2.1. Study Area
The Antaibao Open-pit Coal Mine is located in Shuozhou city, Shanxi Province, 112°10′58″~113°30′ E, 39°23′~39°37′ N [
19], in the east of the Loess Plateau. The climate is characterized as a continental monsoon climate. The landform type is a loess hill with an elevation that ranges from approximately 1300 to 1400 m. Seasons are distinct, characterized by lower rainfall and snow in the spring, concentrated precipitation in the summer, less rain in the autumn, and lower snowfall in the winter. The annual average temperature ranges between 5.4 and 13.8 °C, and average annual precipitation ranges between 428 and 449 mm.
The vegetation cover types of the four dumps are distinct. The vegetation coverage within the South Dump (SD) is dominated by tree species. The vegetation here has formed as tall tree bodies in a long-term stable environment, with roots that have an independent trunk and the trunk is distinct from the canopy. According to the actual site investigation, the main tree species is black locust (
Robinia pseudoacacia Linn.), and also contains a small amount of sea buckthorn (
Hippophae rhamnoides Linn.), caragana (
Caragana Korshinskii Kom), and poplar (
Populus L.). The West Dump (WD) is dominated by shrubs and contains a high density of sea buckthorn. Types of grasses include alfalfa (
Medicago sativa L.) and erect milkvetch (
Astragalus adsurgens pall) within the West Expansion Dump (WED). The Inner Dump (ID) is dominated by a mix of grasses, shrubs, and trees, covered by sea buckthorn, narrow-leaved oleaster (
Elaeagnus angustifolia Linn.), elm (
Ulmus pumila L.) and caragana. The differences between the four dumps are mainly reflected in the differences between vegetation cover. At the same time, invasions of different species may exist within a small scope. Based on this, the following four study areas were determined (
Figure 1).
Currently, Antaibao is one of the largest open-pit coal mines in China. Its mining began in 1985, and during the process of mining, the boundary of the mining area has been continuously adjusted. For more than 30 years, a relatively mature “integration of mining, transportation, drainage and rehabilitation” was carried out [
20]. In this study, four dumps in the Antaibao were selected as the study objects, namely, the South Dump, the West Dump, the West Expansion Dump, and the Inner Dump. Each dump has undergone a relatively long period of reclamation work, and currently forms a relatively stable ecosystem. The relevant information of each dump is shown in
Table 1 [
21,
22,
23,
24].
Due to dump occupation during mining, many environmental problems, such as terrain destruction, vegetation damage, and reduction in biodiversity, have continuously taken place, causing the dump to become a key site in land reclamation and ecological restoration [
25]. In the Antaibao Open-pit Coal Mine, the integrated process of “mining-transport-dumping-reclamation” from west to east was adopted. Due to variable regional topography, dumps adopted different reclamation methods, leading to differences in the ecological restoration speed. Although the South Dump was first abandoned, the spontaneous combustion of coal gangue within the dump not only burned the vegetation, but also directly affected the soil moisture content, thus hindering the growth of surrounding vegetation. Later, the South Dump was reclaimed repeatedly, but the recovery speed was still slow. The West Dump was reclaimed in 1994 and 1996, with increasing vegetation coverage as the reclamation goal, but the recovery speed was also slow. In 2010, due to the industrial adjustment, the area occupied by the interference increased to 7.12% [
23]. In the early formation of the West Expansion Dump, it was covered by natural forest, and later part of it was developed as an open-pit mining dump. Due to the short period of reclamation, there are still major disturbances and the ecosystem has not yet reached a stable state. The Inner Dump was an open pit in the early stage, and it was discharged after the pit is closed. Its recovery speed after reclamation was better than that of the other dumps.
2.2. Data Collection
In order to ensure the consistency of the vegetation spectrum information as much as possible, and in reference to existing research, August was selected for collecting images due to the excellent condition of the vegetation growth at this time [
26]. Compared with other remote sensing satellites (such as ZY satellite and GF satellite), Landsat 8, Sentinel-2, and HJ satellites have earlier launch times and longer estimated service lives, and span a larger monitoring time. These are more suitable for long-term mining monitoring. The Landsat 8 and Sentinel-2 data used in the study were downloaded from the U.S. Geological Survey website (
https://earthexplorer.usgs.gov/, accessed on 14 July 2021), and the HJ1A data were downloaded from the China Resources Satellite Application Center website (
http://www.cresda.com/CN/, accessed on 22 July 2021). The relevant information of the obtained data is as follows (
Table 2).
The Spectral Response Function refers to the ratio of the received radiance to the incident radiance at each wavelength of the sensor. Due to the limitation of sensor hardware, the spectral response functions of different sensors are quite different, which has a very direct impact on the reflectivity of vegetation [
27]. In order to reflect the differences in sensors more clearly, the sensor-specific spectral response function is shown in
Figure 2.
Landsat satellites are jointly managed by USGS and NASA, and have the longest monitoring time. Since 1972, nine satellites have been launched, but Landsat 6 failed to launch, Landsat 7 was lost in 2012, and Landsat 8 was launched in 2013, equipped with OLI sensors. The 30 m visible light waveband, 15 m panchromatic waveband, and 30 m shortwave infrared waveband of Landsat enabled realization of long-term ground monitoring [
28]. The Sentinel-2 optical satellite was launched in 2015, and managed by ESA (European Space Agency, Paris, France). It carried a high-resolution and multi-spectral image device to obtain 10 m visible light waveband, 20 m infrared waveband, 20 m red edge waveband, and 20 m shortwave infrared waveband information. The satellite is mainly used to monitor the growth of vegetation and land use cover. HJ (Environmental Disaster Satellite) was launched in 2008, and specially developed by China for detecting natural disasters. It is widely used in disaster monitoring and ecological assessment [
29].
2.3. Method
In order to compare the correlation between sensor monitoring results under different types of vegetation cover (
Table 1), NDVI monitored by HJ, Sentinel-2, and Landsat 8 were obtained separately. Based on the NDVI results, the correlations under different vegetation cover were obtained by unary linear regression analysis.
2.3.1. Preprocessing
In order to reduce the influence of sensor parameters, the image data from three satellites were preprocessed, including radiation calibration and atmospheric correction. After decompressing the Landsat 8 data, ENVI5.3 software was used for radiometric calibration to convert the DN value into a radiance value, and for FLAASH atmospheric correction to eliminate the influence of atmospheric factors and convert surface reflectance to land surface reflectance. Using Sen2Cor 02.09.00 software, Sentinel-2 data was subjected to radiation calibration and atmospheric correction. Sen2Cor software is a tool plug-in formatted for Sentinel-2 Level 2A products produced by ESA, which performs radiation calibration and atmospheric correction on L1C data. After atmospheric correction, SNAP software was used to convert the data format and resample images to 30 m spatial resolution to ensure the spatial resolution was consistent with that of other data. The original HJ data were subjected to radiometric calibration and band composition using patch tools, and the data after radiometric calibration were subjected to format conversion, geometric correction, and FLAASH atmospheric correction using ENVI5.3 software.
2.3.2. NDVI
The rapid development of remote sensing technology has enabled many vegetation indices to be obtained by formula calculations based on image band data, which has promoted the wide application of various vegetation indices in vegetation evaluation and monitoring [
30]. The Normalized Difference Vegetation Index, reflecting characteristics of plant growth, vegetation coverage, and biomass, is widely used as an indicator to monitor vegetation growth. It has been used by many scholars both locally and globally for a long time to study the vegetation condition [
31]. In this study, NDVI was selected as an indicator for vegetation monitoring to reflect the growth status of vegetation. Relevant studies have shown that this index can eliminate the influence of other factors (terrain, shadow, and atmosphere) to a certain extent.
As the most commonly used vegetation index, NDVI is defined as the ratio of the difference of the near-infrared and infrared bands to the sum of the near-infrared and infrared bands [
32], and the threshold is [−1,1].
where
NDVI is the normalized difference vegetation index,
NIR is the near-infrared band, and
R is the visible red band. Generally speaking, when −1 <
NDVI < 0, the land cover type is cloud, water, snow, etc., which are without vegetation cover; when
NDVI = 0, the ground cover is rock or bare soil; when 0 <
NDVI < 1, the land is usually covered by vegetation. The larger the
NDVI value, the better the vegetation condition. Based on the preprocessed remote sensing image,
NDVI is calculated by the above formula.
2.3.3. Unary Linear Regression Analysis
Currently, regression analysis is a widely used predictive analysis method that can discover the association among variables through time series models, providing the correlation coefficient and linear regression equation [
33,
34]. Regression analysis originated from basic statistical concepts; based on the statistical results, related mathematical methods are used to obtain the basic assumptions, statistical inferences, and regression diagnosis. This is very useful for practical research, including dummy variables, interaction, auxiliary regression, polynomial regression, spline function regression, and step function regression.
Origin software has an effective data analysis function and is often used for analysis of statistical variables in scientific research and practical applications [
35]. When calculating NDVI in this study, it was found that, due to the influence of design parameters and solar altitude angles of sensors, NDVI calculated by different sensors has certain differences. Exploring the differences between NDVI results from monitoring using different sensors is of great significance to the mutual substitution and supplementation of sensor data. In this study, based on the results of NDVI, 1000 random pixel points were established within dumps for regression analysis.