A Novel Approach to Automatically Identify Open-Pit Coal Mining Dynamics Based on Temporal Satellite Images

Abstract

Open-pit coal mining drives socioeconomic development but imposes significant environmental impacts. The timely monitoring of mining dynamics is essential for sustainable resource exploitation and ecological restoration. However, existing studies often rely on predefined mining boundaries, limiting their applicability in unknown regions. This study proposes an innovative approach that leverages the intra-annual coal frequency index (ACFI) to identify potential open-pit mining areas, and integrates the Rays method to monitor their temporal changes. By applying a predefined discriminative rule, this approach effectively distinguishes open-pit coal mines from other disturbances and enables spatiotemporal monitoring without the need for prior knowledge of their locations. Applied to the Chenbarhu Banner coalfield, Inner Mongolia, the method achieved 92% accuracy and a kappa coefficient of 0.84 in identifying mining areas. It effectively distinguished active and closed mines, detecting key temporal features with 94% accuracy (kappa = 0.86). The study also identified mining directions and extents, such as 4–13° for the Baorixile mine and 69–141° for the Dongming mine, while excluding non-mining areas with high precision. A strong correlation (r = 0.929, p < 0.01) between annual mining area and coal production further validated the approach. This method provides accurate, scalable tools for monitoring mining dynamics and supports decision-making in regulatory and ecological management processes.

1. Introduction

The mining industry, serving as a cornerstone of the global economy, plays a pivotal role in driving economic growth on a global scale [1,2,3]. Among various resources, coal stands out as one of the most critical energy sources due to its abundant reserves, wide distribution, and cost-effectiveness [4,5]. Open-pit coal mining holds a significant role in coal resource development due to its high efficiency and cost-effectiveness [6]. While improving resource extraction efficiency, it is typically accompanied by topsoil removal and significant alterations to topography and geomorphology [7,8,9]. These activities profoundly impact land-use patterns and ecosystem structures [10,11,12,13], posing substantial challenges to the sustainability of the regional environment.

In this context, particularly in scenarios where mining location information is unavailable, accurately monitoring open-pit coal mining dynamics—encompassing critical information such as initiation and cessation times, mining extents, mining directions, and mining statuses (e.g., active or closed)—is essential for the sustainable development and effective management of coal resources [14,15]. It not only reveals the spatiotemporal trajectories of resource extraction but also provides a reliable basis for evaluating the extent and long-term consequences of mining activities on surrounding areas, including the impacts of open-pit mining on hydrological regimes [16], air quality [17], climate change [18,19], and biodiversity [20,21]. Furthermore, distinguishing mining statuses helps identify the current phase of resource development and offers scientific guidance for land reclamation planning and concurrent mining and reclamation efforts.

Existing open-pit coal mining monitoring methods that rely on bare coal spectral features exhibit fundamental mechanistic limitations [22,23,24,25,26]. Extensive accumulations of bare coal occur not only in open-pit coal mines but also in coal storage yards and temporary coal piles, resulting in highly similar spectral signatures across these areas. Consequently, reliance solely on the bare coal index hampers the accurate discrimination of open-pit coal mines. Although manual interpretation can partially address this issue when detailed mine information is available, this approach is labor-intensive and prone to classification errors [27]. This challenge is particularly acute in the absence of mining area information, significantly complicating large-scale monitoring of open-pit coal mines. Although previous studies have attempted to combine publicly available POI data with multi-buffer analysis to locate open-pit mining areas without prior mining information [28,29], the effectiveness of this approach is limited by the spatial accuracy and update frequency of the POI data [30,31]. Furthermore, while the ACFI method has been successfully applied to monitor open-pit coal mining areas by constructing an intra-annual coal frequency index (ACFI) time series for each pixel to capture pixel-level changes [32], it relies solely on the frequency of bare coal pixel occurrences to filter out disturbances such as misidentified areas and temporary coal storage sites. Consequently, it remains challenging to fully distinguish between open-pit mining areas and long-term bare coal regions (e.g., plant storage areas and underground mine areas). Furthermore, when employing the ACFI for open-pit coal mine identification, bare coal patches are required to contain at least 50 bare coal pixels to be recognized as an open-pit mining area. This threshold may exclude smaller mining areas, thereby compromising the comprehensiveness and accuracy of monitoring. More importantly, the method is limited in its ability to capture key dynamic information of mining activities, such as the mining direction and mining statuses (e.g., active or closed). Additionally, while some studies have attempted to monitor open-pit coal mine information by analyzing vegetation index changes induced by mining activities in conjunction with change detection algorithms [33,34,35], these approaches are often severely affected by natural factors in arid and semi-arid grassland regions. Consequently, it remains challenging to discern whether observed changes in vegetation indices are attributable to mining-induced vegetation damage or to natural drought fluctuations [36]. In summary, current monitoring methods typically consider only a single factor. Given that open-pit coal mining is a dynamic process, most existing studies focus either on spectral information or vegetation dynamics, without effectively integrating bare coal area data and their temporal variations for a comprehensive analysis.

Based on the aforementioned issues, the main objectives of this study are: (i) to integrate the bare coal index with dynamic mining process detection to develop an automatic method for identifying large-scale open-pit coal mining areas in the absence of predefined location information, and (ii) to enable the rapid and accurate capture of open-pit coal mining dynamics, including key information such as mining start and stop years, mining extent, mining direction, and mining status. We believe that the proposed approach holds significant potential for rapidly and accurately identifying the open-pit coal mining dynamics without prior knowledge of mine locations. Additionally, it offers robust technical support for government agencies in their regulatory and verification efforts.

2. Materials

2.1. Study Area

Chenbarhu Banner coalfield (49°23′24″N, 119°42′57″E) is located in the northeast of the Inner Mongolia Autonomous Region (Figure 1). The general geomorphological structure of the coalfield is defined by the Chenbarhu Banner coalfield basin, extending in a northeast to east–west direction at the center. To the south lies the Hailar river valley plain, while hills encircle the region to the north, east, and west. The Chenbarhu Banner coalfield basin features a high plain in the east and an alluvial plain in the west, with scattered hills, plateaus, and river valleys forming transition zones between major geomorphological units, resulting in a complex and intricate topography. The overall terrain slopes from high in the north to low in the south, and from high in the east to low in the west, with elevation generally decreasing from northeast to southwest. The coalfield lies within a semiarid region of the mid-temperate zone, characterized by cold, extended winters and hot, brief summers. The mean annual temperature is −1.5 °C, with an average wind speed of 3.5 m/s. Annual precipitation averages 306.8 mm, predominantly occurring between July and August, while average annual evaporation reaches 1329.0 mm.

Based on the previous survey, the study area comprises two active open-pit coal mines, two underground coal mines, and one open-pit mine that has ceased operations. The basic information for each mine is presented in

Table 1. Overview of Chenbarhu Banner coalfield coal mine.

Coal Mine ID	Coal Type	Mine Construction Time	Mode of Production
Dongming (a)	lignite	2008	open-pit
Baorixile (b)	lignite	2000	open-pit
Husheng (c)	lignite	2005	underground
Tianshun (d)	lignite	2006	underground
Shunxing (e)	lignite	2006	open-pit

. The Dongming coal mine was established in 2008 with an initial production capacity of 0.6 Mt/y. By 2021, its capacity had been expanded to 4 Mt/y. The Baorixile open-pit mine is located approximately 30 km north of Hailar District. The construction of the mine began in 1998, with production commencing in 2000. The mine sits at an elevation ranging from 660 to 545 m. As of 2022, the mine’s approved production capacity was 35 Mt/y. The Husheng coal mine began construction in February 2006, with trial production initiated in March 2011 and an initial designed capacity of 1.2 Mt/y. By the end of March 2021, its approved production capacity had increased to 1.8 Mt/y. The Tianshun coal mine began construction in early 2007 and formally entered production in the latter half of 2008. By the end of March 2021, its approved production capacity had reached 1.2 Mt/y. The Shunxing coal mine operates at an elevation between 724 and 601 m above sea level. Mining began in 2006, with a design production capacity of 0.6 Mt/y. The mine was closed and decommissioned in 2022.

2.2. Image Preprocessing

Due to poor image quality in the study area during winter, we utilized all available standard level 1 Terrain-Corrected (L1T) orthorectified Landsat images with less than 10% cloud cover from 1 April 2000 to 30 September 2024 (

Table 2. Overview of data used in the study.

	Acquisition Time of Satellite Imagery	Number of Satellite Images
Landsat 5 TM	2000–2011	103
Landsat 7 ETM+	2000–2023	225
Landsat 8 OLI/TIRS	2013–2024	112
Landsat 9 OLI-2/TIRS-2	2022–2024	32

), archived as part of the United States Geological Survey (USGS) Landsat 5/7/8/9 Surface Reflectance (SR) dataset on the Google Earth Engine (GEE) platform [37]. Additionally, to ensure the quality of the acquired images, we applied the CFMASK to remove pixels affected by cirrus, cloud shadows, ice, and snow [38,39]. Figure 1 illustrates the total number of images used for data analysis within the study area from 2000 to 2024. All pre-processing steps, as described above, were conducted on the Google Earth Engine (GEE) platform.

3. Methods

3.1. Overview of Methodology

The major data processing and methodologies for the study were divided into three parts and are illustrated in Figure 2: (i) Construction of the intra-annual coal frequency dataset to access potential open-pit mining areas. (ii) Calculation of inter-annual distance between the ray origin (RO) and the bare coal boundary based on the Rays method for potential open-pit mining area. (iii) Identification of open-pit coal mines and monitoring mining dynamics. First, we utilized the GEE platform to identify bare coal areas within the study region. The Landsat 5/7/8/9 surface reflectance (SR) data were pre-processed by masking snow and cloud cover. The intra-annual coal frequency index dataset was subsequently calculated using the Bare Coal Index (BCI) and ACFI from images spanning 2000 to 2024 [26], enabling to the determination of potential open-pit mining areas during this period. However, this area encompassed not only open-pit mining sites but also coal storage zones, factory coal piles, and shaft mines. To accurately identify the open-pit mining areas and their mining dynamics, we constructed the ray network based on potential open-pit mining areas. For each independent bare coal patch, a ray origin was selected, and the spacing angle (SA) between adjacent rays was defined. By analyzing the distance variation in rays in different directions within the network, directional changes in the open-pit mining areas were automatically extracted. The accuracy of identification results was subsequently validated.

3.2. Construction of Intra-Annual Coal Frequency Timeseries

The first step in identifying open-pit coal mining dynamics is to accurately delineate the extent of the bare coal. This study developed an intra-annual coal frequency dataset using BCI and ACFI to address limitations in existing bare coal identification methods, enabling a more accurate delineation of potential open-pit mining areas, as detailed below [26,32,40,41,42].

As illustrated in Figure 3, all available Landsat imagery within the study area was screened and preprocessed, and the BCI was applied to generate a time series dataset of bare coal region. By overlaying all valid bare coal region images, we captured the spatial distribution of regions classified as bare coal at least once during the study period, representing the maximum cumulative bare coal extent. However, this approach also includes misclassified areas alongside accurately identified bare coal regions. To mitigate this issue, we computed the bare coal frequency for each pixel throughout the time series and introduced the Annual Coal Frequency Index (ACFI). By applying an optimal threshold to the ACFI, potential open-pit coal mines can be accurately identified and their spatial distribution effectively mapped. The BCI and ACFI calculation formula are presented as follows:

$$\mathrm{BCI}={\displaystyle \frac{R_{Swir2}-R_{Swir1}}{R_{Nir}\times \left(R_{Nir}+R_{Swir1}\right)}},$$

(1)

$$\mathrm{ACFI}={\displaystyle \frac{\sum N_{BareCoal}}{\sum N_{Total}-\sum N_{Bad}}}.$$

(2)

In Equation (1), $R_{Nir}$, $R_{Swir1}$, and $R_{Swir2}$ are the reflectance of the $Nir$, $Swir1$, and $Swir2$ bands, respectively. In Equation (2), $\sum N_{BareCoal}$ represents the number of images in which a pixel is classified as bare coal across all years based on BCI, while $\sum N_{Total}$ denotes the total number of available images, and $\sum N_{Bad}$ indicates the number of images exhibiting high cloud cover or poor quality. The term $\sum N_{Total}-\sum N_{Bad}$ represents the total number of images that meet the criteria for time-series monitoring. The ACFI ranges from 0 to 1, where a value of 0 indicates that no bare coal pixels are detected in any image, while a value of 1 signifies that bare coal pixels are present in every image. To determine the optimal identification threshold for ACFI, 51 candidate thresholds were generated within the range of [0,1] at 0.02 intervals. The extraction accuracy of bare coal regions and their spatial distribution patterns was assessed for each threshold (Figure 4). The results indicate that when ACFI = 0.2, the accuracy of exposed coal extraction reaches its peak. Consequently, the final threshold for OCFI was set to 0.2.

3.3. Deriving Inter-Annual Distance Measurements Using Rays Method

The potential open-pit mining areas identified in the previous step still include disturbances such as temporary coal piles, coal storage areas from shaft mines, and industrial coal heaps. To eliminate these, this study applied the Rays method to obtain inter-annual distance lists for each direction of each bare coal patch. By analyzing the annual expansion of these patches, we were able to delineate the true open-pit mining areas, while also determining the mining direction and intensity. The basic steps for implementing the Rays method for a single bare coal patch are as follows: (1) select a ray origin and establish a 0° reference line in the due east direction; (2) choose a specific spacing angle and project rays sequentially in a counterclockwise direction; (3) record the distance from the ray origin to the intersection points where the rays meet the boundary of the bare coal area for each year of the study period.

(1)

Selection of the ray origin

When using the Rays method to extract boundary distance variations in each patch in a potential open-pit mining area, the choice of ray origin significantly impacts the results. Thus, selecting an appropriate ray origin is crucial. Due to the study area’s long mining history and the lack of relevant data on past coal mining activities, it is impossible to determine the initial mining position for each block. Thus, two key principles were established for selecting ray origins for each bare coal patch in the study: first, the ray origins were chosen within the small bare coal patches that were initially generated in potential open-pit mining areas; second, the selected ray origins had to offer the best fit to the boundaries of the bare coal when constructing the ray network. The fitting accuracy was assessed for each ray origin using the coefficient of determination ($R^2$), calculated as follows:

$$R^2={\displaystyle \frac{SSR}{SST}},$$

(3)

where SSR represents the sum of the squared differences between the fitted data and the mean of the original data, while SST denotes the sum of the squared differences between the true data and its mean. The value of $R^2$ ranges from 0 to 1, with a value closer to 1 indicating a higher degree of fit. SSR and SST can be computed using the following formulas:

$$SSR={\sum }_{i=1}^n{(\widehat{y_{\mathrm{i}}}-\overline{y})}^2,$$

(4)

$$SST={\sum }_{i=1}^n{(y_{\mathrm{i}}-\overline{y})}^2,$$

(5)

where $\widehat{y_{\mathrm{i}}}$ is the fitted distance from the bare coal boundary to the ray origin, $\overline{y}$ is the mean of the actual distances, and $y_{\mathrm{i}}$ is the actual distance from the bare coal boundary to the ray origin.

(2)

Determination of the optimal ray spacing angle

When using the Rays method to monitor the bare coal areas, the more complex the shape of the boundary, the smaller the ray spacing angle should be. However, if the chosen interval angle is too small, it will result in an exponential increase in data volume, significantly impacting the computation speed. Therefore, it is crucial to select the ray interval angle reasonably to balance accuracy and efficiency. An analysis of the actual boundary of bare coal patch revealed that the boundary between adjacent rays could be categorized into three forms, as illustrated in Figure 5. In scenario (a), the bare coal boundary was a straight-line segment; regardless of the number of densification operations, the cumulative difference in distances between two rays remained constant. In scenarios (b) and (c), the cumulative distance difference increased after ray densification, with the increase in (b) being significantly smaller than that in (c). This occurred because the densified rays were able to capture trend variations in the boundary. For example, in scenario (c), fitting the boundary with a straight line without densification failed to detect the trend change at point A. After densification, the actual boundary was better approximated, and the fitted boundary exhibited more pronounced trend changes, resulting in a larger cumulative distance difference between adjacent rays, as shown in (d). Thus, a greater increase in the sum of these differences during ray encryption indicated that the process was more effective in monitoring boundary changes in the bare coal zone, making the encryption operation more essential.

(3)

Constructing the ray network

Following the identification of the ray origin for each bare coal patch and the determination of the angular spacing between neighboring rays, the rays were projected outward toward the corresponding bare coal boundaries in a 360° sweep around the origin. This process continued until the rays completed a full circle and returned to the 0° position, thereby forming a comprehensive ray network that covered all bare coal patches. The ray distance, defined as the length from the ray origin to the boundary of the mining area, calculated in GEE from the coordinates of the two points, and was measured annually for each directional angle. The change in ray distance between consecutive years was calculated by subtracting the ray distance of the subsequent year from that of the preceding year. This difference was used to assess dynamics of change in the mining boundary and characterize the progression of mining activities.

3.4. Monitoring Open-Pit Coal Mining Dynamics

This study used the Rays method to obtain distance change data in all directions for potential open-pit mining areas. Based on these data, the expansion factor and the expansion duration were comprehensively analyzed to determine the open-pit mining area and its mining dynamics. The expansion factor was used to analyze the rate of expansion and was calculated based on the change in distances over the last two years in the potential open-pit mining area. The calculation was performed as follows:

$$k={\displaystyle \frac{d_n-d_l}{d_l}},$$

(6)

where $d_n$ represents the distance from the ray origin to the bare coal boundary in the second year, and $d_l$ is the distance from the ray origin to the bare coal boundary in the first year. The variable $k$ denotes the expansion factor, with larger values of $k$ indicating a faster rate of increase in ray distance in that direction, which corresponds to more rapid mining activity. In order to determine the segmentation threshold for the expansion coefficient that distinguishes mining from non-mining areas, this study first analyzed the actual mining conditions. Theoretically, the expansion coefficient for non-open-pit mining areas should be zero. However, due to monitoring errors in bare coal identification and biases inherent in the distance calculation process, the measured values may deviate either positively or negatively. To establish the optimal segmentation threshold, the expansion coefficients of all bare coal patches were arranged in ascending order and their mutation points identified through data analysis. Figure 6 indicates that a significant change occurs when the expansion coefficient reaches 0.0136; considering potential error factors, the segmentation threshold was ultimately set at 0.01.

However, the expansion factor for a single year alone is insufficient to confirm that an area is an open-pit mining site. Thus, a three-year time window for expansion was adopted, enabling an interannual analysis of the expansion factor to determine the mining status of coal mines by assessing the stability of each patch over three years. The open-pit mining area and its direction was extracted using the following rule: if the expansion factor exceeded 0.01 in the last three years (e.g., 2022–2024), the area covered by the ray was classified as an open-pit mining area, with the direction of mining aligned with the angle of the ray and labeled as “expansion”. The year when bare coal was first observed in this area was considered the start of mining activities (Figure 7a). Conversely, if the average expansion factor over the last three years was lower than 0.01, a 3-year sliding window was applied to traverse the entire observation period in descending order of years. If the situation depicted in the pink area of Figure 7b occurred—where the expansion factor was lower than 0.01 during the first three years, but higher than 0.01 in the subsequent three years—the area covered by the ray was still classified as an open-pit mining area. The first year with an expansion factor below 0.01 marked the cessation of mining, and the ray was labeled as “stable”. In the second scenario, as illustrated in Figure 7c, if the average expansion factor remained below 0.01 for any consecutive three-year period within the entire observation timeframe, it indicated that the area covered by the ray represented a temporary coal pile.

3.5. Data Post-Processing and Validation

First, 50-pixel samples from open-pit coal mining areas and 50 from non-open-pit coal mining areas were randomly selected on Google Earth. The overall accuracy was then calculated using a confusion matrix to verify the accuracy of open-pit coal mining location detection. To verify the accuracy of the open-pit mining years, we selected 50 sample points per year from 2002 to 2024 (Identification of mining activities began in 2002) using stratified random sampling. Time-series imagery from Google Earth was used to interactively and visually calibrate the mining years for 1150 sample points. Using the sample labels and algorithm-generated results, we constructed an error confusion matrix to calculate overall accuracy (OA), producer accuracy (PA), user accuracy (UA), F1 score (F1 = 2 × PA × UA/(PA + UA)), and Kappa coefficient, assessing the detection accuracy of mining and reclamation events. These metrics provide an effective and straightforward approach for evaluating classifier performance.

Additionally, this study superimposed the identified mining direction results with the outputs of the center of gravity migration model to assess the method’s ability to detect mining direction. The center of gravity migration model is computed as follows:

$$\left(X,Y\right)=\left\{\begin{array}{c}{X}_t={\displaystyle \frac{{\sum }_{i=1}^n{Z}_i{x}_i}{{\sum }_{i=1}^n{Z}_i}}\\ Y_t={\displaystyle \frac{{\sum }_{i=1}^n{Z}_i{y}_i}{{\sum }_{i=1}^n{Z}_i}}\end{array}\right.,$$

(7)

where $n$ denotes the total number of pixels in year $t$; $Z_i$ represents the area of the $i$-th in year $t$; $x_i$ and $y_i$ are the longitude and latitude coordinates of the geometric center of the $i$-th in year $t$, respectively; $X_t$ and $Y_t$ represent the longitude and latitude coordinates of the center of gravity in year $t$, respectively. It is important to note that the center of gravity migration model only captures shifts in the annual center of gravity of bare coal areas and, therefore, cannot fully represent the mining direction. As such, this validation serves solely as an auxiliary assessment of the mining direction results, and is not intended to verify the accuracy of the mining direction itself.

4. Results

4.1. Identification of Potential Open-Pit Mining Areas

This study utilized ACFI with Landsat time-series imagery spanning 2000 to 2024 to detect potential open-pit coal mining areas in the Chenbarhu Banner coalfield. As shown in Figure 8, a total of 22 bare coal patches were detected. While the ACFI demonstrated high effectiveness in identifying bare coal areas, the results also included numerous patches unrelated to open-pit coal mining. To achieve accurate identification of open-pit mining areas, further integration with the Rays method is required.

4.2. Ray Network Analysis Results

(1)

Ray origin selection

In this study, 22 potential open-pit mining patches from the Chenbarhu Banner coalfield were analyzed for their suitability as ray origins based on the established selection criteria. As illustrated in Figure 9, all but the fifth bare coal patch exhibit R² exceeding 0.7. Notably, for larger bare coal patches, the R² values consistently surpass 0.8, indicating a high level of reliability in ray origin selection for these areas. Additionally, the area of each independent bare coal patch was calculated individually, with the area serving as the weighting factor for the corresponding patches, obtaining a weighted R² of 0.8422 for the study area. This demonstrates a high goodness of fit in ray origin selection across most bare coal patches, offering a robust foundation for accurately identifying the mining direction and operational status of the open-pit mining areas.

(2)

Neighboring ray spacing angle selection

In this study, five large bare coal areas were selected, and rays were constructed at 1° intervals from the year of initial identification to 2024, generating a total of 30,515 distance difference datasets. These datasets were reordered from smallest to largest, forming a distance difference curve between neighboring rays, as shown in Figure 10. The analysis revealed a significant change in the curve when the distance difference exceeded 210 m, with 99.15% of all distance differences falling below this threshold. Therefore, distances greater than 210 m were identified as anomalies, indicating the need for further ray refinement.

Subsequently, the sum of the distance differences and the number of anomalies between neighboring rays at intervals of 0.25°, 0.5°, 2°, and 4° were calculated. Groupings were formed as follows: 4° and 2° for the first group, 2° and 1° for the second, and so on, forming a total of four groups. The angle corresponding to a significant increase in both the sum of the distance differences and the number of excluded anomalies was then selected as the optimal spacing angle for ray construction. The ray network was constructed for each of the five selected angles, and the sum of the distance differences between all neighboring rays, as well as the number of instances where the distance difference exceeded 210m, were recorded. Based on the results shown in Figure 11, it is evident that when the interval angle between neighboring rays is reduced from 2° to 1°, the sum of the distance differences between rays increases the most, and the greatest number of outliers are eliminated. Therefore, 1° was selected as the optimal interval angle for neighboring rays.

(3)

Ray network construction

Based on our previous analyses, we emitted rays in a full 360° pattern from a defined ray origin using an appropriate ray spacing angle and calculated the corresponding distances. This approach allowed us to compile a comprehensive list of distances for all potential open-pit mining areas from the initial formation of the bare coal patches through to 2024. Subsequently, by integrating the computation of expansion coefficients with established open-pit coal mine determination rules, we achieved precise localization of the open-pit coal mines and extracted their operational statuses.

Figure 12 illustrates the ray network constructed at the Dongming mine, where rays were emitted at 1° intervals over the period from 2018 to 2024. In the figure, the bare coal boundaries for 2018, 2020, 2022, and 2024 are depicted as black, red, green, and blue curves, respectively, while the corresponding-colored straight lines represent the expansion distances in each direction. Each ray records seven distance measurements, which provide a quantitative description of the open-pit mining area’s evolution at discrete time points over the study period. As mining activities progress, the expansion of the open-pit mining area is reflected by increased ray distances. Analysis of these measurements explicitly reveals the spatiotemporal evolution of the bare coal patches, thereby offering a robust basis for determining subsequent mining directions and operational statuses.

4.3. Identification of Open-Pit Coal Mines and Operational Status

Using the ACFI, 22 potential open-pit mining patches were identified within the study area. When combined with the Rays method, the results indicated that three of these patches were open-pit mining areas, while the remaining patches were coal storage facilities or temporary coal piles (Figure 13). The proposed method achieved an overall accuracy of 92% and the F1 score of 0.91 for the identification of open-pit coal mines, with a user accuracy of 100%, a producer accuracy of 84%, and a Kappa coefficient of 0.84. Furthermore, the temporal accuracy of identifying mining activities from 2000 to 2024 was validated, yielding an overall accuracy of 94% and a Kappa coefficient of 0.86 (Figure 14).

These findings demonstrate that the ACFI-Rays method combination effectively distinguishes open-pit mining areas from factory storage sites, temporary coal piles, and areas misclassified due to spectral confusion. Furthermore, during the construction of the rays, each ray was assigned an attribute of either “stable” or “expanding” based on the inter-annual variation in the expansion coefficient, enabling the assessment of the mining status of the patch. As illustrated in Figure 13, among the three identified open-pit coal mines, (a) and (b) represent mines that were still being actively mined, while (e) represents a mine where mining had ceased. For the latter, the last mining year of the identified patch was determined to be 2022, consistent with the actual cessation time of mining activities. It is worth noting that non-mining rays could still be present within coal mines, either due to a low annual mining progress in that direction, or because it was the first recorded instance of expansion in that direction.

4.4. Detection of Open-Pit Coal Mining Direction

Mining direction identification relies on rays with expansion properties, each encoded with corresponding angular information. The mining direction identification results were validated using the center of gravity migration model. As shown in Figure 15, the Rays method results were directionally consistent with those from the center of gravity migration model, confirming the reliability of the ray method’s extraction. Compared to the center of gravity migration model, which provides only approximate information on mining direction, the Rays method offers superior spatial information processing capabilities, enabling comprehensive monitoring of mining dynamics in all directions of an open-pit coal mine. Additionally, the ray distribution can be selectively refined in specific areas based on research requirements, allowing for more detailed and targeted monitoring and analysis. Based on the identification results, the current mining direction of the Baorixile mine is predominantly between 4° and 13°, while the Dongming mine’s mining activity is concentrated between 69° and 141°. Notably, no new mining activity was detected between 127° and 134°, which aligns well with the spatial extent of new open-pit mining areas observed in Google Earth imagery from 2024, further verifying the accuracy of the ray method in extracting open-pit coal mining information. Moreover, since the information contained within each ray is independent and unaffected by adjacent rays, the method maintains high recognition accuracy and strong resistance to interference, even in densely distributed open-pit coal mines where ray intersections are possible. These findings demonstrate the stability and applicability of the ray method in complex mining environments, providing reliable technical support for the dynamic monitoring of open-pit coal mining activities.

5. Discussion

5.1. Evaluation of Coal Production in Relation to Open-Pit Mining Area

Coal production, as a key indicator of production efficiency and the extent of resource extraction in coal mines, plays a critical role in the monitoring of open-pit mining operations. High production levels are often associated with widespread land disturbance, resulting in significant ecological impacts. Thus, evaluating coal production is essential for maintaining a balance between resource development and environmental protection. In this study, the ray method was employed to delineate open-pit mining areas actively mined in the current year and to calculate their respective areas. From 2002 to 2024, the total open-pit mining areas of Dongming coal mine, Shunxing coal mine, and Baorixile coal mine were 399.419 ha, 131.518 ha, and 1585.194 ha, respectively. Notably, this study only obtained coal production data for the study area for the period 2018–2023. Consequently, the analysis focuses on the relationship between the mined area and coal production during this period. Between 2018 and 2023, the mined area exhibited a positive correlation with coal production (Figure 16), with the fitting degree R² being 0.863, Pearson correlation coefficient of 0.929, and the significance-test P-value of less than 0.01. The results of the study demonstrated a strong alignment between the expansion trend of the mining area and the growth trajectory of coal production during the study period. This consistency further validated the accuracy and reliability of the proposed method for monitoring open-pit coal mining. However, while the coal production of Dongming coal mine and Baorixile coal mine reached their highest levels in 2023, their corresponding mined areas were not at their maximum. This phenomenon may be attributed to the fact that coal production in open-pit mining is influenced not only by the extent of the mined area but also by the mining depth. Variations in the spatial distribution of coal seams result in differences in mining depth across locations, leading to a mismatch between the mined area and coal production.

5.2. Applicability of Mining Dynamics Monitoring in Large-Scale Open Pit Coal Mines

This study proposes an automated method for large-scale monitoring of open-pit coal mining dynamics. Notably, the method is capable of accurately identifying the distribution of mining areas and efficiently tracking the mining dynamics of open-pit coal mines, even in regions where mining location information is unavailable. All data processing and calculations were conducted on the Google Earth Engine (GEE) platform. GEE significantly enhances the efficiency of remote sensing data acquisition and processing by integrating petabytes of satellite pictures and geospatial information catalogs with line-star analysis capabilities [43]. In addition, GEE offers comprehensive preprocessing support that surpasses the capabilities of standard remote sensing image processing tools. In conclusion, this study is grounded in a rigorous logical analytical framework, demonstrates the adaptability and effectiveness of combining the ACFI with the Rays method for dynamic monitoring of open-pit coal mining.

In addition, due to the ideal natural conditions in the study area selected for this paper, the spectral differentiation between bare coal and other features is highly pronounced. To further validate the effectiveness of the proposed method, we also monitored the mining conditions in the Yimin and Shengli open-pit coal mine areas. Notably, the Shengli mine area is constrained by suboptimal local natural conditions, with the region predominantly characterized by bare soil and vegetation. The obtained monitoring results are shown in Figure 17. Based on the monitoring results, we calculated the identification accuracy of open-pit coal mines for two case study areas. In the Yimin mine area, the overall accuracy was 90%, with a user’s accuracy of 100%, a producer’s accuracy of 80%, and a Kappa coefficient of 0.80. In the Shengli mine area, the overall accuracy was 89%, with a user’s accuracy of 100%, a producer’s accuracy of 78%, and a Kappa coefficient of 0.78. These results indicate that, despite significant differences in natural conditions, the proposed method can accurately identify open-pit coal mines, thereby confirming its general applicability. However, it is noteworthy that significant discrepancies between the producer’s and the user’s accuracy were observed across the three monitored regions. A detailed analysis of the satellite imagery indicated that this disparity is attributable to the fact that the delineated potential open-pit mining areas capture only the exposed coal portion while excluding the mining platform (Figure 13 and Figure 17 illustrate this phenomenon, with the gray and white area within the red circle representing the mining platform). For newly developed coal mines, the bare coal area is typically small relative to the larger mining platform, which impacts the mining area’s recognition accuracy. However, this effect only slightly reduces the recognized extent of the mining area compared to the actual extent, and does not impact on the determination of the location of the open-pit mining area. As mining progresses, the exposed coal area will expand to cover the platform, and with increased mining time, the impact on recognition accuracy will gradually diminish. Additionally, Figure 17 shows that a portion of the bare coal region is missing in R2. This discrepancy arises because the spectral characteristics of the omitted area differ from those of the majority of the bare coal, leading to inaccurate recognition and reduced overall accuracy [26]. This observation indicates that the current bare coal identification method requires further refinement, potentially by integrating additional information from other coal types.

In summary, the identification results for open-pit coal mines across different environments indicate that, despite some missing mining information, the method remains effective in locating open-pit coal mines under a wide range of conditions without requiring a priori information. Using this approach, future studies will be able to rapidly and accurately identify the global distribution of open-pit coal mines and create a detailed geodatabase. We believe the method has significant potential for scalability and optimization, making it applicable not only to open-pit coal mining monitoring but also to information extraction for other surface cover types, as well as the detection of stability or extensibility across multiple targets.

5.3. Uncertainty Analyses and Future Research Perspectives

While the proposed method demonstrates significant advantages in the long-term monitoring of open-pit mining disturbances, certain limitations may affect its applicability in specific contexts. First, identifying the initial mining period presents a challenge, as the exposure of bare coal typically occurs after a delay following the stripping of the topsoil. Since this study relies on the presence of bare coal for detection, the initial stages of mining activity may not be immediately identified, leading to a temporal lag in the assessment. Second, when using the Rays method to assess the status of open-pit mining, the approach relies on the intersection of rays with bare coal boundaries to determine whether a coal patch is expanding or stable. As a result, changes occurring within existing bare coal patches may not be accurately captured, potentially omitting some mining information. Finally, the ACFI threshold can be dynamically adjusted based on the study area, alternative approaches such as Otsu’s method also offer robust solutions for large-scale studies.

In light of the inherent limitations of Landsat satellite data (extended revisit intervals and the absence of vertical dimensional information), future research should focus on developing a collaborative monitoring framework that integrates an unmanned aerial vehicle (UAV). The integration of an UAV offers several advantages: an UAV can conduct dynamic monitoring during intervals between satellite passes, effectively capturing sudden mining events and providing more immediate data; utilizing techniques like tilt photogrammetry, an UAV can generate high-precision three-dimensional point cloud data of mining areas, facilitating accurate quantification of volumetric changes. Specific methodologies to achieve this integration include: (i) employing the methods proposed in this study to accurately identify the locations and mining dynamic information of open-pit coal mines on a broad scale; (ii) establishing an emergency response mechanism for drones, so that when satellite monitoring detects signs of abnormal mining activity, drones are immediately deployed for targeted verification. By constructing a multi-scale monitoring system that integrates both macro- and micro-level observations, more accurate data support can be provided for the supervision and verification of open-pit coal mines.

6. Conclusions

Identifying the distribution and extraction extent of numerous open-pit coal mines across large scales is a challenging task, particularly in the absence of location information. This study presents a novel approach for automatically locating open-pit coal mines and monitoring their operational status under conditions of unknown mining information. The method was successfully applied to the Chenbarhu Banner coalfield in Inner Mongolia, with the following key conclusions:

(i) The proposed approach enabled rapid and accurate identification of open-pit mining areas without the need for prior knowledge of mining locations. (ii) The approach successfully distinguished between active and closed mines, while accurately identifying key temporal information. (iii) The study accurately identified both the direction and extent of mining for open-pit coal mines. Furthermore, a significant correlation between the annual mining area and coal production (r = 0.929, p < 0.01) was established, confirming the reliability of the proposed approach. Finally, the generalizability of the method was validated based on the identification results for open-pit coal mines across diverse environmental settings.

In conclusion, the proposed approach offers a powerful tool for large-scale, accurate monitoring of open-pit coal mining dynamics, even in the absence of prior location information.