Spatio-temporal variation and propagation direction of coal fire in Jharia Coalfield,India by satellite-based multi-temporal night-time land surface temperature imaging

Nrendr Singh,R.S.Chtterjee,Dheerj Kumr,D.C.Pnigrhi

a Department of Mining Engineering,Indian Institute of Technology (Indian School of Mines),Dhanbad 826004,India

b Geosciences Department,Indian Institute of Remote Sensing,Dehradun 248001,India

Keywords:Satellite-based night-time imaging Land surface temperature Coal fire Spatio-temporal variation Propagation direction Jharia Coalfield

ABSTRACT In this paper,the spatio-temporal variation and propagation direction of coal fire were studied in the Jharia Coalfield (JCF),India during 2006–2015 through satellite-based night-time land surface temperature (LST) imaging.The LST was retrieved from Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) night-time thermal-infrared data by a robust split-window algorithm based on scene-specific regression coefficients,band-specific hybrid emissivity,and night-time atmospheric transmittance.The LST-profile-based coal fire detection algorithm was formulated through statistical analysis of the LST values along multiple transects across diverse coal fire locations in the JCF in order to compute date-specific threshold temperatures for separating thermally-anomalous and background pixels.This algorithm efficiently separates surface fire,subsurface fire,and thermally-anomalous transitional pixels.During the observation period,it was noticed that the coal fire area increased significantly,which resulted from new coal fire at many places owing to extensive opencast-mining operations.It was observed that the fire propagation occurred primarily along the dip direction of the coal seams.At places,lateral-propagation of limited spatial extent was also observed along the strike direction possibly due to spatial continuity of the coal seams along strike.Moreover,the opencast-mining activities carried out during 2009–2015 and the structurally weak planes facilitated the fire propagation.

1.Introduction

Coal fires are a catastrophe in the coal mining area leading to continuous burning of the high-grade coal deposits across the world,especially India[1,2],China[3,4],the U.S.[5],and Australia[6].Besides,coal fires are a risk to the environment,and cause health hazards [7].Coal fires are primarily caused by the selfignition of coal.Self-ignition is triggered when an exothermic reaction between coal and oxygen takes place and the heat released by the reaction is not dissipated [8,9].The Jharia Coalfield (JCF) in India has the highest vulnerability for coal fires [10].Due to coal fires,the JCF has been experiencing challenges in the mining of good quality coal deposits apart from health hazards for a long time.The spatial-temporal analysis of coal fire in the JCF has become necessary for deciding suitable preventive measures to reduce the adverse consequences of the coal fire.Due to the dynamic nature and irregular distribution of coal fire,traditional ground-based techniques are neither economical nor advisable to monitor sporadically occurring coal fire over a wide area.

The space-based remote sensing technology has the potential to the monitoring of coal fire over a large area at a regular time interval and is affordable as well.The studies conducted in the past have reported to delineate coal fire in the JCF using space-based coarser as well as medium spatial resolution thermal infrared(TIR) data [1,11–16].However,coarser spatial resolution TIR data of space-borne sensors such as Moderate Resolution Imaging Spectroradiometer (MODIS)is not recommended for quantitative analysis of coal fires[13].The medium resolution TIR data of Landsat-5,7,8 and Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) have been explored to overcome constraints of coarser-resolution TIR data for quantitative analysis of coal fire in the JCF[17–21].Further,the nigh-time TIR data of medium spatial resolution is more appropriate for precise detection and temporal monitoring of coal fire as compare to daytime because the temperature contrast between coal fire and background features is maximum at night due to the absence of solar flux [1,22,23].Apart from night-time TIR data,retrieval of atmospheric and emissivity corrected land surface temperature (LST) also precisely enhances the temperature contrast between coal fire and background pixels [24].By considering both factors (night-time TIR data and LST retrieval) as essential components for improvement in the detection of coal fire,Singh et al.[25]implemented a robust split window algorithm(SWA) using ASTER night-time multispectral TIR data to retrieve precise night-time LST and manifested that retrieved LST enriches the coal fire detection.

Most researchers studied the spatial extent variations of coal fire based on multi-temporal daytime LST retrieved by simple Planck’s thermal inversion.Raju et al.[19] used daytime ASTER LST images for trend analysis of spatial extent of the coal fire in the JCF.The study revealed that during 2000–2004 the coal fire area shows a declining trend followed by an increasing trend during 2004–2009.Recently,Mishra et al.[21] made use of daytime Landsat LST images for mapping the surface and subsurface fires in the JCF and evaluating its six stages in terms of area at successive five-year intervals during 2001–2016.However,the study of spatial extent variations of coal fire based on precisely estimated multi-temporal ASTER night-time LST by robust SWA with due consideration to emissivity and atmospheric correction is nearly untouched,which is scientifically more accurate than earlier studies with multi-temporal daytime LST by simple Planck’s thermal inversion.Besides,previous research works also lack significant analysis of propagation directions of LST-based coal fire.

The main focus of the research is to assess spatio-temporal variation and propagation direction of the coal fire in the JCF during 2006–2015 using precisely retrieved satellite-based multitemporal night-time LST.This research article adopted the following key steps:(1) retrieval of precise LST from ASTER night-time TIR data of 2006,2009,and 2015 by robust and refined SWA using night-time atmospheric transmittance and land surface emissivity(LSE),(2) accurate detection of thermal anomalies related to surface and subsurface coal fire by LST-profile-based statistical approach,(3) assessment of the spatio-temporal variations of surface and subsurface coal fire in the JCF from 2006 to 2015,(4)evaluation of different coal fire states such as new,consistent and extinguished fire during the observation period,and (5) analysis of propagation direction of the coal fire in the light of geological features and mining activities.The results of the present study are verified with ground-based observations.The satellite-based estimates and techniques adopted in this study are scientifically more appealing and precise.Further,the inferences in consultation with essential factors and ground-based observations have become more significant.Thus,it is assumed that this study shall be of interest for quality research in satellite-based coal fire studies.The results may provide decisive ideas to decision makers for identifying future potential risk zones,controlling fire,and hazard-free sustainable coal mining.

2.Study area and datasets

The JCF is situated in the Jharkhand state of India.This region stretches from latitude 23.58°N to 23.91°N and from longitude 86.08°E to 86.55°E within an area of approximately 450 km2(Fig.1).The JCF is severely impacted by coal fires.Sixty-seven fires have been reported in the JCF out of which fifty-eight fires are in the active phase which require preventive measures to control them and nine underground fires are in the dormant phase [26].All fires are restricted to Barakar formation in the JCF as all coal seams are confined to this formation [19].This formation is also severely influenced by faults and fractures.The majority of faults are east-west (E-W) or northwest-southeast (NW-SE) trending and in the vicinity of opencast mines [27].

ASTER night-time TIR Level-1T products (AST_L1T) have been acquired at approximately 16:32:39 GMT on 25 October 2006,27 November 2009,and 11 October 2015 over the area under study.Commonly,day-time ASTER visible near and short wave IR data(VNIR and SWIR) of the overlapped or adjacent dates as AST_L1T data were utilized for generating normalised difference vegetation index(NDVI)and normalized difference water index(NDWI)maps.In absence of cloud-free ASTER VNIR and SWIR data,the study used Landsat-5 or Landsat-8 data for generating NDVI and NDWI maps.The JPL (Jet Propulsion Laboratory) emissivity spectral library was used for calculating emissivity values of soil,vegetation,and water for TIR bands 13 and 14.

In-situ coal fire data was obtained from field-survey based report made by BCCL (Bharat Coking Coal Limited) of CIL (Coal India Limited) and field survey.The extensive field work was conducted at various locations in the JCF during October and November of the year 2015 with Raytek PhotoTemp MX6 Photographic TIR gun to measure fire temperature.The global navigation satellite system (GNSS) receivers operated in differential global positioning system mode were used to locate coal fire sites.The TIR gun operates at a spectral range of 8–14 µm and a temperature range of -30 to 900 ℃.The accuracy of the instrument is ±0.75℃ (https://www.instrumart.com/assets/raymx6.pdf).The tested horizontal accuracy of GNSS in the study area is approximately±5 to 10 cm after differential correction.

3.Methodology

3.1.Retrieval of land surface temperature

The present study adopted SWA based procedure as proposed by Singh et al.[25]for LST retrieval,especially for the coal fire area.The simplified SWA equation for LST is

whereTSis the LST;T13andT14the brightness temperatures of TIR bands 13 and 14,respectively;andA0,A1,andA2the coefficients determined by the atmospheric transmittance and LSE in both TIR bands.

Initially,AST_L1T products were converted to raw spectral radiance,and then recalibrated radiances were computed using recalibration coefficients obtained from an online RCC (radiometric calibration coefficient) model proposed by Tonooka et al.[28](http://tonolab.cis.ibaraki.ac.jp/ASTER/RECAL/).The brightness temperature of TIR bands 13 and 14 (T13andT14) was estimated using recalibrated spectral radiance of respective bands by inverse Planck’s radiance formula.The scene-specific regression coefficients,band-specific land surface emissivity,and atmospheric transmittance are needed to deriveA0,A1,andA2parameters which are involved in Eq.(1).The regression coefficients were simulated from linear approximation between parametric spectral radiance(Li) and scene-specificT13andT14.Liis a function of temperature and empirically computed from the ratio of radiance to Planck’s radiance derivative with respect to temperature.For detailed discussion about these parameters refer to Singh et al.[25].

Fig.1.LST image of the JCF retrieved from ASTER night-time TIR data of 2015 by SWA showing a set of eight transects across selected eight coal fire test sites.

The hybrid LSE model [25] was adopted to estimate the LSE of ASTER TIR bands.This model includes normalized difference vegetation index threshold(NDVITHL),thermal emission and vegetation cover (TVC) model,modified normalized difference water index(MNDWI),and red band reflectivity model.The land surface cover pixels were categorized byNDVIvalues as bare land pixels(NDVI<0.2),mixed pixels(0.2 ≤NDVI≤0.5),and vegetation pixels(NDVI>0.5).For mixed pixel (mixture of soil and vegetation) LSE,the TVC model was implemented using vegetation fraction cover,and radiance ratio of soil (rs) and vegetation (rv).The TVC also used JPL average spectral emissivity values of soil (TIR13=0.9682,TIR14=0.970) and vegetation (TIR13=0.9745,TIR14=0.9725) types present in the study area.Thersandrvwere approximated from linear regression between numerically computedrsandrv,and vegetation fraction cover.Thersandrvvalues used to establish linear regression were calculated by dividing the thermal radiance of soil pixels and vegetation pixels by the thermal radiance of mixed pixels,respectively [25].The bare land pixels (NDVI<0.2) comprise soil,sand,urban area,overburden dump,and opencast mines except for water body pixels.For such pixels,the study used red band reflectivity based emissivity of Landsat-8 TIRS(thermal infrared sensor) 10 [29] for ASTER TIR band 13 and bandwidthweighted average red band reflectivity based emissivity of Landsat-8 TIRS bands 10 and 11 for ASTER TIR bands 14 [25].In the case of water body pixels,theMNDWI>0.123 was used to extract the water pixels [30] and it was assigned average emissivity of water for TIR bands 13(0.9901)and 14(0.9904)from the JPL emissivity spectral library.Similarly,in the case of dense vegetation pixels (NDVI>0.5),the average emissivity values of dense deciduous vegetation for TIR bands 13 and 14 computed from the JPL emissivity spectral library were used.

To remove the atmospheric effect,atmospheric transmittance of TIR bands 13 and 14 were estimated from water vapor content(WVC) approximated by linear regression between band-specific atmospheric transmittance and WVC [31].The modified splitwindow co-variance and variance ratio (MSWCVR) technique was adopted to retrieve WVC based on night-time brightness temperature of the two ASTER TIR bands 13 and 14[25,32].This technique involved a quadratic empirical relationship of WVC with transmittance ratio.The transmittance ratio was determined by the ratio of covariance to the variance of brightness temperature of the two adjacent TIR bands [25].

3.2.Approach for detection,spatiotemporal variation,and propagation direction of coal fire

LST differs with time,land surface cover,climate conditions,and nature of coal fires [33].So,the use of a fixed temperature threshold for multi-date LST is not scientifically correct for multitemporal coal fire detection.Raju et al.[34] reviewed various dynamic thresholding approaches (temperature threshold varies with time).However,these methods are not applicable to classify thermally anomalous pixels into surface and subsurface fire pixels.The LST profile-based statistical approach was formulated to extract surface coal fire,subsurface coal fire,and thermally anomalous transitional pixels.The eight prominent coal fire test sites were selected based on prior knowledge of fire-affected areas in the JCF obtained from the BCCL report and field observations(Fig.1).Two transects for each test site were taken in such a way that they pass through fire as well as non-fire temperature pixels,intersecting each other.The temperature values for transacting were plotted against spatial distance.The peak temperature of a profile was taken as the maximum temperature (Tmax) of the profile and the nearly flat laying tails on both sides of the LST profile represent background temperatures (TBG1,TBG2).

Fig.2 demonstrates the retrieval of the background and maximum temperatures for a set of eight LST profiles with respect to transects shown on the LST image of the JCF for 2015.The average of two background temperatures of a profile was referred to as the background temperature of that particular profile (TBG) (Fig.2).Thus for the LST image of each year,eight pairs ofTBGfor eight test sites in the JCF were formed.Now,the mean and standard deviation of 16 computedTBG(and σBGrespectively) was calculated and then+σBGwas considered as the maximum background temperature(TBGmax)for an LST image which was used to delineate thermally-anomalous transitional pixels from background pixels.Similarly,the maximum and standard deviation of observed maximum temperatures of these profiles(MTmaxand σmaxrespectively)were also determined,and then,MTmax-2σmaxwas set as a threshold temperature to detect the maximum possible surface fire pixels.

Fig.2.Background and maximum temperature retrieval for a set of eight LST profiles (AA to HH).

Next,the temperature difference (ΔT) betweenTmaxandTBGwas determined for 16 profiles and then the standard deviation of ΔT(σΔT) was derived.The σΔTgave an estimate of temperature variability due to differences in nature and intensity of transitional area and coal fires over an entire JCF viz.seam fire,dump or debris fire,and shallow and deep subsurface fire.The transitional area is a high temperature (thermally anomalous) region heated owing to either the radiation effect of surrounding fire or deep subsurface coal fire.The temperature threshold ofTBGmax+σΔTwas used to segregate subsurface fire thermally anomalous pixels from transitional pixels.The thermally anomalous pixels were categorized into three classes based on the following three criteria:(1) transitional area:TBGmaxtoTBGmax+σΔT,(2)subsurface fire area:TBGmax+-σΔTtoMTmax-2σmax,and (3) surface fire area:greater thanMTmax-2σmax.The computation of these statistics was done for each of the three dates independently using their respective LST values.

Fig.3.Night-time LST-based surface coal fire,subsurface coal fire,and transitional thermal anomalies in 2006,2009,and 2015,and in-situ coal fire observations in 2006 and 2015 overlaid on high resolution Cartosat-2 image.

Fig.3 (continued)

The coal fire-induced thermally anomalous pixels extracted from night-time LST product also included thermally anomalous pixels(false alarm pixels)due to the high thermal inertia of water bodies.These false alarms removal are essential for accurate detection and quantification of the coal fire.MNDWI threshold value greater than 0.123 was used to filter-out false alarm pixels from the coal fire thermally anomalous pixels.The obtained coal fire pixels were validated by the means of BCCL fire map and field observations.Subsequently,the spatial extent of the delineated surface fire and subsurface fire in the JCF was quantified for the years 2006,2009,and 2015.The quantified surface and subsurface fire areas for three years were compared to assess the changes in spatial extents of these fires.

Moreover,coal fires(surface plus subsurface fires)were broadly characterized into three states:new,consistent,and extinguished based on their existence during the observation period.Three conditions for categorizing coal fire states:(1) if coal fire existed for a year only but it did not exist in all previous years it was considered as new,(2) if coal fire did not exist in a year but existed in any of the previous years it was considered as extinguished,and(3)if coal fire existed in all years it was considered as consistent.Later,the three states of coal fire were mapped and quantified during these three time periods:2006–2009,2009–2015,and 2006–2015.The direction of propagation is one of the important aspects of the coal fire.In this study,the extracted coal fire pixels of 2006,2009,and 2015 were vectorized by converting them to polygons and then analyzed together to track the propagation direction of coal fire during 2006–2015 in the light of geological features and mining activities.

4.Results and discussion

In the present work,precise night-time LST products for all stipulated datasets were retrieved by robust and refined SWA using brightness temperature,scene-specific regression coefficients,and precise estimates of LSE and night-time atmospheric transmittance.The temperature distribution in the night-time LST image of the JCF for 2015 is shown in Fig.1.The LST-profile-based statistical approach was applied on retrieved LST products of 2006,2009,and 2015 to fix LST thresholds for extracting thermal anomalies related to surface fire,subsurface fire,and transitional area.The LST thresholds to delineate transitional area,subsurface fire,and surface fire respectively were found to be 295.41,299.13,and 304.78 K for 2006,290.43,295.50,and 305.29 K for 2009,and 299.8,302.6,and 309.1 K for 2015.Significant differences were observed among LST thresholds of all years which were probably due to variation in climatic conditions,nature of coal fires,land cover changes,etc.The thermally anomalous pixels of surface fire,subsurface fire,and transitional area for all years were mapped(Fig.3).

4.1.Ground observation and validation of coal fire

The coal fires in JCF mapped from ASTER LST of 2006 were validated with in-situ coal fires given by the BCCL report.It was observed that coal fires extracted from LST are comparable with in-situ coal fires (Fig.3a).However,some in-situ coal fires were not detected well from LST 2006.In-situ fires comprise surface fires,shallow and deep underground fires,and dormant fires.Sometimes,in-situ surface fire is much smaller than the pixel size because of which pixel cannot efficiently represent the coal fire temperature.Besides,deep underground fire produces weak thermal anomaly on the surface such that it becomes difficult to detect from space-based LST.It was also noted that due to low intensity,dormant fire does not get detected from satellite LST.Additionally,high speed wind and high temperature regions nearby fire can hinder the detection from LST data.Coal fires extracted from LST image 2015 were verified with locations (Fig.3c) and field photographs of coal fires taken during the field survey (Fig.4).It was found that coal fire pixels obtained from LST image 2015 matched well with coal fires observed on the ground.

Fig.4.In-situ coal fire observed at Block-II,Gaslitand,Tetulmari,Sendra Bansjora,Godhar,Alkusa,Bhowra,and Angarpathra.

Some of the on-site coal fire locations pointed out on the LSTderived coal fire map are discussed here(Fig.3c).The smouldering of coal chunks mixed with fragmented mining waste (overburden dump) was observed near Sudamdih village in eastern Jharia(location-20).The temperature range of fire-affected overburden dump (OBD) was recorded from 67 to 96 ℃.The high-intensity vent with a recorded temperature of 399.5 ℃(Fig.4g) and fireaffected coal waste pile adjacent to vent near Bhowra locality in the Eastern Jharia area were detected as subsurface fire pixel extracted from LST(location-19).The high-temperature cracks/fissures adjacent to south Tisra (location-17) and Ghanoodih colliery(location-16),due to the burning of shallow depth coal,weredelineated as subsurface fire pixels.The coal seam fires were spotted in Alkusha and Godhar mines in Kusunda area near the roads and townships.In Godhar mine,pockets of the blazing fire in exposed coal seam (Fig.4e) represent surface fire pixels extracted from the LST image.The breakage of opencast mine(OCM)benches was observed in Alkusa colliery due to the uncontrolled burning of combustible materials (carbonaceous shales or thin coal layers) in OCM benches(Fig.4f).The underground fire,as well as coal debris surface fire in Kankanee colliery,were identified on LST based coal fire map (location-9).

In Sendra Bansjora OCM,the intense and exposed fires were detected in the thin coal layer intercalated in shale strata (Fig.4d,location-8).A smoldering fire in a coal waste heap/OBD with a large spatial extent was found,with recorded temperatures ranging from 104 to 137 ℃.Due to underground fires,a large number of fractures and cracks,destruction of infrastructures,and disintegration of land surfaces were found near Tatulmari road of Sijua area in JCF (Fig.4c).These underground fires were delineated as subsurface fire pixels (location-7).The burning of coal waste heap in Gaslitand OCM in central JCF (Fig.4b) induces surface and subsurface fire anomaly pixels in the coal fire map (location-5).The smoke-emitting cracks were seen on the ground surface due to the presence of underground fire in the Angarpathra area.The temperature for most of the in-situ cracks was recorded within the range of 51–60.2 ℃(Fig.4h).It was observed that these cracks occupy a small part of 90m pixel.So,low-intensity heat dissipated from small extent cracks is not enough to raise the temperature of 90m pixel above a threshold of coal fire pixels.Consequently,thermal anomalies associated with these cracks were detected as the thermally-anomalous transitional pixels instead of coal fire pixels(location-4).In the western part of JCF,high intensity OBD fire near northeast of Muraidih workshop in Barora area was delineated as surface fire pixels (location-2).The temperature of the OBD fire was measured at more than 274.4 ℃.The surface and shallow subsurface fire with a maximum temperature of more than 350 ℃was observed adjacent to OCM of Block-II area in western JCF (Fig.4a,location-1).Field investigation revealed that the majority of the large spatial extent fires were either coal waste heap fires or OCM bench fires in abandoned mines.

4.2.Assessment of spatio-temporal variations of surface and subsurface coal fire

Surface coal fire is a manifestation of exposed coal seam fire which has extremely high temperature and small spatial extent.The spatial extent of surface fire has been quantified for 2006,2009,and 2015 in the parts of JCF (Table 1).In 2006,most of the surface fires in the JCF were located in and around Shatabdi colliery near Muraidih workshop in Barora area,West Mudidih colliery near southwest of Tetulmari in Katras area,Kusunda colliery near southeast of Kendwadih in Kusunda area,ENA Industry and Rajapur colliery in Kustore area,and Kujama colliery in Bastacolla area(Fig.3a).In 2009,the spatial extent of surface fire in these areas,except Barora area,has declined (Table 1).It was observed that from 2006 to 2009 the total surface fire area significantly decreased by 0.3159 km2(57.35%) where the largest contribution to the total decrease was of Kusunda area (Table 1).

Table 1Area-wise distribution of spatial extent of surface fire in the JCF from 2006 to 2015.

Table 2Area-wise distribution of spatial extent of subsurface fire in the JCF from 2006 to 2015.

Fig.5.Spatial distribution of coal fire states in the JCF during 2006–2009 and 2009–2015.

Surface fires in Lodna area were extinguished in 2009(Fig.3b) however,they re-appeared in 2015 (Fig.3c) with a spatial extent of 0.0405 km2(Table 1).In a period of six years i.e.2009–2015,the shallow subsurface coal fires have been exposed to surface fires due to extensive mining operations at various locations in the JCF.Consequently,new surface fires emerged in 2015 at Block-II colliery of Block-II area,Jogta,Kankanee,and Sendra Bansjora collieries of Sijua area,Bhowra colliery of Eastern Jharia area,and Chasnala colliery which were not in existence from 2006 to 2009 (Fig.3).A rapid increase of 0.3240 km2(137.93%) in the spatial extent of surface fire in the JCF was observed during 2009–2015,where the largest contribution to the total increase was of Katras area (Table 1).From 2006 to 2015,due to a decline in the spatial extent of surface fire in four areas of the JCF (including significantly decreased fire area in Bastacolla),the study observed just a 0.0081 km2increase in the total surface fire area (Table 1).

Table 3Area-wise spatial coverage of coal fire states during 2006–2015.

Fig.6.Spatial distribution of coal fire states in the JCF from 2006 to 2015.

The fire in the coal seam beneath the ground surface has been termed as subsurface coal fire.The spatio-temporal variation of subsurface coal fire was analyzed in the JCF from 2006 to 2015.Despite the increasing trend of the spatial extent of subsurface fire in the five areas of the JCF,an overall decline of 5.99%(0.2187 km2)in the subsurface fire area was noted in the JCF from 2006 to 2009(Table 2).This is because of a considerable decrease of 0.4212 km2in the spatial extent of subsurface fire in Lodna area (Table 2).In contrast,a substantial increase of 2.5677 km2(74.76%) in subsurface fire area from 2009 to 2015 was observed primarily due to new subsurface fires spotted in Katras Choitudih and Gaslitand collieries in the Katras area,and Jogta and Kankanee collieries in the Sijua area in 2015 (Fig.3c).The subsurface fire area in Katras and Sijua showed a rapid increase of 0.7857 and 0.9315 km2,respectively during 2009–2015.

It was noted that the annual average increase (0.428 km2/a) in subsurface fire area during a time span of six years i.e.2009–2015 is significantly dominating over the annual average decrease(0.073 km2/a) during 2006–2009.So,over a period of nine years i.e.from 2006 to 2015,subsurface fire has shown an increasing trend leading to a substantial increase in area by 2.349 km2.Eventually,total coal fire area (including surface and subsurface fire area) in 2006,2009,and 2015 was found to be 4.2039,3.6693,and 6.5609 km2,respectively.The total coal fire area was observed to decline by 12.72% during 2006–2009 followed by a dramatic increase of 78.81% during 2009–2015 and a significant increase of 56.10% during 2006–2015 with an annual average increase of 0.2619 km2/a.

Fig.7.Coal fire polygons for 2006,2009,and 2015,faults[27]and coal seams[35]shown on Cartosat-2 image.A,B,C,D,E,F,and G are selected test sites for the delineation of propagation directions of coal fire.

4.3.Evaluation of coal fire states

The spatial distribution of new,consistent,and extinguished coal fire pixels during 2006–2009 and 2009–2015 are shown in Fig.5.It was observed that the new fire area was significantly larger during 2009–2015 than 2006–2009.The reason behind it is the extensive opencast mining from 2009 onwards in Katras and Sijua area of the JCF,which exposed a large coal-bearing area to oxygen for ignition,resulting in many new coal fires which triggered in these areas during 2009–2015.It was also observed that the extinguishing rate of the coal fire area is comparatively higher(roughly 0.6956 km2/a) during 2006–2009 as compared to 2009–2015(0.2902 km2/a).This was due to the extensive reduction of fires in the eastern parts of the JCF (especially in Lodna,Bastacolla,and Kustore area) during 2006–2009 (Fig.5a).The fires in these parts were observed in the shallow depth coal seam and the mitigation efforts to excavate and remove these shallow fires have been carried out by BCCL on a large scale.

Furthermore,new,consistent,and extinguished coal fires in the JCF from 2006 to 2015 were identified by observing coal fires in all three years cumulatively i.e.2006–2009–2015.The area-wise spatial distribution of coal fire states during 2006–2015 is summarized in Table 3.The new coal fires in the JCF from 2006 to 2015 signify that the fire existed in 2015,however,it did not exist in 2006 and 2009.The Katras and Sijua area of central JCF accumulated more number of new fires,which were mostly concentrated in Gaslitand,and Kankanee and Jogta OCM respectively whereas Kusunda area contains the largest number of consistent fires(Fig.6,Table 3).Other than consistent fires from 2006 to 2015,some coal fires existed in 2006 which become extinct in 2009,however,these coal fires re-ignited in 2015;therefore,it was treated as consistent coal fire in 2006 and 2015 but not in 2009(Fig.6).Improper concealing or adopting ineffectual extinguishing techniques may be responsible for the re-ignition of extinct coal fires.Such fires cover a small area of 0.3321 km2in the JCF (Table 3)and most of the fires have been spotted in Kusunda area of the JCF.Likewise,some coal fires which triggered in 2009 remained active even in 2015,but these fires did not exist in 2006,i.e.the coal fires consistent in 2009 and 2015 but not in 2006 (Fig.6).

The total coal fire-affected area in 2015 was 6.5609 km2which comprised 65.56% area of new fire,16.17% area of fire which was consistent in 2015,2009,and 2006,13.21% area of fire which was consistent in 2015 and 2009 but not in 2006,and 5.06% area of fire which was consistent in 2015 and 2006 but not in 2009.It was observed that a fire area of 3.4990 km2has extinguished from 2006 to 2015.The major vanishing of past fires was found in Barora,Kusunda,Kustore,Bastacolla,and Lodna areas(Fig.6).This was probably due to various coal fire prevention and reclamation measures implemented by BCCL as well as the extinguishing of many past and shallow fires as carbonaceous materials were depleted due to uncontrolled and continuous burning for a long time.

4.4.Analysis of propagation direction of coal fire

The information of coal fire spreading direction is useful for coal mine authority for effective mine planning and for selecting the right strategy for controlling them.Coal fire propagation analysis was done for the affected areas such as A,B,C,D,E,F,and G where fire consistently existed in 2006,2009,and 2015.This analysis was carried out in consultation with the concentration of weak planes such as fractures and faults which serve as avenues for atmospheric oxygen supply to the subsurface coal fires and spatial disposition of the coal seams and geological faults (Fig.7).Opencast mining operations expose coal and combustible materials to air,resulting in the spread of fire along the direction of mining operations.It also develops multiple openings/fractures in OCM benches(coal seam and shale strata),which magnifies coal fire and facilitates rapid propagation.

Fig.8.Propagation directions of coal fire in respective test sites A,B,C,D,E,F,and G.From high-resolution Google Earth images of 2006 and 2015 for test site C,the mining area expanded in the south direction from 2006 to 2015.Cyan and red polygons indicate spatial extent of coal fire in 2006 and 2015 respectively.

At test site A (Block-II area),fire elliptically expanded in two opposite directions i.e.NW and SE (Fig.8).The NW propagation of fire was due to burning of the coal seam in up-dip direction whereas SE propagation was due to combustion of left-out coal and combustible materials during opencast mining operations.Coal fire in this area was continuously fuelled by oxygen from the fractures,developed during the mining operations and preexisting structurally weak planes (e.g.faults and fractures) which further enhanced the propagation of coal fire.At test site B (Barora),the fire was gradually shrinking during 2006–2015 (Fig.8).This was probably due to the lack of air supply through lessabundant surface cracks and fire preventing measures implemented by BCCL.At test site C (Katras),it was found that the coal mining area extended towards the South during 2006–2015(Fig.8h and i),resulting in the fire propagation in the same direction (Fig.8c).At test site D (Sijua),coal fire propagated along the direction of structurally weak planes (e.g.faults and fractures).Moreover,the spatial extent of coal fire was found to be consistently increasing during 2006–2015.

At test site E comprising Kusunda (Godhar and Kusunda collieries) and Kustore (Rajapur colliery) areas,blazing fire was observed in the exposed fire-affected coal seam through open cracks (Fig.4e).A large number of structural weak planes (faults and fractures) facilitating the supply of atmospheric oxygen into the fire-affected coal seams were observed.In Kusunda area,a spatially-extensive coal fire area has been consistently burning since 2006 despite extensive measures adopted by BCCL for mitigation of fire.Coal fire propagation along the strike direction of the coal seam was also observed (Fig.8).The direction of propagation is primarily towards NW in the vicinity of NH-32 highway and habitation near Godhar mine.It has also propagated towards SE near Jharia Township,however,with a limited rate of propagation.In Kustore area,the southern part of Rajapur colliery was affected by the abundance of structurally weak planes,surface cracks and heavy mining activities during the observation periods.Coal fire propagation was observed towards SE-SSE direction near Jharia town perhaps due to the spatial abundance of faults and fractures,and the trend of the coal seams along NW-SE direction (Fig.8).

At test site F,coal fire was spatially extensive in 2006 (from South Tisra to Ghanoodih)which splited into isolated areas during the observation period (Fig.8) perhaps due to continuous burning and opencast mining operations.The fire propagation was observed along the strike direction of the coal seam in Ghanoodih,Kuzama,and North Tisra due to spatial continuity of coal seam in strike direction and occurrences of offsets/trenches/barriers in the dip direction(Fig.8).In contrast,at test site G,coal fire propagation was noticed to be largely reduced due to consistent fire preventing measures implemented by the mining agency and less number of surface faults and lineaments.Coal fire essentially propagated along the strike direction following spatial continuity of the coal seams (Fig.8).

5.Conclusions

The spatio-temporal variation and propagation direction of the coal fire in the JCF has been successfully analyzed during 2006–2015 using precisely retrieved LST from satellite-based multitemporal ASTER night-time TIR data by a robust SWA and statistically vigorous LST-profile-based coal fire detection algorithm.Additional inputs employed in the study such as geological factors and ongoing mining operations have further added newer insights into the propagation direction of the coal fire.The major conclusions of this study are as follows.

(1) The coal fire detection algorithm holds the potential to derive date-specific threshold temperatures for efficiently separating surface coal fire,subsurface coal fire,and thermally-anomalous transitional pixels.

(2) During 2006–2009,the spatial coverage of surface and subsurface fires showed a significant decrease of 57.35% and 5.99% respectively followed by a rapid increase of 137.93%and 74.76%respectively during 2009–2015.Overall,a significantly high increasing trend in total coal fire area was noticed with an increase of 56.10%during 2006–2015 which largely resulted from new coal fires at many places (mainly in Katras and Sijua areas) that surfaced due to extensive opencast mining operations.

(3) Coal fire propagation occurred primarily along the dip direction of the coal seams.At places,coal fire propagation of limited spatial extent was also observed along the strike direction of the coal seams perhaps due to the spatial continuity of the coal seams and occurrences of offset/barrier in the dip direction.The opencast-mining activities carried out during 2009–2015 and the abundance of structurally weak planes facilitated the coal fire propagation.

(4) The coal fire propagating towards roads and habitation in the coalfield emphasizes that there is a requirement for prevention of further lateral-propagation of the coal fire.

(5) The time series night-time medium resolution LST images along with collateral metrological information has been found to be useful for more precise and accurate coal fire propagation studies.

Acknowledgements

The research work was funded by Indian Space Research Organization.The authors are thankful to Indian Institute of Remote Sensing,Dehradun and India Institute of Technology(Indian School of Mines),Dhanbad for providing laboratory facilities for conducting the research work.The authors are also grateful to the editor and reviewers whose constructive comments and suggestions significantly improved this manuscript.