A Study on the Propagation Trend of Underground Coal Fires Based on Night-Time Thermal Infrared Remote Sensing Technology

Abstract

Underground coal fires in coal fields endanger the mine surface ecological environment, endanger coal resources, threaten mine safety and workers’ health, and cause geological disasters. The study of methods by which to monitor the laws that determine the way underground coal fires spread is helpful in the safe production of coal and the smooth execution of fire extinguishing projects. Based on night-time ASTER thermal infrared images of 2002, 2003, 2005 and 2007 in Huangbaici and Wuhushan mining areas in the Wuda coalfield, an adaptive-edge-threshold algorithm was used to extract time-series for underground coal fire areas. A method of time-series dynamic analysis for geometric centers of underground coal fire areas was proposed to study the propagation law and development trend of underground coal fires. The results indicate that, due to the effective prevention of the external influences of solar irradiance, topographic relief and land cover, the identification accuracy of coal fires via the use of a night-time ASTER thermal infrared image was higher by 7.70%, 13.19% and 14.51% than that of the daytime Landsat thermal infrared image in terms of producer accuracy, user accuracy and overall accuracy, respectively. The propagation direction of the geometric center of the time-series coal fire areas can be used to represent the propagation direction of underground coal fires. There exists a linear regression relationship between the migration distance of the geometric center of coal fire areas and the variable-area of coal fires in adjacent years, with the correlation coefficient reaching 0.705, which indicates that the migration distance of the geometric center of a coal fire area can be used to represent the intensity variation of underground coal fires. This method can be applied to the analysis of the trends of underground coal fires under both natural conditions and human intervention. The experimental results show that the Wuda underground coal fires spread to the southeast and that the area of the coal fires increased by 0.71 km² during the period of 2002–2003. From 2003 to 2005, Wuda’s underground coal fires spread to the northwest under natural conditions, and the area of coal fires decreased by 0.30 km² due to the closure of some small coal mines. From 2005 to 2007, due to increased mining activities, underground coal fires in Wuda spread to the east, south, west and north, and the area of coal fires increased dramatically by 1.76 km².

1. Introduction

Underground and surface coal fire hazards cause considerable harm to natural resources, the environment, human health and infrastructures in which they not only drain an abundance of coal resources but also release toxic gas (CO, SO₂, H₂S, N₂O, etc.) and microelements (AS, F, Se, Hg, etc.) that threaten the physical and mental health of local residents [1]. Additionally, the emission of greenhouse gases such as CO₂ and CH₄ results in global warming and causes surface subsidence, collapse, fractures, venting and landslides, which all disrupt surface ecosystems. The study of underground coal fire detection techniques and the analysis of the trends of underground coal fire development have significant practical value and relevance to the prevention and monitoring of underground coal fire, ecological environment assessment and restoration, and urban planning and construction of coal resources.

Domestic and foreign scholars have used methods based on geomagnetic [2], electric [3], chemical [4], and temperature fields [5] over the years to thoroughly detect the location of underground coal fires. In particular, those methods that utilize satellite thermal infrared (TIR) remote sensing to determine thermal anomalies of underground coal fires has been widely adopted [6,7,8,9,10,11,12]. For example, low-resolution NOAA/AVHRR and MODIS thermal infrared remote sensing images (1 km resolution) [13], medium-resolution Landsat, ASTER, and CBERS-04 satellite thermal infrared images (60–120 m resolution), and high-resolution unmanned aerial vehicle (UAVs) thermal infrared images (<5 m resolution) [14,15,16,17,18,19] have all identified the location of underground coal fires to varying degrees.

Scholars have conducted extensive studies on the dynamics and trend analysis of underground coal fires, relying on their recognized range. Song et al. [20] found an eastward migration trend of coal fires in the Wuda coalfield in Inner Mongolia based on a daytime Landsat time-series thermal anomaly map of coal fires between 2013 and 2014. Using coal fire maps of the Jharia coalfield, India—produced from daytime Landsat thermal infrared imagery between 1988 and 2013—Pandey et al. [21] investigated the lateral variation of surface and subsurface coal fire fields during propagation to new areas. Jiang et al. [22] analyzed the stages of coal fire changes in the Wuda coalfield identified by daytime Landsat satellite imagery from 2000 to 2015, evaluated the efficiency of fire suppression in coalfield fire areas and divided the propagation of coal fires into three stages: expansion, reduction and stable. Mishra et al. [23] used daytime Landsat imagery to classify surface and subsurface coal fires into six types and estimated the general trend of coal fires in the Jharia coalfield based on the size of the coal fire area. However, many studies provide information only on the spatial scale and area variation of underground coal fires, but do not determine the detailed direction of coal fire propagation. Laboratory experimental results [24] have shown that the rate of greenhouse gas emissions from coal fires resulting from mining activities is more significant than surface air seepage. A negative correlation between temperature and subsidence has been reported by monitoring thermal anomalies and land subsidence [25]. Huo et al. [26] manually delineated the propagation direction of coal fire areas in the Wuda coalfield under the natural environment from 1999 to 2006 using the coal fire area baseline method to address this issue. Vu et al. [27] monitored the spatio-temporal changes during 2008–2016 in Khanh Hoa coal field by using Landsat images. Li et al. [28] used ISODATA unsupervised clustering algorithm to find the clustering center of coal fire areas, and dynamically analyzed the direction of coal fire extension caused by open-pit mining in coalfields through the clustering centers of coal fire areas. Du et al. [29] found the fire centers in the Wuda coalfield moved from a deep part to a shallow part along coal seams or along the coal seam outcrops.

In conclusion, most of the current studies have focused on the dynamic changes of the area and spatial location of underground coal fires using daytime thermal infrared images, and few of these have used more accurate night-time thermal infrared images to delineate the extent of underground coal fires, and few studies have been conducted to summarize the development pattern and propagation trend of underground coal fires. Therefore, based on night-time ASTER thermal infrared imagery data from 2002 to 2007, this paper seeks to identify the time-series spatial extent of underground coal fires through surface temperature retrieval and thermal anomalies of underground coal fires in the mining area of Wuda coalfield in Inner Mongolia. It also seeks to summarize and analyze the propagation direction and migration law of underground coal fires by using the time-series dynamic analysis of the geometric centers of coal fire areas, which in turn provides research insights for environmental management, fire suppression engineering, ecological resource protection and sustainable development of the mining area.

2. Study Area and Data Sources

The study area, Wuda coalfield, as shown in Figure 1, is located in the west of Wuda district, Wuhai city, in Inner Mongolia, China. It has the north of the Helan mountain on its west, is 6 km from the Yellow river to the east, has Ulanbuhe desert to its north, and Wuda industrial park to its south, with a specific range of 39°28′41″~39°32′36″ and 106°36′32″~106°39′06″, and it contains two mining areas: Huangbaici and Wuhu mountain. The study area is approximately 22 km² with an average elevation of 1200 m. The climate is temperate continental and has sparse rainfall and much sandstorm throughout the year. The annual average precipitation is around 168.5 mm, mainly concentrated in July and August, which accounts for 59% of the annual precipitation, and the annual evaporation is 3496 mm, which is 21-times the amount of the precipitation. The Wuda coal fires started in 1961, and mainly burned in its natural state before 2004. In 2005, a number of small coal mines in the Inner Mongolia Autonomous Region were shut down due to successive mining accidents, and the underground coal fires subsequently subsided. From 2006 to 2009, the number of small coal mines increased rapidly again, and the surface of the coalfield was severely damaged, with the original landforms widely destructed, and with disorganized stripping pits and coal gangue piles visible throughout the coalfield.

The ASTER satellite images contain three spectral bands: visible light/near infrared, short-wave infrared and thermal infrared, and among these, the thermal infrared band includes five bands between 8.125 to 11.65 μm with 12-bit high radiometric quantitative level and 90-m spatial resolution. The ASTER satellite sensor has a high detectable temperature limit and can measure small radiation differences caused by low-temperature phenomena such as underground coal fires [30], and the accuracy of surface temperature retrieval using the temperature emissivity separation (TES) algorithm is as high as ±1.5 K. Given the accessibility of ASTER data, night-time thermal infrared images of the Wuda coalfield were collected on 21 September 2002, 15 September 2003, 21 March 2005, and 21 October 2007 were used as the data source for this study, and daytime Landsat 5 TM thermal infrared images with spatial resolution of 120 m on 18 September 2007 were used as the comparison data with the night-time ASTER images. ASTER imagery was downloaded from Land Processes Distributed Active Archive Center (LP DAAC) (https://search.earthdata.nasa.gov/search, accessed on 17 January 2021) and the Landsat 5 image was provided by United States Geological Survey (USGS) (https://earthexplorer.usgs.gov/, accessed on 27 January 2021).

3. Methodology

3.1. Thermal Infrared Images Surface Temperature Inversion Method

First, the radiant brightness value R_b of ASTER images in the B10 to B14 band was calculated based on Gain and Offset. Then, according to the radiant brightness value, the TES algorithm can be used to retrieve the surface temperature in the study area. Then, the surface temperature of the study area is retrieved based on radiant brightness values using the TES algorithm, which consists of three main modules, the normalized emissivity method (NEM), ratio algorithm (RAT) and maximum minimum difference (MMD) [31]. These can remove the reflected down-welling sky irradiance, generate accurate unbiased emissivity estimated value and improve the accuracy of the estimated value of the surface temperature. The NEM module is calculated as follows:

$$R_b=L^{\prime }-(1-{\epsilon }_{\max })S_{\downarrow }$$

(1)

$$T_b=\frac{c_2}{{\lambda }_b}{(\ln (\frac{c_1{\epsilon }_{\max }}{\pi R_b{\lambda }_b^5}+1))}^{-1}$$

(2)

$$T^{\prime }=\max (T_b)$$

(3)

$${\epsilon }_b^{\prime }=\frac{R_b}{B_b(T_b)}$$

(4)

where ε_max is initial maximum emissivity at 0.99; L′ denotes the value of the surface radiation that includes the down-welling sky irradiance by eliminating the path radiation; S_↓ represents the down-welling sky irradiance; λ_b is the wavelength of band b; c₁ and c₂ are Planck constant; T_b represents the temperature of band b; T^′ is the output temperature of the NEM model; B_b(T_b) denotes radiation value of black bodies.

The RATIO module calculates the relative emissivity β_b with the following equation:

$${\beta }_b=5{\epsilon }_b/{\displaystyle \sum {\epsilon }_b}$$

(5)

where b = 10, 14; 0.7 < ε_b < 1.0; 0.75 < β_b < 1.32.

The formula used to calculate the estimated emissivity and temperature for the MMD module is as follows:

$${\Delta }_{MMD}=\max ({\beta }_b)-\min ({\beta }_b)$$

(6)

$${\epsilon }_{\min }=0.994-0.687{\Delta }_{MMD}{}^{0.737}$$

(7)

$${\epsilon }_b={\beta }_b{\epsilon }_{\min }/\min ({\beta }_b)$$

(8)

$$T=\frac{c_2}{{\lambda }_{\max }}{(\ln (\frac{c_1{\epsilon }_{\max }}{\pi R_{\max }{\lambda }_{\max }^5}+1))}^{-1}$$

(9)

where b = 10~14; and T represents the surface temperature of inversion. When △_MMD < 0.032, the accuracy of the gray body will be very low, allowing ε_min to equal 0.983.

Furthermore, the algorithm for the Landsat 5 TM thermal infrared image retrieval of surface temperature is undertaken by utilizing the mono-window algorithm from Qin et al. [32]. This method mainly includes calculation of land surface emissivity, calculation of atmospheric water vapor content, and calculation of satellite-observed brightness temperature, etc.

3.2. AET Algorithm to Identify Underground Coal Fires

In previous studies, several thermal anomaly extracting methods have been proposed and used successfully, including the fixed threshold method [33], composite pixel method [34], contextual or moving window (MW) method [35,36,37], and hot spot analysis method [38]. In this study, the adaptive-edge-threshold (AET) method [16], simplified from self-adaptive gradient-based thresholding method (SAGBT) [39], was adopted to recognize the underground coal fire regions from which it calculates the gradient values of the ASTER thermal infrared surface temperature values retrieved in Section 3.1 by the Sobel edge detection operator [39]. The formula of the Sober operator is shown in (10) and (11). Li et al. [28] compared the ranges of subsurface coal fires extracted from ASTER and the China and Brazil Earth Resource Satellite (CBERS) thermal infrared images using previous Roy, SAGBT, and MW algorithms, respectively, and concluded that the AET algorithm had the highest accuracy in monitoring coal fire dynamics over long-time spans.

$$G_x=\left[\begin{array}{l}-101\\ -202\\ -101\end{array}\right]\ast I(i,j)\&G_y=\left[\begin{array}{ccc}-1& -2& -1\\ 0& 0& 0\\ 1& 2& 1\end{array}\right]\ast I(i,j)$$

(10)

$$G=\sqrt{G_x^2+G_y^2}$$

(11)

where G, G_x and G_y represent the overall gradient, horizontal gradient, and vertical gradient of the surface temperature, respectively; I (i,j) denotes the pixel temperature value at (i,j) in the temperature pattern.

The sum of the mean value of surface temperature and the standard deviation of temperature, i.e., T_m + std(T), can be applied to distinguish the high-temperature regions and low-temperature regions. Therefore, the edge of the high-temperature thermal anomaly of the coal fire at the surface of the mine is located in the intersection part G ∩ T which is from the intersection of the temperature region above T_m + std(T) and the gradient region of the Sobel operator. The average value of the temperature of the intersection part Mean(T) is taken as the threshold value T_t of the thermal anomaly of the coal fire, which is expressed by Equation (12) as:

T_t = Mean [T_{[T>Tm + std(T)]} (G ∩ T)]

(12)

The threshold T_t binarization of the coal fire thermal anomaly is used to segment the ASTER surface temperature image to obtain the underground coal fire area image. The raster coal fire area is then regularly gridded, and the outermost closed contour is extracted as the vector polygon of the coal fire area by application of the contour method.

3.3. Underground Coal Fire Propagation Tendency Analysis Method

Because of the numerous coal fire areas and complex graphics, it is difficult to find the development direction of coal fires with the traditional overlapping method of using a time-series of coal fire maps. The coal fire baseline method is also difficult to conduct due to complex manual interpretation. Therefore, this study proposed a method of time-sequential dynamic analysis of the geometric center of coal fire areas to analyze the development trend of underground coal fire, and this method can greatly improve the efficiency of the analysis of the propagation trend of underground coal fires. Based on the extracted underground coal fire areas by the method in Section 3.2, the x and y coordinates of the vector polygons composing each coal fire areas are selected, and the average of the x and y coordinates is calculated as the geometric center of the coal fire areas C(X_i, Y_i), and the underground coal fire areas and their geometric center are superimposed in different periods according to the time series. For the same coal fire area, using traversal lines to directly connect the geometric center of the coal fire area at the same spatial location, the migration traverse lines of the coal fire area over time are able to be formed, i.e., through the trend lines of coal fire propagation development, and according to the propagation trend lines of coal fire, we can further analyze and predict the future development tendency. The geometric center of underground coal fire areas is calculated as follow:

$$C(X_i,Y_i)=C(\frac{1}{n}){\displaystyle \sum _{i=1}^n{x}_i},\frac{1}{n}{\displaystyle \sum _{i=1}^n{y}_i}$$

(13)

4. Results and Analysis

4.1. Accuracy Analysis of Night-Time Thermal Infrared Remote Sensing Coal Fires Identification

The temperature in the coal fire thermal anomaly region of the mine attenuates rapidly with increasing outward distance, with this attenuation producing a considerable number of spots with very high gradient values. From the thermal cracks observed in the field investigations, the high temperatures generally do not extend beyond 2–3 m; moreover, the temperature variation in the coal fire area over a distance of less than 20 m typically exceeds 500 °C, and the resulting temperature gradient is significant [40]. We have employed the Canny, Zerocross, Sobel, Roberts, Prewitt, and Log edge detection operators to detect thermal anomaly gradients. The Canny, Zerocross, and Log operators generated a large number of excess gradients including high and low temperatures, while the Prewitt and Roberts operators missed some high-temperature gradients. Only the Sobel operator identified most of the high-temperature gradients. In addition, during the development of the AET algorithm, we tested T_m + 0.5std(T)~T_m + 1.5std(T) to determine the best demarcation line between the high-temperature and low-temperature areas. After comparative analysis it is known that the high-temperature areas greater than T_m + 1.0std(T) yielded the closest range to the coal fire areas. Consequently, we consider the average temperature of the overlapping part between high-temperature areas and the Sobel operator as the best threshold for the identification of the thermal anomaly of the coal fires.

Figure 2 demonstrates the coal fire areas identified by night-time ASTER on 21 October 2007 and by daytime Landsat thermal infrared images on 18 September 2007. Comparing the historical images of Google Earth, we found that the underground coal fire areas which were automatically identified through thermal infrared remote sensing images are consistently mingled with some false spots related to the high temperature thermal anomaly. These false reports included, for instance, the No.23 fire area in Figure 2a, fire areas 22, 25 and 32 in Figure 2b, which belong to the coal pile areas of the mine, and were manually rejected according to the visual interpretation results.

Table 1. Accuracy comparison of coal fire identification from night-time ASTER images and daytime Landsat TIR images.

Classification	ASTER Night-Time Images			Landsat Daytime Images
	Coal Fires	Coal Pile Fires	Total	Coal Fires	Coal Pile Fires	Total
Coal fires	11	2	13	10	4	14
Coal pile fires	2	0	2	3	0	3
Total	13	2	15	13	4	17
Accuracy	Producer accuracy: 84.62% User accuracy: 84.62% Overall accuracy: 73.33%			Producer accuracy: 76.92% User accuracy: 71.43% Overall accuracy: 58.82%

shows the comparison results of the accuracy in identifying coal fire areas between night-time ASTER imagery acquired on 21 October 2007 and the daytime Landsat images acquired on 18 September 2007. Combined with Figure 2, it can be seen that two omission errors in ASTER images were found in the east and west slag discharge fields of the northern No. 21 fire area, and two commission errors were found in the slag discharge fields of No. 3 and No. 22; Three omission errors were located in the extreme north of three slag discharge fields, and four commission errors were located at No. 3, No. 7, No. 8, No. 27 slagging discharge fields.

We have found that the omission errors and commission errors of the daytime Landsat thermal infrared images were obviously higher than those of the night-time ASTER thermal infrared images, and the producer accuracy, user accuracy, and overall accuracy of Landsat imagery for identifying fire areas were 7.70%, 13.19%, and 14.51% lower than those of night-time ASTER images, respectively. Furthermore, the total area of reference fire areas, fire areas identified by night-time ASTER images and fire areas identified by daytime Landsat images were 3.62 km², 2.80 km² and 1.65 km², respectively, and the total area of coal fires identified by daytime Landsat images was 42.71% lower than the total area of night-time ASTER-identified fires.

Due to the increase of the total radiation on the surface of the mine area caused by solar irradiation during the daytime, the surface of the mine area is heated, and the temperature of some coal piles or gangue, which can easily absorb the heat of solar irradiance, will increase. Furthermore, the mine surface is often undulating, with different slopes and aspects receiving different solar irradiance. Normally, the surface temperature of the steep slope to the north is lower while the surface temperature of the gentle slope to the south is higher. Areas where coal fires did not originally occur may display high temperature thermal anomalies in daytime thermal infrared images, while the high-temperature thermal anomalies in shadows may be covered, and hence, daytime thermal infrared images will generate more coal fire omission and commission errors. In addition, the increasing surface temperature in the mine areas during the daytime leads constantly to the increase of radiation of daytime thermal infrared images and the rise of the retrieved surface temperature values. Therefore, the temperature value along the edge of the high-temperature thermal anomalies identified by the AET algorithm become larger, which results in the increase of the segmentation threshold of coal fires. To sum up, the daytime thermal infrared images will not only lead to the appearance of extra omission errors, but also cause the identified coal fire areas to be considerably smaller than those of the reference coal fire areas. Compared to the Landsat 5 thermal infrared image with spatial resolution of 120 m, the ASTER thermal infrared image features a spatial resolution of 90 m and a higher temperature sensitivity of the thermal infrared sensor. Therefore, night-time ASTER thermal infrared imagery is an ideal data source for subsurface coal fire identification in mining areas because it is not affected by temperature differences at the mine surface caused by solar irradiance, topographic relief and land cover [41].

4.2. Analysis of the Tendency of Underground Coal Fire Propagation

In order to conveniently analyze the temporal evolution of coal fires directly, the directions of coal fire propagation are formed by connecting geometrical centers of coal fire regions in adjacent years using directional lines. As shown in Figure 3, Figure 3a–c represent the evolution process of coal fires in 2002 to 2003, 2003 to 2005, and 2005 to 2007, respectively.

As shown in Figure 3a and

Table 2. Coal fire areas of Wuda coalfield from 2002 to 2007.

Coal Fire No.	Area in 2002/m2	Area in 2003/m2	Area in 2005/m2	Area in 2007/m2
5	8546	0	0	21,404
6	0	17,091	0	245,581
7	141,794	353,708	30,431	246,086
8	0	19,374	510	345,486
9	13,969	12,123	0	305,780
10	41,915	231,483	0	112,651
11	264,797	349,133	247,422	651,408
12	165,875	254,042	440,833	289,977
13	0	0	5425	14,096
14	0	107,767	295,318	261,728
16	0	0	17,873	1181
18	0	0	2070	307,073
Total	636,895	1,344,721	1,039,881	2,802,452

(column “Area in 2003”), during 2002 to 2003, except for No. 7 fire area which developed to the south, No. 5, No. 9, No. 10, No. 11, No. 12, No. 13 fire areas all developed to the southeast, and the remaining No. 6, No. 8, No. 14 fire areas were new emerging fire areas. Except for the decreased areas in No. 5, No. 9 fire areas, the rest of the fire areas showed a significant growth trend, and the total fire areas increased by 707,844 m².

As shown in Figure 3b and Table 2 (column “Area in 2005”), between 2003 and 2005, in addition to the No. 7 fire area toward the northeast and the remaining No. 8, No. 11, No. 12, No. 13, No. 14 fire areas toward the northwest, the No. 5, No. 6, No. 9, No. 10 fire areas disappeared. The No. 13, No. 16, No. 18 fire areas were new emerging fire areas. No. 6 to No. 11 fire areas appeared to show a significant reduction, while the area size of No. 12, No. 13, No. 14, No. 16, No. 18 fire areas increased, and the total fire areas decreased by 304,840 m².

As shown in Figure 3c and Table 2 (column “Area in 2007”), From 2005 to 2007, fire areas No. 7, No. 8, No. 16 and No. 18 spread in the southeast direction, fire areas No. 13 and No. 14 spread in the southwest direction. Fire area No. 11 developed in the middle from two directions, southeast and southwest, and eventually connected into a large fire area. Fire area No. 12 spread in the southeast direction in the northern part of the fire area and in the northwest direction in the southern part, that is, from the north and south ends toward the middle, and finally connected into a large fire area. In addition to the area size of No. 12, No. 14, No. 16 fire areas which had a small reduction, the remaining No. 5 to No. 11, No. 13 and No. 18 fire areas had increased significantly, of which No. 6, No. 8, No. 9, No. 18 fire areas had different numbers of scattered small fire areas, with a sharp increase in the total fire area of 1,762,570 m².

Using a direction line to connect the geometric centers of coal fire areas between 2002 to 2007, a time series coal fire propagation trend was mapped. According to the time-series dynamic shift of geometric center in every single coal fire area, we analyzed the general trends of coal fire propagation.

The propagation direction of the coal fires in the No. 5 fire area is: (Southeast → Northwest), with a general development to the south.

The propagation directions of the coal fires in the No. 6 fire area are southeast and northwest, with a general expansion in two directions to the north and south.

The propagation direction of the coal fires in No. 7 fire area is: (Southeast → Northeast → Southeast), with a general migration to the east.

The propagation directions of the coal fires in No. 8 fire area are: (Southeast → Northeast), (Southeast), with a general rapid expansion to the east and south.

The propagation direction of coal fires in No. 9 fire area is: (Southeast → Northeast and Southeast), with overall sharp expansion in three directions to the north, east and south.

The propagation direction of coal fires in No. 10 fire area is: (Southeast), (Northwest), (Northeast), with overall sharp expansion in three directions to the north, west and south.

The propagation directions of coal fires in No. 11 fire area are: (Southeast → Northwest → Southwest), (Southeast → Northeast), the overall expansion in two directions to the east and south.

The propagation directions of coal fires in No. 12 fire area are: (Southeast → Northwest → Southeast), (Southeast → Northwest), the two overall fire areas are concentrated in the middle area, with a continuous trend.

The propagation direction of the coal fire in No. 13 fire area is: (Southwest); the propagation direction of the coal fires in No. 14 fire area is: (Southeast → Northwest), with a general westward expansion.

The direction of coal fire propagation in No. 16 fire area is: (Southeast); the direction of coal fire propagation in No. 18 fire area is: (Southeast).

Furthermore, by constructing the regression fitting of underground coal fire areas with coal fire propagation distance values in adjacent years, we have discovered that these have a linear correlation relationship with a determination coefficient of 0.705 as shown in Figure 4. This indicates that the propagation distance of coal fires can indirectly indicate the intensity of coal fire propagation.

The results of the coal fire propagation tendency indicate that the coal fire propagation directions were unified in 2002 to 2005, and the distances of coal fires migration were significant. However, during 2005 to 2007, coal fire propagation directions appeared in four directions, east, west, south and north, and the coal fire migration pattern was relatively complex. The area of coal fire has increased in 2002 and 2003, but suddenly decreased in 2005 and raised again in 2007.

The reason for this situation of underground coal fires is that: from 2002 to 2005, underground coal fires in the Wuda coalfield were mainly burned in the natural state, and most of the underground coal fires were distributed along the outcrop lines of coal seams, and the coal fire propagation distance was remarkable. The turning point came in 2005, when the local government shut down a significant number of small coal mines in the Wuhu Shan mining area due to a series of mining accidents, resulting in a sharp reduction in the area of underground coal fires.

After 2006, small coal mines spread without restraint so that the original ground surface was serious damage in mining areas. As a result, the burning pattern of underground coal fires has changed from a natural state to a disturbed state caused by mining activities around the coal field. The area of underground coal fires, the, has not only has grown, but the distribution of coal fires has also shifted from concentrated to scattered distribution. Therefore, from 2005 to 2007, coal fire propagation trends have become more complex and shown in multiple emerged centers and different propagation directions.

5. Conclusions

(1)

Night-time ASTER thermal infrared images can effectively prevent external disturbance caused by solar irradiance, topographic relief and land cover, at the mine sites. The producer accuracy, user accuracy, and overall accuracy of identified coal fires using night-time ASTER thermal infrared images are 7.70%, 13.19%, and 14.51% higher than those using daytime Landsat thermal infrared images, respectively.

(2)

The results of this monitoring by night-time thermal infrared images indicate that underground coal fires in the Wuda coal field spread to the southeast from 2002 to 2003, and the total area of coal fires increased by 0.71 km². From 2003 to 2005, underground coal fires in the Wuda coal field spread to the northwest due to the closure of small mines, reducing the total area by 0.30 km². However, the coal fire area increased dramatically from 2005 to 2007 by 1.76 km² and spread to the east, south, west and north due to increased mining activity.

(3)

The proposed time-series dynamic analysis method of geometric centers of underground coal fire areas can be used not only for underground coal fire propagation trends in adjacent years but can also be superimposed on the time-series coal fire maps for underground coal fire propagation trend analysis. The migration direction of the geometrical center of the coal fires can be used to represent the actual underground coal fire propagation direction, and the migration distance of the geometrical center of the coal fires can be used to indicate the magnitude of the underground coal fire. This approach is applicable to an analysis of the developmental tendencies of underground coal fires both in their natural state and under artificial conditions of exploitation.