Remote Sensing for Surface Coal Mining and Reclamation Monitoring in the Central Salt Range, Punjab, Pakistan

Abstract

The expansion and exploitation of mining resources are essential for social and economic growth. Remote sensing provides vital tools for surface-mining monitoring operations as well as for reclamation efforts in the central Salt Range of the Indus River Basin, Pakistan. This research demonstrates the applicability of remote sensing techniques to the coal mining monitoring scheme to allow for effective and efficient monitoring and to offset the adverse consequences of coal mining activities. Landsat 8 OLI images from June 2019 and 2020, and a Landsat 7 ETM+ image from June 2002, were used for this study. A three-phase methodology including Normalized Difference Vegetation Index (NDVI) analysis, land cover mapping, and change detection approaches was adopted. Image classification based on Tasseled Cap Transformation and the brightness temperature At-satellite using the K-means algorithm was implemented in a GIS program to identify seven land cover classes within the study area. The results show some level of surface disturbance to the landscape due to the coal mining reclamation activities that had taken place over the 18-year time period. From 2019 to 2020, about 3.622 km² of coal mines or barren land were converted into bare agricultural land. Over the years, it was also observed that reclamation areas exhibited higher values of NDVI than coal mining areas. The mean NDVI for coal mining areas was 0.252 km², and for areas of reclamation, it was 0.292 km² in 2020, while in 2019, the value for coal mining sites was 0.133 km², and 0.163 km² for reclamation sites. This trend suggests that coal-mining operations can be monitored using satellite data, and the progress of reclamation efforts can be assessed using satellite NDVI data from the target locations. This study is beneficial to agencies responsible for monitoring land cover changes in a coal mine because it provides a cost-effective, efficient, and robust scientific tool for making mine site allocation decisions and for monitoring the progress of reclamation efforts.

1. Introduction

Coal is a primary traditional energy source, and a major factor in national economic growth [1,2,3]. Over the last few decades, the surface mining of this resource has increased worldwide [4]. As of 2016, Pakistan was globally ranked as the 34th most productive coal mining country, generating more than 4.5 billion tons of coal annually (https://www.worldometers.info/coal, accessed on 25 September 2021). Additionally, wood is used in significant quantities by the coal industry [5]. Mining timber is used at a rate of roughly 56 m³ for 1000 tons of coal [6].

Pakistan is currently facing an energy shortage challenge [7]. Thus, coal is considered one of the most essential energy sources [8]. The biggest lignite coal reserves in the world have been discovered in Pakistan. Pakistan contains more than 90 billion tons of gas; 97% of its coal reserves are lignite, with only 3% maturing from sub-bitumen to bitumen [9]. The coal sector of Pakistan, which includes the water and power development authorities (WAPDA), produces sugar, steel, cement, and brick ovens for domestic use and use by other small industries [10].

Furthermore, recent projects that use coal as the main power supply, such as the One Belt One Road initiative, include industrial zone developments in various Pakistani cities. The number of large coal deposits, their potential for good returns on investment, and the speed at which coal is being burned are great threats to Pakistan’s climate. Some of the effects of charcoal mining on the climate are air pollution, deforestation, carbon emission, and problems with the quality of both surface water and groundwater [11,12,13,14,15]. One of the major threats to Pakistan’s environment is the depletion of water resources and land supplies [16]. Effluent release induces the destruction of these properties mechanically, chemically, and biologically, and leachate constantly pollutes local water and surface resources. The Indus River is Pakistan’s main river, fulfilling the water requirements of all industries for both domestic and industrial consumers. Coal is mined in different parts of the Indus basin, thereby changing the ecosystem of the basin and posing serious risks to human and environmental survival. Thus, the effects of coal mining on the Indus Basin must be evaluated [17,18]. The restoration process should facilitate the return of the land to a pre-mined or comparable condition [19].

Ground movements, collisions in mining cavities, and the distortion of aquifers are all possible outcomes of mining operations, which can also cause other geological alterations. The groundwater table could rise as a result of these processes, potentially leading to gradual subsidence of subsurface soils or to an unanticipated collapse [20,21]. Damage to soil cohesiveness and subsequent soil compression may be irreparable as a result of the extraction techniques and machinery utilized to reach mine vaults [21,22]. This can lead to the formation of unexpected water bodies, as well as to flooding due to groundwater intrusion. Consequently, this can lead to a number of detrimental long-term effects on the environment, namely the degradation of flora, the erosion of soil, inundation, the development of sinkholes, the pollution of water and soil, and the destruction of infrastructure [23,24,25]. If mines are not effectively balanced, the geological modifications and environmental problems associated with them may persist after reclamation efforts have been completed.

In order to recognize areas of successful remediation and those where management challenges occur, or reclamation activities fail, a successful remediation program should have a monitoring aspect [26]. In some situations, it may not be possible to use manual controls on time, people, and money at all mining sites where a resource manager is in charge. A cost-effective and productive alternative to tracking systems is environmental monitoring by remote sensing imagery [27]. With remote sensing software, a single analyst can look at different aspects of the mining process without ever going to the mine itself [28]. The general public is becoming more and more aware of satellite pictures. In April 2008, the United States Geological Survey (USGS) announced the full archive of satellites for Landsat earth observation (USGS, 2008) [29]. Landsat consists of six satellites deployed by NASA between 1972 and 1999. These satellites repeatedly blanket the earth, scanning many parts of the electromagnetic spectrum. When they are acquired and stored, new photographs become visible. The increase in Landsat data access dramatically expands the possibilities for the application of environmental monitoring systems using remote sensing techniques [30].

Remote sensing technology has been used in recent years to evaluate the impact of mining operations on the features of the outer surface [31,32,33,34]. For instance, Townsend et al. [35] demonstrated that the concentration of mining operations caused extensive degradation throughout the Central Appalachian Mountains in the United States prior to surface reclamation in 1977 [35]. However, both the degradation as well as variations in the LULC sequence have drastically decreased after the reclamation as of 2006. In a separate research project, Obodai et al. found that mining operations were the primary cause of deforestation in the Ankobra river basin in Ghana between the years 2008 and 2016 [36]. Cano Londoo investigated the detrimental consequences of mining operations on three different aspects: life cycle analysis, exergy examination, as well as energy accountancy [37]. The life cycle analysis examines the sustainable development of a procedure regarding environmental impacts that are caused by the pollution that is exposed to the environment. Energy has been evaluated based on its consumption of the required resources to undertake the task, and energy is evaluated based on how efficiently the process is carried out. (a) Previous research on the impacts of mining operations on surface physical characteristics has a number of drawbacks, including the fact that it mostly focused on how mining operations affected the LULC changes in the areas around mines. (b) Through this research, optical remote sensing pictures have been the only technique used to assess the influence of mineral processes upon surface physical properties [21].

If no precautionary steps are taken, the environmental consequences in arid and semi-arid areas can be greater [38]. The environmental effects include disruption of water and water demand and supply [39]. The degradation of water quality, water degradation, biodiversity, and soil quality problems impacts coal mining that directly affects farm productivity and human health [40]. In addition, the regional hydrological cycle may be affected due to changes in surface and groundwater transformation, the increasing rate of infiltration, and a decrease in the rate of evaporation [41]. An increase in weathering solutes in streams near mining areas has been recorded in previous studies [42]. The adjustments were noticed before and after coal mining operations in the concentration release relationship for the watershed [43]. In order to properly plan and manage ecological resources, especially water resources, it is important to consider the effects of coal mining caused by humans [44].

The purpose of this research is to evaluate how remote sensing techniques can be utilized as a tool to monitor surface coal mining operations. By integrating remote sensing into a monitoring regime for the Central Indus River Basin, agencies charged with monitoring surface mining activities can more efficiently help avoid potential adverse outcomes of surface mining. This research was conducted in three main phases; the first phase consists of image classification using tasseled cap data with temperature data, phase two includes NDVI analysis and land cover mapping, and the final phase consists of change detection analysis. The methods described in this research achieve rapid and efficient extraction of spatiotemporal coal mining and land reclamation processes and provide basic data for sustainable mining production and land reclamation. The primary objective of this research is to look at how the surface area of mined land, as well as LULC, has changed over time. The results of this study will also be used to describe the effect of increased coal production in the study area.

2. Materials and Methods

2.1. Study Area

The study area is situated in the Salt Range, a hill system in the Punjab province, Pakistan, with a longitude of 720,260–720,500 east and latitude of 320,220–320,440 north (Figure 1). It is a major geomorphological and ecological feature, bordered by the Thal desert in the west and Potohar Plateau in the northeast. This range begins from Potohar and ends on the northern side of the Jhelum River [45]. The Jhelum River is considered as the central hydrological unit and is one of the major tributaries of the Indus River, which passes through this region. The region is known to have low rainfall of approximately 50 cm and more precipitation periods in July, August, and September. The key category of vegetation is subtropical dry and evergreen shrubs. The location of the central Salt Range is shown in Figure 1.

The central basin of the Indus River system is sedimentary, with vast deposits of coal and other mineral resources, reflecting the Precambrian to recent stratigraphy [46,47,48]. Paleocene and Permian stratigraphy dominate carbon geology in the central Indus Basin. The Chakwal Division, Rawalpindi extracts Permian coal. Tertiary coal is also mined in the east and the central Salt Range. Coal is found in the late Paleocene Patala Formation in the Chakwal and Jhelum Districts, Province of Punjab, Pakistan. These horizons are friable, high in ash and sulfide, ripen between lignite, and are highly volatile [49]. The coal beds are typically thinly banded (bands > 3–5 feet thick) and characterized by bright bands isolated in a matrix that is dominated by dark, resinous organic content [50]. In this area, the chemical composition of coal has a fixed carbon and ash content of 13.21% to 32.79%, respectively. The sulfur level is between 5.45% and 10.63%, and the humidity content is between 3.14% and 4.26%. The proximate analysis for coal in the Salt Range is presented in the Punjab Mines and Minerals (PUNJMIN) findings, as shown in Figure 2, which show the next tests for Salt Range coal bitumen [50].

2.2. Data Collection and Methods

The process of utilizing Landsat data for monitoring surface mining in the Central Indus Basin (CIB) followed a multi-phase approach (Figure 3). Before beginning analysis, satellite images and GIS data pertaining to the study area were obtained, and preprocessing operations were completed. The first analysis phase consisted of creating classified images using tasseled cap data with temperature data. Phase two included the calculation of NDVI and creating different images to identify areas of degradation and improvement. The final phase of analysis consisted of identifying areas of change, the generation of mask images to determine the types of change that had occurred, and the inspection of possible problems or violations in the field. These phases are explained more thoroughly in the following sections.

2.2.1. Data Acquisition and Preprocessing

Before analysis, it is necessary to obtain satellite images and other required GIS data of the study area. Three Landsat scenes were selected and downloaded from the USGS Landsat archive. The satellite images of June 2002, 2019, and 2020 were acquired to study the area while vegetation growth had started, maximizing differences between vegetated and non-vegetated areas. The images selected were verified for cloud cover with the fmask plugin in QGIS for the study area and found to be free of cloud cover across the study area section. Figure 3 depicts a flow chart describing the steps involved in this study.

Two time series (2002–2019 and 2019–2010) were selected for change detection. The 2002 to 2019 series indicates long-term changes that have occurred in the Indus River Basin in the past, representing the extraction of several billion tons of coal and other development that accompanies it. The 2019 to 2020 series was chosen to detect short-term changes in the study area and identify the current situation. In terms of mining patterns, these time series are anticipated to be valuable.

Before conducting image analysis, several steps were taken to help in the process. Geo-referencing was checked to ensure the images were properly aligned with each other. The study area was extracted from the images; the extraction was helpful in shortening the processing time and reducing the amount of excess data produced by the analysis. Two subsets were made for each imagery; one contains the thermal band, and the other consists of six bands.

The thermal band has a higher resolution than the other bands of the Landsat satellite. While the non-thermal bands have a spatial resolution of 30 m, band 8 has a 15 m resolution. Files used in conjunction with thermal data were resampled to match this coarser resolution. This study employed Tasseled Cap Transformation data and at-satellite brightness temperature to map the land cover of the study area for the three chosen time periods.

Tasseled Cap Transformation (TC)

The TC was first defined for Landsat MSS, an earlier Landsat sensor, by Kauth and Thomas in 1976. The TC was later adapted for use with the six non-thermal bands of the Landsat TM sensor by Crist and Cicone in 1984. The weights below were used for the Landsat 7 ETM+ data following the procedure of Crist and Cicone [52].

The Tasseled Cap Transformation (TC) is a linear transformation that projects soil and vegetation information into a single plane in multispectral data space. The transformation permits the user to view the major spectral components of an agricultural scene as a two-dimensional figure [53].

The transformation consists of linear combinations of the six spectral bands (1–5 and 7) of the Landsat 8 TM sensor to create a set of three new variables. The first variable is interpreted as brightness and is a weighted sum of all the bands. The second new variable is greenness, which represents information pertaining to the abundance and vigor of living vegetation. The third variable represents soil wetness. The brightness and greenness components typically contain the most information in a given scene [54].

$$\mathrm{Brightness}=0.3037\mathrm{TM}1+0.2793\mathrm{TM}2+0.4343\mathrm{TM}3+0.5585\mathrm{TM}4+0.5082\mathrm{TM}5+0.1863\mathrm{TM}7$$

(1)

$$\mathrm{Greenness}=-0.2848\mathrm{TM}1-0.2435\mathrm{TM}2-0.5436\mathrm{TM}3+0.7243\mathrm{TM}4+0.0840\mathrm{TM}5-0.1800\mathrm{TM}7$$

(2)

$$\mathrm{Wetness}=0.1509\mathrm{TM}1+0.1793\mathrm{TM}2+0.3299\mathrm{TM}3+0.3406\mathrm{TM}4-0.7112\mathrm{TM}5-0.4572\mathrm{TM}7$$

(3)

The weights below were used for the Landsat 8 OLI data following the procedure of Muhammad et al. [55].

$$\mathrm{Brightness}=\mathrm{TM}2\times 0.3029+\mathrm{TM}3\times 0.2786+\mathrm{TM}4\times 0.4733+\mathrm{TM}5\times 0.5599+\mathrm{TM}7\times 0.1872$$

(4)

$$\mathrm{Greenness}=\mathrm{TM}2\times \left(-0.2941\right)+\mathrm{TM}3\times \left(-0.243\right)+\mathrm{TM}4\times \left(-0.5424\right)+\mathrm{TM}5\times 0.7276+\mathrm{TM}7\times \left(-0.1608\right)$$

(5)

$$\mathrm{Wetness}=\mathrm{TM}2\times 0.1511+\mathrm{TM}3\times 0.1973+\mathrm{TM}4\times 0.3283+\mathrm{TM}5\times 0.3407+\mathrm{TM}7\times \left(-0.4559\right)$$

(6)

A tasseled cap image was created for each image. The best data identifying mining features found in the greenness band of the three bands generated with the TC transformation. While brightness and wetness values varied over different portions of a given mine, greenness values were very uniform, producing a mostly complete footprint of each mine. Because of these qualities, the greenness band was chosen for unsupervised classification, while the brightness and wetness bands were left out.

At-Satellite Brightness Temperature

Reflectance in the far-infrared (thermal) band of Landsat data can be manipulated into an approximate measure of surface temperature. Even though this method may be a little bit off when it comes to the actual surface temperatures, it can still be used to study how temperatures vary across an area.

Brightness temperature is first converted into radiance values using the following equation:

$$\mathrm{L}=\left(\frac{\mathrm{Lmax}-\mathrm{Lmin}}{255}\times \mathrm{DN}+\mathrm{Lmin}\right)$$

(7)

L is the radiance of a given pixel, Lmax is the maximum radiance detectable by the satellite, Lmin is the minimum reflectance measurable, and DN is the pixel’s digital number or brightness value. Radiance is subsequently converted to temperature using the equation:

$$\mathrm{T}=1282.7108/\ln \left(\frac{666.093+\mathrm{L}}{\mathrm{L}}\right)$$

(8)

T is the apparent surface temperature, and L is the radiance value of the pixel [56]. Using the study area subset of the far-infrared band, brightness values were converted to radiance, then to temperature in Kelvin, and finally to degrees Celsius. Temperature data were combined into a single file with tasseled cap data and subsequently used in unsupervised classification.

Unsupervised Classification

Two main types of image classification are supervised classification and unsupervised classification. Supervised classification requires the user to select training areas that represent the desired classes to be identified. A classification algorithm then places each pixel into a category based on which training area or land cover category the pixel is most similar, spectrally. It requires strong aerial photography interpretation skills and adds an opportunity for human error.

Unsupervised classification is the process of placing pixels into categories or classes [7,53,57,58]. Image classification is based on the idea that similar objects have similar spectral properties. For example, water reflects and absorbs light differently than grass or concrete. Digital classification considers brightness over multiple data bands—for each pixel within an image—to group similar pixels (or similar land uses/land covers).

Unsupervised classification attempts to find natural groupings of pixels within an image without adding input from the user. Following classification, the user must identify what each class represents and determine whether it corresponds with its land use/land cover class or is part of another category. Unsupervised classification is advantageous in this study because the risk of operator error is much less compared to supervised classification, and extensive knowledge of the study area is not necessary.

The unsupervised classification was conducted using the K-means algorithm. The K-means algorithm begins by arbitrarily locating a point for each desired class within a given image. Next, each pixel within the image is classified into the nearest cluster. In the third step, new mean vectors are calculated based on all the pixels in each class. The second and third steps are repeated until the change between each iteration reaches a user-defined threshold. This change can be defined as the distance mean cluster vectors move between iterations or the number of pixels that change classes.

A classified image was created for each image date using the thermal band and the greenness portion of the tasseled cap transform. The k-means algorithm initially identified 15 classes: Mine/Barren, Grassland, Shrub/Scrub, Rangeland, Riparian and Reclaimed areas, etc. Some of the classes were grouped, and four major classes were acquired. After combining classes, each class was assigned a color representing the land cover type it symbolized. The final two images created a mask image for the Mine/Barren landcover class. These images show what land cover classes existed at new mining areas in the previous image date’s classification.

Normalized Difference Vegetation Index Change Analysis

The NDVI was developed in the late 1960s and has proven to be a useful indicator for measuring photosynthetic activity in vegetation. There is an inverse relationship between vegetation reflectance in the visible red and near-infrared portions of the electromagnetic spectrum. Chlorophyll absorbs red light, while mesophyll tissue reflects infrared radiation. Healthy vegetation absorbs more red light and reflects more infrared light than unhealthy or strained vegetation. NDVI is useful for revealing the latent information within this inverse relationship [59]. The NDVI is calculated as the difference between the red and infrared bands divided by the sum of the red and infrared bands. For Landsat data, this is simply:

$$\frac{\mathrm{Band}3\left(\mathrm{Red}\right)-\mathrm{Band}4\left(\mathrm{IR}\right)}{\mathrm{Band}3\left(\mathrm{Red}\right)+\mathrm{Band}4\left(\mathrm{IR}\right)}$$

(9)

For the Landsat 7 ETM+ data, it is:

$$\frac{\mathrm{Band}4\left(\mathrm{Red}\right)-\mathrm{Band}5\left(\mathrm{IR}\right)}{\mathrm{Band}4\left(\mathrm{Red}\right)+\mathrm{Band}5\left(\mathrm{IR}\right)}$$

(10)

This difference/sum ratio reduces influences from the atmosphere, local topography, and image aspect or shading, which increases its value compared to other vegetation indices. While NDVI has been shown to work best in highly vegetated areas, such as forests, it can still be a useful tool in regions with less vegetative cover [60]. Live vegetated areas typically have values from 0.4 to 1.0, whereas other non-vegetated areas normally have values below 0.4 and often negative values [53]. In the same way, places with healthy plants are likely to have higher values than places with sick or stressed plants.

NDVI images were generated for each image using the red and near-infrared bands, and the difference was calculated by subtracting the earlier or initial image from the final. The 2002 NDVI image was subtracted from the 2019 image and the 2019 image from the 2020 image. The resultant values lie within −0.7 to 0.8. Areas that had improved vegetative cover contained values greater than zero. In contrast, areas that were disturbed or had decreased vegetative cover consisted of values less than zero.

Post-Classification and Processing Analysis

After creating land cover maps and NDVI difference images, the two were compared. Coal mining and reclamation areas were analyzed by comparing the classified images with the NDVI images created from each time series. This method helped recognize areas affected by mining, pre-recovered areas disturbed by mining, previously disrupted areas where vegetation was not recovered, and areas without any changes.

Utilizing statistical indices derived from satellite images, including the NDVI, could be a very useful tool for providing an adequate evaluation of the detrimental effects that mining activities have on surface features. For surface features modeling, NDVI is the most essential and useful indication [61,62]. Because interactions between surfaces are complicated, integrating optical and thermal remote sensing data from the surface can improve surface modeling [63,64]. Modeling the effects of anthropogenic activity on surface features is critical. Furthermore, forecasting future patterns of these alterations is vital and valuable to planning and management to mitigate these variations’ detrimental impact [65,66,67].

3. Results and Discussion

This study was based on seven coal mines in the central Indus River Basin. This region contributes enormous quantities of coal to the coal production mined in Pakistan and will likely see an increase in mining over the next few decades. Mine monitoring and reclamation efforts in the region are unsuitable due to various factors, including extraneous workloads and economic constraints. Utilizing Landsat satellite images through the U.S.G.S. archive makes it possible to monitor surface mining and reclamation operations and characterize land cover changes resulting from mining

3.1. NDVI Analysis

The NDVI images of the study area produced clear and complete footprints of each mine within the study area (Figure 4, Figure 5, Figure 6, Figure 7 and Figure 8). Active mine areas were strongly associated with low NDVI values compared to the rest of the landscape. A striking pattern of change can be seen across the area from 2002 to 2020. Mining operations make up a much smaller percentage of the total land area in 2020 than in 2002. Mining areas are clearly in evidence in the NDVI images. In this figure, low NDVI values correspond to the lighter tone and indicate areas where vegetation has been disturbed or removed. A comparison of the dates illustrates the intensification in mining activities between 2002 and 2020 and a shift in mining locations as they slowly progress toward the center of the Central Indus Basin.

Areas with the highest NDVI values include reclaimed or revegetated areas and riparian zones near the rivers and streams in the study area. Areas undergoing active reclamation have higher NDVI values. Riparian zones exhibit high NDVI values because vegetation grows more densely in these areas. However, the highest NDVI value is observed in the 2020 NDVI image (Figure 4), where values ranged between 0.5 and 0.7. It is likely due to differences in the precipitation patterns between the three years, 2002, 2019, and 2020 (Figure 4, Figure 5 and Figure 6). Areas of vegetation exhibit higher NDVI values as depicted from the NDVI analysis based on Landsat 8 OLI image for the study year June 2020, June 2019, and June 2002.

Typically, wet areas, including rivers and wetlands, exhibited low NDVI values across the landscape and in active mining areas. Few areas are vegetated in the June 2002 image compared to June 2019 and June 2020. In the difference image created by subtracting the 2019 image from the 2020 image (Figure 7) and subtracting the 2002 image from the 2019 image (Figure 8), newly disturbed areas have the lowest values, while reclaimed areas typically have the highest. Most images with values between disturbed and reclaimed areas show vegetation that has not changed between the dates of the images.

NDVI Analysis—Coal Mines and Areas of Reclamation

Figure 9, Figure 10, Figure 11, Figure 12, Figure 13 and Figure 14 show the distribution of NDVI values over the coal mining and reclamation areas within the study area over 2002, 2019, and 2020. In every case, areas that had been reclaimed had higher NDVI values than areas where coal was mined.

Across the years, it has been observed that reclamation areas exhibit higher values of NDVI than areas of coal mining, and the reclamation works have increased over time simultaneously with mining works. For instance, the mean NDVI for coal mining areas in 2020 was 0.252, while for areas of reclamation, it was 0.29 (

Table 1. Statistical comparison of NDVI values for coal mining and reclamation sites from 2002 to 2020.

Variation in NDVI Values across the Areas of Coal Mining and Areas of Reclamation within the Study Area
	Coal 2020	Reclaim 2020	Coal 2019	Reclaim 2019	Coal 2002	Reclaim 2002
1st quartile	0.03	0.07	0.05	0.08	0.09	0.07
2nd quartile	0.19	0.23	0.11	0.14	0.12	0.12
Median	0.25	0.29	0.13	0.16	0.13	0.13
4th quartile	0.30	0.34	0.15	0.18	0.14	0.15
5th quartile	0.46	0.50	0.20	0.24	0.17	0.19
Mean	0.25	0.29	0.13	0.16	0.13	0.13

).

3.2. Land Cover Mapping

Land cover mapping was conducted to assess the effect of mining on the landscape and track the mines’ progress. Land cover mapping was accomplished by applying the K-means unsupervised classification algorithm to a composite image of the greenness band of the Tasseled Cap Transform and a measure of at-satellite brightness temperature. The classification found 20 spectrally unique classes, subsequently combined into seven mainland cover categories (

Table 2. Classification System for Landcover Mapping.

Class Name	Description
Coal Mine/Barren Lands	Land disturbed by active mining and other non-vegetated areas.
Reclaimed Lands	Areas undergoing the revegetation stage of reclamation
Shrub	Land covered by sparse vegetation.
Water bodies	Wet areas are largely composed of the river but also streams
Cropped Agricultural Lands	Lands with crops
Bare Agricultural Lands	Agricultural lands that are currently without crops
Wetlands	Swampy areas or areas that always have water

). The results of land cover mapping confirm and support the findings of NDVI analysis.

Land cover classification produced noticeably clear results. A visual comparison of the three images reveals the trend in mining activities over 2002, 2019, and 2020 (Figure 15, Figure 16 and Figure 17). Several land cover change regimes were identified by utilizing these imageries. The most striking trend is the undisturbed land (shrubs) in an earlier image that has been disturbed by mining (Figure 16 and Figure 17). Some areas change from active mining sites to reclaimed areas or shrub designations. Other areas classified as reclaimed have reverted to shrub classifications as vegetation has become stable and integrated into the surrounding area (

Table 3. Area (km²) of each land cover type from 2002 to 2020.

Area (km2)
Land Cover Type	2002	2019	2020
Coal Mines/Barren Lands	174.20	149.84	77.20
Reclaimed Lands	95.26	180.75	243.82
Shrubs	138.67	110.44	119.93
Water bodies	87.02	26.08	17.37
Cropped Agricultural Lands	159.76	234.46	274.81
Bare Agricultural Lands	326.42	269.52	309.09
Wetlands	141.51	149.73	80.64

and

Table 4. Area changes between 2019 and 2020. The highlighted blue color shows reclaimed lands of coal mines.

Change (2019–2020)	Area Change (km2)
Bare Agricultural Lands—Bare Agricultural Lands	249.72
Bare Agricultural Lands—Coal Mines/Barren Lands	0.12
Bare Agricultural Lands—Cropped Agricultural Lands	13.93
Bare Agricultural Lands—Reclaimed Lands	4.82
Bare Agricultural Lands—Shrubs	0.48
Bare Agricultural Lands—Water Bodies	0.00
Bare Agricultural Lands—Wetlands	0.43
Coal Mines/Barren Lands—Bare Agricultural Lands	3.62
Coal Mines/Barren Lands—Coal Mines/Barren Lands	23.19
Coal Mines/Barren Lands—Cropped Agricultural Lands	25.00
Coal Mines/Barren Lands—Reclaimed Lands	84.04
Coal Mines/Barren Lands—Shrubs	10.78
Coal Mines/Barren Lands—Water Bodies	0.07
Coal Mines/Barren Lands—Wetlands	3.08
Cropped Agricultural Lands—Bare Agricultural Lands	22.43
Cropped Agricultural Lands—Coal Mines/Barren Lands	26.80
Cropped Agricultural Lands—Cropped Agricultural Lands	159.48
Cropped Agricultural Lands—Reclaimed Lands	9.94
Cropped Agricultural Lands—Shrubs	1.57
Cropped Agricultural Lands—Water Bodies	1.26
Cropped Agricultural Lands—Wetlands	12.967
Reclaimed Lands—Bare Agricultural Lands	3.27
Reclaimed Lands—Coal Mines/Barren Lands	12.79
Reclaimed Lands—Cropped Agricultural Lands	20.70
Reclaimed Lands—Reclaimed Lands	115.28
Reclaimed Lands—Shrubs	28.52
Reclaimed Lands—Wetlands	0.16
Shrubs—Bare Agricultural Lands	0.14
Shrubs—Coal Mines/Barren Lands	0.68
Shrubs—Cropped Agricultural Lands	4.52
Shrubs—Reclaimed Lands	26.88
Shrubs—Shrubs	78.20
Shrubs—Wetlands	0.07
Water Bodies—Bare Agricultural Lands	0.09
Water Bodies—Coal Mines/Barren Lands	9.42
Water Bodies—Cropped Agricultural Lands	0.497
Water Bodies—Reclaimed Lands	0.08
Water Bodies—Shrubs	0.09
Water Bodies—Water Bodies	15.97
Water Bodies—Wetlands	0.17
Wetlands—Bare Agricultural Lands	29.56
Wetlands—Coal Mines/Barren Lands	3.96
Wetlands—Cropped Agricultural Lands	49.91
Wetlands—Reclaimed Lands	2.42
Wetlands—Shrubs	0.06
Wetlands—Water Bodies	0.01
Wetlands—Wetlands	63.76

).

Mines were haphazardly developed in 2002 (Figure 15). However, as time progressed, the level of the organization increased, and the development trends tended to offer some level of sanity. So, by 2019 and 2020, the mining sites looked organized (Figure 16 and Figure 17). Between 2019 and 2020, the number of pixels in the Mine/Barren category decreased (Table 3, Figure 18). More reclaimed areas are visible in the 2019 and 2020 land cover characterizations.

Several land cover changes took place throughout coal mining within the landscape of the central Salt Range in the Indus Basin. It is observed that within a span of almost 1 year (i.e., 2019 to 2020), the total mining area reclaimed amounted to 84.04 km² (Table 4 and Figure 19) (highlighted blue). However, from 2002 to 2019, the total coal mining areas that were reclaimed were 51.4 km² (

Table 5. Area changes between 2002 and 2019. The highlighted yellow color shows reclaimed lands of coal mines.

Change (2002–2019)	Area Change (km2)
Bare Agricultural Lands—Bare Agricultural Lands	241.64
Bare Agricultural Lands—Coal Mines/Barren Lands	5.39
Bare Agricultural Lands—Cropped Agricultural Lands	36.80
Bare Agricultural Lands—Reclaimed Lands	3.57
Bare Agricultural Lands—Shrubs	0.26
Bare Agricultural Lands—Water Bodies	0.00
Bare Agricultural Lands—Wetlands	38.40
Coal Mines/Barren Lands—Bare Agricultural Lands	6.79
Coal Mines/Barren Lands—Coal Mines/Barren Lands	76.08
Coal Mines/Barren Lands—Cropped Agricultural Lands	11.60
Coal Mines/Barren Lands—Reclaimed Lands	51.42
Coal Mines/Barren Lands—Shrubs	20.26
Coal Mines/Barren Lands—Water Bodies	2.81
Coal Mines/Barren Lands—Wetlands	4.80
Cropped Agricultural Lands—Bare Agricultural Lands	3.81
Cropped Agricultural Lands—Coal Mines/Barren Lands	14.79
Cropped Agricultural Lands—Cropped Agricultural Lands	105.88
Cropped Agricultural Lands—Reclaimed Lands	14.88
Cropped Agricultural Lands—Shrubs	3.17
Cropped Agricultural Lands—Water Bodies	3.76
Cropped Agricultural Lands—Wetlands	12.83
Reclaimed Lands—Bare Agricultural Lands	0.04
Reclaimed Lands—Coal Mines/Barren Lands	12.44
Reclaimed Lands—Cropped Agricultural Lands	1.20
Reclaimed Lands—Reclaimed Lands	74.32
Reclaimed Lands—Shrubs	7.05
Reclaimed Lands—Water Bodies	0.08
Reclaimed Lands—Wetlands	0.11
Shrubs—Bare Agricultural Lands	1.60
Shrubs—Coal Mines/Barren Lands	23.77
Shrubs—Cropped Agricultural Lands	2.34
Shrubs—Reclaimed Lands	30.87
Shrubs—Shrubs	79.13
Shrubs—Water Bodies	0.01
Shrubs—Wetlands	0.56
Water bodies—Bare Agricultural Lands	3.78
Water bodies—Coal Mines/Barren Lands	11.87
Water bodies—Cropped Agricultural Lands	43.23
Water bodies—Reclaimed Lands	2.81
Water bodies—Shrubs	0.53
Water bodies—Water Bodies	19.14
Water bodies—Wetlands	5.43
Wetlands—Bare Agricultural Lands	11.83
Wetlands—Coal Mines/Barren Lands	5.462
Wetlands—Cropped Agricultural Lands	33.39
Wetlands—Reclaimed Lands	2.85
Wetlands—Shrubs	0.02
Wetlands—Water Bodies	0.32
Wetlands—Wetlands	87.58

and Figure 20) (highlighted yellow).

The results of land cover mapping illustrate decreased surface disturbance across the study area caused by reduced mining from 2002 to 2020. A direct comparison of land cover maps produced from two different image dates was useful for visually assessing change. However, determining exactly where that change occurred and exactly which land cover types had been converted proved more difficult. Class masks were used to aid in this process and learn more about the land cover changes across this portion of the central Indus River Basin over the 18-year time horizon. Class masks are useful for observing areas where the land cover has changed between image dates. These images show areas a class occupies in a later image compared to the previous image data (Figure 16 and Figure 17). For example, a mask for the Mine/Barren designation from the 2019–2020 time series allows the user to quickly identify lands within the study area disturbed by mining over the previous year (Figure 19 and Table 4). Masks can also help to confirm the locations of lands currently undergoing reclamation. A mask of the 2019 Mine/Barren class compared to the 2002 land cover classification shows what land cover types were present in 2002 in areas mined in 2019 (Figure 20, Table 5).

4. Conclusions

The central Salt Range of the Indus River Basin is contributing fairly to the coal resources of Pakistan, and coal production is expected to increase during the next 30 years. Increased coal production increases the risk of environmental degradation. Monitoring the mining and reclamation of these lands is critical to ensure they are returned to their pre-mining state or a similar ecological trajectory. Reclamation efforts within coal mines in Pakistan are in distress as ever-growing workloads, and budgetary concerns already stress government agencies charged with monitoring mining activities.

Landsat images were acquired in 2002, 2019, and 2020 for analysis. The NDVI was used to map active mining and reclamation areas, and NDVI difference images were created to assess the growth of mining operations over the last 18 years. The unsupervised classification was performed using the greenness band of the Tasseled Cap Transform and at-satellite brightness temperature to generate land cover maps for each of the three years. It is observed that over the period from 2002 to 2019, the total coal mining areas that were reclaimed were 51.423365 km², while for a span of almost 1 year (i.e., 2019 to 2020), the entire mining areas reclaimed amounted to 84.046902 km².

The results of this study illustrate the environmental changes brought on by increased coal mining in the Indus River Basin. Some of the streams in the study area have been negatively impacted or destroyed by mining activities. While mining permits may have authorized these changes, monitoring mining activities is critical to minimize the negative effects on the environment. The remote sensing techniques used in this study were useful tools capable of monitoring disturbance from surface mining activities. NDVI analysis provided clear, complete footprints of active mines that can be used to determine if mining disturbance exceeds what is permitted. Differencing NDVI images helped to quickly identify areas where vegetation had increased or decreased. Land cover mapping with Tasseled Cap Transformation and thermal data was a practical method of monitoring long-term land cover and use changes. When used in conjunction with NDVI change data, land cover masks were also helpful for monitoring disturbance and reclamation in the study area annually. These techniques are also applicable to other types of surface mining operations. These techniques could be used to delineate the extent of surface mining anywhere there is some sort of vegetation cover surrounding the mining area.

Future research in this area could include monthly attempts to detect changes in mines and reclaimed lands and determine how NDVI changes throughout the year. In addition, monitoring using a sensor with a smaller spatial resolution may enable the user to identify smaller mining features such as waste overburden piles or topsoil storage areas. A satellite with a finer resolution may also allow for better monitoring of reclamation areas.