Application of unmanned aerial vehicle (UAV) thermal infrared remote sensing to identify coal fires in the Huojitu coal mine in Shenmu city, China

China is a major coal-producing country that consumes large amounts of coal every year. Due to the existence of many small coal kilns using backward mining methods, numerous worked-out areas have been formed. The coal mines were abandoned with no mitigation, so air penetrates into the roadways and contacts the coal seams; as a result, the residual coal seams spontaneously ignite to form coal fires. These coal fires have burned millions of tons of valuable coal resources and caused serious environmental problems. To implement fire suppression more effectively, coal fire detection is a key technology. In this paper, thermal infrared remote sensing from unmanned aerial vehicle combined with a surface survey is used to identify the range of coal fires in the Huojitu coal mine in Shenmu city. The scopes and locations of the fire zones are preliminarily delineated, which provides an accurate basis for the development of fire suppression projects.


Methodology
TIR is a remote sensing method that detects variations in heat on Earth's surface 34 . The use of airborne TIR for mapping and studying coal fires has greater resolution and availability than satellite TIR 11 imagery. Recent advances in UAVs equipped with global positioning systems (GPSs) and digital cameras are reducing the cost of collecting imagery 30 . High-resolution thermal cameras have been successfully mounted on aircraft platforms and on UAVs, increasingly using high-performance sensors with smaller size and weight and greater spectral and spatial resolutions. The thermal cameras can reach centimetre-scale ground resolution and provide sufficient accuracy.
In our research, TIR Zenmuse XT2 cameras mounted onto a UAV DJI M210 were used to acquire data. The Zenmuse XT2 gimbal and cameras, which included a forward looking infrared detector and a visual camera, provided both infrared and visual images simultaneously. The forward looking infrared camera performed Scientific RepoRtS | (2020) 10:13895 | https://doi.org/10.1038/s41598-020-70964-5 www.nature.com/scientificreports/ high-sensitivity infrared scanning at 640/30 fps and was equipped with an uncooled vanadium oxide (VOx) microbolometer to measure longwave radiation in the spectral range 7.5 ~ 13.5 μm and a temperature range of -20 to 135 °C (high gain); it had a 25 mm lens and acquired image frames of 640 × 512 pixels as raw 8-bit digital numbers (DNs) at the rate of less than 9 Hz. The visual camera captured 4 K videos and 12 megapixel photos (https ://suppo rt.pix4d .com/hc/en-us). Several studies showed that TIR surveys conducted during the fall or predawn were best for detecting coal fires [35][36][37][38] , but that RGB orthophotos were best obtained during the day. To acquire both types of data simultaneously, the flight was carried out from 7 a.m. to 10 a.m. on 22 October 2019, which was a cloudy day; the heating caused by sunlight was very small, and solar radiation was negligible. As described previously literature, an appropriate flight plan was determined using the DJI Ground Station software 30 . The flight plan was then uploaded to the quadcopter's flight controller using the DJI Vision App 30 . Accordingly, both in-flight navigation and image capture were autonomous 30 . The internal time of the camera was set to the GPS time prior to the flight to ensure that the images could be easily synchronized with the position data in the UAV GPS log file 28 . Ground control points (GCPs) measured with differential RTK GPS were established before the flight so that the resulting orthophoto imagery and digital elevation models (DEMs) could be accurately georeferenced and tested 30,32 . On a clear day with good visibility, the flight was conducted at a relatively low angle with respect to the horizon. Three flights were conducted with a flight altitude of 300 m, the frontal overlap rate was 75%, and the side overlap was 65% to obtain high-accuracy results. When there is high overlap between 2 images, the common area captured is larger, and the key points can be matched. Therefore, the main rule is to maintain high overlap between the images. The route length was approximately 21.6 km, and 431 overlapping images were processed using the Pix4D software (https ://suppo rt.pix4d .com/hc/en-us). A digital orthophoto image and a digital surface model (DSM) of the mining area were generated, with a ground resolution of 3.8 cm.
A program was written in Python code to calculate the maximum and minimum grey values of the TIR image, which were used to invert the surface temperature, and the thermal anomaly distribution area caused by coal fires and the locations of fire area were accurately determined (Fig. 3), with a ground resolution of 40 cm (pixel size). During the flight, the ground temperature was measured simultaneously to obtain the key parameters for temperature calibration and inversion of the TIR image; the measurement used a TIR thermometer

UAV thermal infrared.
Airborne TIR technology has a wide detection range and high image resolution, which provides great spatial detail for mapping coal fires [40][41][42] , especially based on UAVs. Thermal anomalies induced by underground coal fires can be extracted from TIR data using an exclusion method 43 , together with field temperature measurements. TIR data are widely used to delineate subtle surface thermal anomalies associated with underground coal fires 44 . In our research, the digital orthophoto map (DOM) and the TIR image ( Fig. 3) obtained after processing the raw data collected by the drone. The image was clear, and the colours were bright. The TIR data and the RGB orthophoto relied upon ground control points (GCPs) or orientation measurements from inertial measurement units (IMUs) to enable accurate georeferencing of the imagery. Then the thermal infrared DOM was registered with the RGB DOM, and the registration error was 0.4 pixels. From the DOM, the coal seams, fissures, burnt rocks, gangue, pool, backfill area, coal washing plant, initial landforms, residential areas, etc., could be clearly interpreted based on multi-scale segmentation and were verified with surface surveys (Fig. 4). Thermal anomalies in the fire zone are very obvious and appears as high-brightness spots or bands on the TIR remote sensing image. The difference in brightness reflects the temperature difference of the fire zone. The thermal anomaly packet is extracted and superimposed on the DOM to interpret four banded fire zones and seven small temperature anomalies (Tables 1, 2).The development of vents, cracks, subsidence on the surface result from underground coal fires. Such features sporadically extent in spatial and vary in dimension from a few to tens of metres 45 . Combining the investigation of surface fissures and cracks with or without smoke, and temperature measurements, the fire transition zone is delineated. Other areas are normal areas, which are non-fire zones. From east to west, the study area is divided into four coal fire zones I, II, III and IV; other fires are small and sparse, mostly burning gangue extending 30-100 m. The fire zone has the characteristics of high temperature, heat waves, flames, and new burnt rocks. It is the centre of the coal fire zone with many open fires according to the field survey ( Fig. 4). Slumping land surface, large and wide fissures are evidently visible along the edges of the fire. Combustion leads to subsidence and many cracks (Fig. 5), which are very dangerous and basically make access difficult. In the fire transition zone, chimneys form in the fire zone, flames are basically not seen, fissures and cracks develop, and smoke is emitted from the cracks with high-temperature gas. This transition zone is the margin of the coal fire zone. The coal fire area refers to the cumulative area of fire zones and fire transition zones.
Coal fire zone I is a long northwest-southeast strip along the overhanging wall of the pit, ranging from 45 to 225 m wide and approximately 816 m long. There are four roadway openings, six open fires and four collapses with many fissures and cracks, which emit smoke and gas. The area of the fire zone is 33,521.64 m 2 , the area of the fire transition zone is approximately 42,113.32 m 2 , and the total area of this coal fire is approximately 75,634.96 m 2 .
Fire zone II corresponds to fire zone I and is is also a long northwest-southeast strip along the overhanging wall of the pit, 30     In pace with eliminating the underlying coal, the overburden subsides and the air conducts into the burning area and hot gas escape from there though the tension cracks, which promotes combustion. As time going, coal burns deeper into the mountain slope, leading to the overlying rocks to gradually subside into the burned-out void 10 . Therefore, the direction of fire advance is from the fire zone to the fire transition zone.
TIR can detect the location of coal fire based on surface signatures 46-48 but cannot be seen into the subsurface, so the true range of the subsurface burning region cannot be delineated merely from this technique. It is successful to identify and delineated the surface fires with depths less than 10 m, but hard to identify fires deeper than 30 m, because it is need a long time (approximately a decade) to conduct the heat to the surface 35 . Therefore, remote sensing is predominant in revealing near-surface fires but has difficulties identifying fires at greater depths 26,49 .
Delineation of the subsurface fire is essential to extinguish fire project, including surface subsidence and temperatures, cracks and fissures. Investigation of these variables can approximately identify the areal extent of the fire 25 . Field geological work includes the investigation of fire areas, crack, smoke, laneways and old kiln wellheads, and the measurement of surface temperatures. Ground real information about coal fires in the HCM has been acquired from using portable thermometers. A field survey was conducted during the month of November 2019 to obtain results for validation. Temperature were measured at different heights in the opencast mine to comprehend the connection between the thermal anomalies due to subsurface coal fires and background temperatures (Fig. 6). The high-temperature points and cracks are basically located in the coal fire zone and effectively determine the coal fire range.
Drilling. Based on the preliminary delineation of the coal fire zone, the centre of the fire zone is determined by deploying drill holes. A corresponding bore is placed on each side of the fire transition zone and the non-fire zone. One borehole is drilled inside first; if it is a high-temperature hole, drilling continues outside; otherwise, drilling is stopped. Using a drill, the lithologies are determined by the characteristics of fragments carried by the wind pressure, such as sandstone, burnt rocks and coal (Fig. 7a, b, c). Temperature is measured from top to bottom every 5 m through the borehole (Fig. 7d), which can approximately verify and modify the boundaries of coal fires. Figure 6 shows the drilling locations, and Table 3 shows the drilling characteristics. Red solid circles represent holes drilled into burnt rock with high temperature, red open circles represent holes drilled into   www.nature.com/scientificreports/ temperature controls the boundary of the western bifurcation fire zone, which is approximately 25 m from the surface smoke point (46.0 °C). ZK16 is located in the back-fill area, approximately 10 m from the smoke point (37.0 °C); the maximum temperature in the hole is 41.2 °C, and the eastern boundary of the fire zone can be determined. According to the above information, the TIR anomaly bifurcates on the surface to the east and west and may indicate a single fire under-ground. The ZK17 hole close to the fire zone is only 9.40 m deep and difficult to drill deeper due to the high temperature; at 6.20 m, high-temperature hard burnt rocks appear (210.0 °C). ZK18 is approximately 13 m from a crack without smoke emission and approximately 28.00 m from the crack with smoke (24.4 °C), and the bottom of the hole is 34.0 °C.
Coal fire zone II. The main coal seam with spontaneous combustion in the fire zone is the 1 -2 coal. ZK01 is approximately 14 m from the a surface crack with smoke (53.0 °C), and approximately 31 m from a surface high-temperature point (120.0 °C); the bottom temperature is 18.0 °C, and the borehole is considered a nonfire zone; thus, the range between ZK01 and near-surface cracks can roughly delineate the fire zone boundary, which is consistent with the initial fire zone boundary, so no further holes are implemented. Similarly, ZK03 is approximately 6 m from the surface crack with smoke (68.0 °C) and is a high-temperature hole (240 °C), which is determined to be a fire zone. The corresponding hole ZK04 approximately 25 m from ZK03 is a low-temperature hole (17.6 °C), so ZK03 and ZK04 are judged to mark the borders of the fire zone. In summary, the fire zone of the coal seam slowly burns into the mountains along the steep walls, and the coal seam that is completely covered does not burn underground due to insufficient oxygen supply. The burning rate of coal seams is mainly related to the thickness of the overlying strata and the development of fractures. The fire is extinguished naturally where the overburden layer is so enough intact and thick that fractures fail to reach the surface to supply more air 10 .
Coal fire zone III. The main coal seam with spontaneous combustion in the fire zone is the 2 -2 coal, which is located in the lowest part of the mining area where the wall overhangs pit. In the eastern fire zone III-E, the overhanging wall shows the 1 -2 coal with a thickness 1-2 m and ongoing combustion, and the top covers burnt rocks.
ZK21 is approximately 50 m from a crack with hot gas emission, the highest temperature in the hole is 82.9 °C, and the temperature is abnormally high at 1,042-1,052 m on the roof of the coal seam. According to the field survey, the 2 -2 coal roof is exposed at the bottom of the pit with a few open fires, and it can be inferred that the abnormally high temperature in ZK21 is caused by baking, which results in smouldering. ZK22 is approximately 35 m from the crack with hot gas and smoke in the III-W fire zone and is drilled into worked-out area. Because ZK22 is a high-temperature hole (155.7 °C), ZK23 is implemented at a greater interval. The hole is approximately 50 m from ZK22 and 84 m from a crack at the edge of the fire zone, and it is still drilled to a worked-out area with a temperature of 57.1 °C. ZK22 and ZK33 may be connected via roadways with a cap thickness of approximately 28 m. It can be inferred from the exposed laneway with open fire to ZK23 that the fire extends approximately 100 m along the roadway and that the temperature decreases from 407 °C to approximately 60 °C.
Coal fire zone IV. The main seam with spontaneous combustion in the fire zone is the 1 -2 coal. ZK19 is approximately 35 m from a crack, and the temperature at the bottom of the hole is 50.5 °C. Because the terrain is steep and difficult to reach, ZK20 is approximately 90 m away from the cracks, and the temperature at the bottom of the hole is 15.7 °C, which is a non-fire zone. From the comparison of ZK19 and ZK20, it can be seen that the cantilevered fire zone has a baking heating effect on the coal seams smouldering at a short distance, and the temperatures of the strata far from the overhanging walls tend to be normal.
The above drilling data show that the preliminarily delineated fire area is basically accurate, and only some parts need to be modified.

Discussion and conclusion
The largest coal consumer, China experiences the most coal fires in the world. Therefore, it is important for China to monitor and execute coal fire evaluation, and suitable suppression work 17 . The government of China has got a clear understanding of this hazard and its impacts on the economy and health, with initiatives for fighting coal fires since 1988. Supposing know the depth of the coal fires, coal fire-fighting teams could fight the fires more successfully and efficiently 50 . To this end, systematic quantification and investigation of actual scenarios of coal seams are always critical issues for the coal fire research community 45 . Many surface and underground coal fires in northern Shaanxi, such as those in the Longyan, Tanyaoqu, and Huojitu coal mines, are generally less than 10 km 2 , most of them are 2-3 km 2 , and the coal fire distribution is even smaller. In these coal mines, satellite imagery (> 0.5 m) often provides inadequate detail about fissures and coal fire information, and imaging carried out by traditional airborne platforms (< 0.1 m) can provide high temporal and spatial resolutions but with high costs 28,30 . Satellite and conventional platforms are limited in weather, the availability of aircraft, and satellite orbits 32 . Magnetic surveys offer a method for the detection, delineation and monitoring of coal fires 3 , but in the HCM, ground disturbances and destruction block access to coal fires, so the magnetic method cannot effectively delineate the coal fires. Contrast to traditional airborne remote sensing, UAV remote sensing provides fine spatial and higher temporal resolution and low-price to satisfy the critical requirements of spectral, spatial, and temporal resolutions 32,39 . This technique commits to offer the swift and safe survey of thermal areas, often current in dangerous and inaccessible terrain 31 .
As Greene (1969) described previously, it is easy to detect fires less than 10 m in depth on TIR; fires between 10 and 30 m are detected only when the heat is transported to the surface by cracks or is conducted to the surface for several years or more; greater than 30 m in depth, detecting fires at the surface require a decade, or more 35 . In our research, TIR remote sensing technology is very effective in monitoring the high-temperature and thermally anomalous regions formed by surface and near-surface(< 10 m) coal fires, especially in open flame areas. However, it is very troublesome to identify coal fires more than 10 m deep with intact cover; for example, at the positions of ZK5, ZK6, ZK22, and ZK23, with no thermal abnormalities on the TIR image, the borehole temperatures are abnormally high. Therefore, remote sensing interpretation of RGB orthophoto images, ground investigations and drilling are needed to compensate for the shortcomings of TIR images. Surface subsidence, cracks, fissures, hot gas and smoke are all manifestations of the development of coal fires. They are thermally anomalous areas that expand outward from the open flame area and may mark the locations of the next open flame areas.
Our study demonstrates a low cost and effective technique to detect the main coal fires in northern Shaanxi based on UAV remote sensing and provides an accurate basis for fire suppression projects.

Data availability
The data and analysis generated during the current study are available from the corresponding author on reasonable request.