these images than in the 
corresponding airborne images. 
An attempt was made to remove some of the solar heating effects 
by producing a relative solar illumination map of the Kelazha 
area. This was done with the aid of a slope map, derived from a 
digital elevation model, and calculations of the solar position at 
the time of image acquisition. The solar illumination map (values 
0-255) was subtracted from the original TM image. Figure 6 
shows the result of adjusting the 7th April 1995 image in this 
way. It can be seen that most of the thermal anomalies in the top 
image are suppressed but that the others remain. The distribution 
of thermal anomalies now corresponds more closely to that of the 
night-time TM and night-time airborne images. 
As yet, no coal fires have been detected using the AVHRR data 
and we are now fairly sure that none will be. This is not 
surprising given the low spatial resolution of these data. Work on 
the ATSR images is in progress. The RESURS-01 data still have 
to be acquired. 
2. THERMAL MODELLING OF COAL FIRES 
2.1 Soil Temperature Determination 
The methods described so far can successfully locate many coal 
fires approximately. In addition, successful fire-fighting needs as 
much information as possible regarding the depth, temperature 
and extent of the fires. 
Several of the fire detection studies mentioned earlier in this paper 
included descriptions of attempts to determine either the depth or 
temperature of an underground fire. One of these quantities was 
known, or estimated; the other was then calculated using some 
form of heat conduction equation. With the exception of a study 
of coal fires in the Jharia coalfield, India (Prakash et al., 1995), 
no-one has successfully determined both the depth and 
temperature of an underground fire from surface information 
alone, and even in the Jharia study an estimate of the fire’s age 
was needed. The second major part of this research concerns the 
development of thermal models that will allow the depth, 
temperature and size of underground fires to be determined using 
only information obtainable at the surface. 
It is well known that the temperature of an outdoor surface 
depends on many factors: the time of day and of the year, the 
weather conditions, slope, aspect, type of surface, etc. (Sabins, 
1987). This means that the temperature of the ground above the 
coal fires is not affected by the fires alone, as was clearly 
demonstrated by the difficulty in detecting the fires using day- 
time images. For this reason, it was thought that it would be 
easier to begin the thermal modelling using data that were free of 
these surface effects. 
Close to the surface, soil temperatures follow an approximately 
sinusoidal temperature curve on both a diurnal and an annual 
cycle (Jury et al, 1991). The amplitude of these temperature 
waves decreases with increasing depth until at a certain depth, the 
variations cease. It is observed that the annual temperature 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996