Analysis of Vegetation Ecological Anomaly Response in the Xiangshan Uranium Mining Area Based on Multi-Source Remote Sensing Data Fusion

Abstract

The Xiangshan Uranium Mining Area in Jiangxi Province is a pivotal uranium extraction site crucial for China’s nuclear sector. However, vegetation ecology research in this region remains scarce, particularly studies grounded in multi-source remote sensing data. To overcome these challenges, this paper introduces a methodology that combines multi-source remote sensing data with the random forest machine learning algorithm to invert vegetation canopy structure parameters in the Xiangshan Uranium Mining Area. This approach is complemented by the integration of multiple vegetation indices for a comprehensive evaluation. To guarantee the dependability of the inversion results, this study employs Sentinel-1/2 imagery and ICESat-2 spaceborne LiDAR data, which furnish abundant optical information, terrain data, and vertical vegetation structure insights. The experimental findings reveal that the overall vegetation ecology in the Xiangshan Uranium Mining Area is in a satisfactory state, yet the low Radar Vegetation Index (RVI) hints at potential soil degradation concerns within the mining area. Furthermore, notable disparities in vegetation canopy structure between the mining area and the comparison zone underscore that the presence of mining deposits indeed exerts a potential influence on vegetation canopy structure. This study bridges the research gap and offers scientific support for mineral exploration and sustainable mining development.

1. Introduction

As a key mineral resource for China’s energy sector, uranium is fundamental to the development of the nuclear industry. Given the ongoing development and exploitation of uranium resources, it is imperative to precisely assess the influence of uranium mining regions on vegetation ecology. This assessment is crucial for maintaining ecological stability in mining zones and achieving sustainable mineral resource utilization.

The Xiangshan Uranium Mine, located in Fuzhou, Jiangxi Province, stands as China’s largest volcanic rock-type uranium mining deposit and a significant uranium extraction site in southern China. Vegetation in the mining area forms a critical ecosystem surrounding uranium deposits. Soil changes associated with mineralization may affect vegetation growth and canopy structure, leading to unique remote sensing signals. These signatures not only function as vital indicators for ecological monitoring of vegetation in mining areas but also lay a fundamental theoretical groundwork for remote sensing-based mineral prospecting.

At present, scholars across the globe have undertaken comprehensive research on vegetation ecological monitoring in a diverse array of mining regions, encompassing coal mines, metal mines, rare earth mines, and various other mineral categories. In the specific realm of uranium mining, Cai et al. [1] have summarized the distinctive features and metallogenic patterns of uranium resources in China, emphasizing the ongoing necessity for ecological preservation throughout the process of uranium mining development. Meanwhile, Jin et al. [2] utilized remote sensing technology to examine the spatiotemporal variations in vegetation coverage both prior to and following the commencement of operations at a uranium mine, revealing a notable decline in vegetation coverage during the mining development phase, followed by a gradual recovery over time.

In the realm of coal and metal mining, Zhao et al. [3] delved into the spatio-temporal evolution patterns of vegetation greenness in the Xilinhot open-pit coal mining region. Their research uncovered that mining activities led to a substantial decrease in vegetation coverage; however, vegetation greenness rebounded following the implementation of pertinent ecological restoration initiatives. Liu et al. [4] adeptly employed machine learning algorithms to categorize the vegetation encircling the Anshan iron mining area. Furthermore, in their investigation on vegetation coverage monitoring and spatio-temporal change analysis in the vicinity of the Anshan iron mining site [5], they quantitatively evaluated the vegetation coverage around the mining area by integrating remote sensing imagery with ground-measured data and GIS technology, thereby furnishing robust data support for ecological restoration endeavors in the mining region. Wei et al. [6] conducted a time-series analysis of the NDVI and vegetation coverage indices for the spoil ground of the Haizhou open-pit coal mine in China, utilizing remote sensing technology. Their findings demonstrated that these two indices could effectively serve as indicators for assessing vegetation restoration conditions in mining areas.

However, most existing studies rely on time-series remote sensing, focusing on vegetation changes before/after mining and restoration effectiveness. Few studies integrate multi-source data or systematically compare canopy structure between mining and non-mining areas, limiting our understanding of how uranium deposits affect vegetation. Furthermore, systematic investigations leveraging multi-source remote sensing data are still relatively scarce. Additionally, there is a dearth of studies that systematically compare the disparities in forest stand structure of surface vegetation between undeveloped mining regions and non-mining areas. This gap hinders our ability to fully elucidate the potential impact of mining deposits themselves on vegetation ecology.

In addition, diverse vegetation indices are widely used for vegetation ecological monitoring in mining areas. However, Single vegetation indices are easily affected by natural factors such as location and climate. Hui et al. [7] compared the accuracy of NDVI for monitoring vegetation coverage in open-pit coal mines and confirmed that NDVI shows severe errors in coal-covered areas inside mines, which proves the inherent limitations of single vegetation indices for precise vegetation ecological monitoring. Taking Central Asian grasslands as the research object, Muminov et al. [8] explored the effects of land use changes on grassland health under three typical human disturbances, including streams, roads, and mining areas, and further proved that a single NDVI cannot effectively distinguish vegetation response discrepancies caused by natural and anthropogenic disturbances. To overcome the weakness of a single NDVI in identifying vegetation ecological anomalies, Hede et al. [9] proposed the Vegetation Index for Physiological Status (VIGS) derived from reflectance spectra of Landsat ETM+ imagery. This method expands the application scope of remote sensing mineral prospecting from traditional arid and semi-arid regions to densely vegetated areas.

Current studies concerning the Xiangshan uranium mining field have formed a relatively complete technical system covering mining deposit identification and ecological monitoring. Zhuo et al. [10] carried out mineral prospecting practices by integrating geological, geophysical, geochemical, and remote sensing methods in the Luopo area of Xiangshan uranium mining field, and confirmed that integrated remote sensing approaches can greatly improve prospecting efficiency. Zhong et al. [11] established a comprehensive information prospecting model for the study area, providing theoretical and practical references for uranium exploration in the adjacent Yuhua Mountain area. Xu et al. [12] extracted remote sensing mineralization-alteration anomalies in vegetation-covered areas of the Zhuguang Mountain region and conducted prospecting prediction via remote sensing techniques. The results showed that uranium deposits are highly consistent with mineralization-alteration anomaly information, and favorable prospecting areas feature well-developed surface alteration with great potential for further uranium exploration. Zhu et al. [13] employed InSAR satellite technology to monitor surface deformation across the Xiangshan uranium mining field, verifying that satellite remote sensing is capable of large-scale and long-term surface deformation monitoring, which compensates for the spatial limitations of traditional ground surveys. Despite remarkable progress achieved with advanced remote sensing technologies, the existing prospecting and monitoring system for the Xiangshan uranium mines still faces challenges, including poor regional adaptability of prospecting models and accuracy constraints of vegetation characteristic parameters.

To address the aforementioned constraints, certain scholars have put forward monitoring approaches that integrate multi-source remote sensing data with computational algorithms. Li [14] conducted a thorough exploration into the utilization of multi-source remote sensing data for constructing a framework aimed at vegetation monitoring in mining regions, thereby establishing a solid foundation for subsequent research predicated on multi-source data fusion. Li et al. [15] amalgamated satellite remote sensing, unmanned aerial vehicle (UAV) remote sensing, and information technology, complemented by geological surveys, geophysical exploration, and drilling activities, to devise an ecological environment restoration and monitoring model specifically tailored to the Jugeng mining area, yielding favorable application results. Shi et al. [16] extracted hyperspectral remote sensing prospecting information based on vegetation spectral anomalies, confirming the existence of spectral disparities between analogous vegetation in mining and non-mining areas. These studies offer invaluable prior knowledge for harnessing machine learning algorithms to probe into the impact of mining activities on vegetation ecology, and present novel viewpoints on leveraging the detection of vegetation ecological anomalies as indicator factors for remote sensing prospecting.

Although numerous studies have verified the great potential and advantages of multi-source remote sensing data fusion in vegetation ecological monitoring across various regions, this method has not yet been systematically applied to the Xiangshan Uranium Mining Area in Jiangxi Province. Current research fails to effectively reveal the ecological differences in vegetation between this uranium mining area and ordinary vegetation zones, which further restricts the scientific and efficient development of mineral exploration. Against this background, this study takes the Xiangshan Uranium Mining Area as the core research area and focuses on exploring the potential impacts of uranium deposits on local vegetation growth and canopy structure. By fusing multi-source remote sensing data and adopting the random forest machine learning algorithm, this study retrieves canopy structure parameters and carries out a comprehensive analysis combined with optical vegetation indices. The research results can provide a scientific basis for ecological monitoring, vegetation restoration, and green development in Xiangshan Uranium Mining Area.

2. Methodology

The technical workflow employed in this study is illustrated in Figure 1. Initially, diverse data types undergo preprocessing to yield valid datasets within the designated target area. Subsequently, RVI is computed from Sentinel-1 data, while slope and aspect are extracted from DEM data. Following the extraction of band information from Sentinel-2 data, vegetation indices and texture features are calculated, and valid data from ICESat-2 are meticulously screened.

Subsequently, all processed results are integrated and screened, then imported into the inversion model as feature variables to invert canopy height and canopy openness in leaf-on and leaf-off periods, thereby eliminating the interference of seasonal factors.

Finally, vegetation indices and canopy structure parameters are extracted from both mining and non-mining areas. Through a combination of statistical comparison and spatial analysis, similarities and disparities in vegetation characteristics across different regions linked to mining deposit locations are discerned.

2.1. Data Preprocessing

To ensure the consistency and compatibility of data from different sources, all datasets were subjected to preprocessing, including cloud masking, radiometric calibration, band extraction, unified projection, and image clipping. Sentinel-1 and Sentinel-2 data were resampled to a uniform spatial resolution of 30 m. The Extract Multi Values to Points tool in ArcGIS10.8 was used to match sample points with feature variables to ensure consistent data spatial scales.

2.1.1. Sentinel-2 Data

Sentinel-2 Level-2A optical remote sensing images were downloaded from the European Space Agency (ESA) Copernicus Data Space Ecosystem. The Level-2A products have been geometrically and radiometrically corrected and resampled to a spatial resolution of 30 m for consistency with Sentinel-1 and DEM data.

2.1.2. Sentinel-1 Data

Two scenes of Sentinel-1 IW SLC data were procured from the ESA Copernicus Data Space Ecosystem. Data preprocessing was executed using SNAP 13.0.0 software, encompassing procedures such as orbit correction, radiometric calibration, band merging, multilooking, and terrain correction. Subsequently, the processed data were exported in ENVI5.6 format.

2.1.3. DEM Data

The DEM was precisely clipped using the boundary vector data of the study area to isolate data pertinent to the target region. Following this, terrain analysis tools within ArcGIS10.8 software were employed to generate terrain factor layers, including slope and aspect.

2.1.4. ICESat-2 LiDAR Data

Priority was given to selecting valid photon data with high quality, along with associated canopy height, canopy openness, and other pertinent information. The data were then clipped to align with the study area boundary. Subsequently, the data were classified into leaf-off period samples (January to March) and leaf-on period samples (August to October), so as to fuse characteristic variables for subsequent model training. Figure 2 illustrates the spatial distribution of high-quality, valid photon data points within the study area. As evident from Figure 2, the valid photon data points are uniformly dispersed across the study area with an appropriate density.

2.2. Vegetation Index Calculation

Vegetation indices serve as crucial metrics for monitoring vegetation growth status through remote sensing technology. These indices are generated by applying mathematical combination operations to spectral reflectance values obtained from various bands captured by sensors [17]. This approach effectively mitigates interference from external factors such as the atmosphere and soil. Given the specific vegetation types and data characteristics of the Xiangshan mining area, seven representative optical vegetation indices and microwave vegetation indices were carefully selected [18]. The calculation formulas for these indices are detailed in

Table 1. Calculation formulas for the vegetation indices used in this study.

Parameters	Calculation Formula	Brief Description
Normalized Difference Vegetation Index (NDVI)	$\begin{array}{l}(B8-B4)/\left(B8+B4\right)\hfill \end{array}$	It reflects changes in vegetation photosynthesis and leaf biomass, thereby providing reliable information on vegetation growth status [19].
Enhanced Vegetation Index (EVI)	$2.5\times (B8-B4)/(B8+6\times B4-7.5\times B2+1)$	It performs better than NDVI in capturing vegetation changes affected by topography [20] and is more suitable for areas with high vegetation cover.
Fractional Vegetation Cover (FVC)	$\begin{array}{l}(NDVI-NDVIsoil)/(NDVIveg-NDVIsoil)\hfill \end{array}$	Reflects vegetation coverage.
Red-edge Chlorophyll Index (CIRE)	$\left(B8/B5\right)-1$	The red-edge band is sensitive to vegetation chlorophyll content and can reflect the physiological state of vegetation [21].
Green-band Chlorophyll Index (CIG)	$\left(B8/B3\right)-1$	The combination of the green band and near-infrared can reflect vegetation chlorophyll concentration.
Normalized Red-edge Vegetation Index (NDRE)	$(B8\mathrm{a}-B5)/\left(B8\mathrm{a}+B5\right)$	The normalized index of the red-edge band can more accurately reflect the vegetation growth stage [22].
Radar Vegetation Index (RVI)	$4{\sigma }_{VH}^0/({\sigma }_{VV}^0+{\sigma }_{VH}^0)$	A vegetation index in the microwave band that reflects vegetation structure [23,24].

.

In the table, Bi (i = 2, 3, 4, 5, 6, 8, 8A) represents the blue, green, red, red edge 1, red edge 2, near-infrared 1, and near-infrared 2 bands of Sentinel-2, respectively. ${\sigma }_{VH}^0$ and ${\sigma }_{VV}^0$ denote the VH-polarized and VV-polarized backscattering coefficients of Sentinel-1, respectively.

2.3. Canopy Structure Parameter Inversion

2.3.1. Random Forest Regression Model

The Random Forest is an ensemble learning algorithm that comprises multiple decision trees and is employed for both classification and regression tasks. Renowned for its robust resistance to overfitting, effective handling of high-dimensional data, and high computational efficiency [25], the Random Forest algorithm has found widespread application in remote sensing feature inversion. It operates by randomly and independently extracting sub-sample sets and constructing decision trees independently for computation. During the construction of each decision tree, a random subset of features is selected at each node, from which the optimal feature is chosen for splitting. This approach not only enhances the model’s predictive capability but also demonstrates good tolerance to noise and outliers, mitigates overfitting to a certain extent, and effectively integrates feature variables such as spectral data, texture, and terrain.

In this study, the Random Forest algorithm was implemented using the sklearn library in PyCharm 2023.2.3, a powerful and extensively utilized machine learning library that offers a diverse range of algorithm implementations for tasks including classification, regression, and clustering, encompassing the Random Forest algorithm [24]. High-quality sample points were selected from valid photon data. Utilizing the extraction and analysis tools within ArcGIS, feature variable values corresponding to each discrete sample point were extracted. The data were subsequently partitioned into training and test sets in an 8:2 ratio for model training, parameter optimization, and accuracy validation. The number of decision trees was configured to 200, and the number of random features was set to one-third of the total independent variables [26,27,28]. Although the inversion model exhibits limited stability in complex terrain scenarios and, due to constraints in the number of valid photon samples, the prediction accuracy carries a degree of uncertainty, the inversion results can still be utilized for relative comparative analysis between regions.

2.3.2. Feature Selection

From the preprocessed optical remote sensing data and SRTM DEM data, 17 feature variables were selected [26]. The specific descriptions of each feature variable are shown in

Table 2. Inversion characteristic variables.

Index Name	Index Type	Description
NDRE_Band5	Vegetation Index	$(\mathrm{B}8-\mathrm{B}5)/(\mathrm{B}8+\mathrm{B}5)$
NDRE_Band6		$(\mathrm{B}8-\mathrm{B}6)/(\mathrm{B}8+\mathrm{B}6)$
Ratio Vegetation Index (RVI)		$B8/B4$
B3	band reflectance	Sentinel-2A green band
B5		Sentinel-2A red-edge 1 band
B6		Sentinel-2A red-edge 2 band
B8		Sentinel-2A near-infrared band
B8A		Sentinel-2A narrow near-infrared band
Mean3	texture feature (mean)	GLCM mean of Sentinel-2A B3 band
Mean5		GLCM mean of Sentinel-2A B5 band
Mean6		GLCM mean of Sentinel-2A B6 band
Mean8A		GLCM mean of Sentinel-2A B8A band
Dis3	texture feature (heterogeneity)	GLCM heterogeneity of Sentinel-2A B3 band
Dis5		GLCM heterogeneity of Sentinel-2A B5 band
DEM	Topographic Factors	SRTM DEM at 30 m resolution
Slope		reflects the degree of terrain slope
Aspect		reflects slope aspect

.

Image texture features refer to the characteristics extracted from spectral images that can reflect the spatial structure and spectral information of the image [29]. In this study, the gray-level co-occurrence matrix (GLCM) was used to extract the mean and dissimilarity texture features of Bands B3, B5, B6, and B8A from Sentinel-2 images. The extraction window was set to 3 × 3, with a direction of 45° (southeast).

2.3.3. Feature Variable Selection Based on Correlation Analysis

Redundancy among feature variables can compromise the accuracy of the inversion model and its resultant outputs, potentially giving rise to overfitting or diminishing model stability. Consequently, it becomes imperative to filter multi-source remote sensing feature variables prior to their integration into the inversion model. The filtering procedure is illustrated in Figure 3.

Initially, Pearson correlation analysis was employed to compute the Pearson correlation coefficients between each feature and forest canopy height as well as canopy openness. This process facilitated the elimination of redundant features exhibiting weak correlations while retaining those strongly correlated with canopy height and canopy openness as the candidate feature set. Subsequently, the Variance Inflation Factor (VIF) was utilized to evaluate the extent of multicollinearity among features, leading to the removal of those with excessively high collinearity.

Leveraging the random forest algorithm, feature importance was quantified and ranked using the Mean Decrease Impurity (MDI) method, thereby pinpointing core features that exert a significant influence on the inversion results [26]. Ultimately, a comprehensive evaluation of the outcomes from these three dimensions was conducted, culminating in the selection of eight feature variables, including slope, DEM, RVI, NDRE_BAND6, MEAN3, DIS5, B3, and B5, as the definitive feature variables for input into the random forest inversion model.

2.3.4. Accuracy Metrics

When evaluating the inversion accuracy of canopy height and canopy openness, the coefficient of determination (R²) and root mean square error (RMSE) are adopted for assessment. The calculation formulas are presented as follows:

$$R^2=1-{\displaystyle \frac{{\displaystyle \sum _{i=1}^n{(y_i-{\widehat{y}}_i)}^2}}{{\displaystyle \sum _{i=1}^n{(y_i-\overline{y})}^2}}}$$

(1)

$$RMSE=\sqrt{{\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^n{(y_i-{\widehat{y}}_i)}^2}}$$

(2)

where $y_i$ is the true observed value, ${\widehat{y}}_i$ is the model-predicted value, $\overline{y}$ is the mean of the true observed values, and $n$ is the total number of samples.

3. Experimental Results and Analysis

3.1. Study Area

This study took the Xiangshan Uranium Mining Area as the research area and accurately defined the research scope according to the distribution of mining deposits listed in

Table 3. Locations of the major mineral deposits in the study area.

Deposit	Latitude	Longitude
Youjiashan Deposit	27.55° N	115.94° E
Zoujia Deposit	27.56° N	115.95° E
Julong’an Deposit	27.53° N	115.92° E
Xiangshan Deposit	27.55° N	115.95° E

. The study area is located at the junction of Chongren County and Le’an County, Fuzhou City, Jiangxi Province, as shown in Figure 4. It has a typical subtropical monsoon climate, where deciduous broad-leaved forest is the main vegetation type, and the structure of native vegetation communities is well preserved [30]. The dominant tree species in the area include Betula luminifera H. Winkl., Carpinus fargesiana H. Winkl. var. fargesiana, Castanopsis eyrei (Champ. ex Benth.) Tutch., Schima superba Gardner et Champ., Taxus chinensis var. mairei (Lemée) L.K.Fu and Nan Li, and Fokienia hodginsii (Dunn) Henry et Thomas.

3.1.1. Comparison Area 1 (West Side of the Mining Area)

This area was carefully chosen to meet specific criteria. It has an identical area size to the Xiangshan mining area and shares a roughly similar latitudinal range, being situated to the west of the mining area. Geographically, it lies within a subtropical monsoon climate zone, much like the mining area itself. The vegetation here is predominantly composed of deciduous broadleaf forests, which form the dominant ecological community. In terms of land use, natural forest land makes up the majority. The selection of this Comparison Area is aimed at eliminating any potential interference that latitudinal factors might have on the vegetation ecological conditions.

3.1.2. Comparison Area 2 (North Side of the Mining Area)

The choice of this Comparison Area was based on the principle of ensuring it has the same area as the Xiangshan mining area and a roughly comparable longitudinal range, positioned to the north of the mining area. It generally shares the same climate type and topographical features as the mining area. The vegetation is mainly made up of deciduous forests, and the major land use types include forest land, along with a small proportion of cultivated land. This area is designated to exclude any interference that could be caused by differences in longitude.

3.1.3. Comparison Area 3 (Southeast of the Mining Area)

To circumvent the indirect impacts that might stem from the development of the mining area, an area with a high vegetation cover was selected. This area is located at a relatively greater distance from the Xiangshan mining area, making it an ideal reference standard for assessing the vegetation ecological conditions. The dominant vegetation consists of Phyllostachys edulis (Carrière) J.Houz., Castanopsis faberi Hance, and Cyclobalanopsis glauca (Thunb.) Oerst. and Machilus thunbergii Sieb. et Zucc., forming mixed bamboo-broadleaf forests.

3.2. Experimental Data

This study adopts a multi-source remote sensing data synergy strategy, capitalizing on the distinct technical strengths of each data source to achieve mutual complementarity. This approach guarantees the timeliness, accuracy, and complementarity of the data while ensuring spatial–temporal consistency. The detailed data information is presented as follows:

3.2.1. Sentinel-2 Satellite Data

In this study, Sentinel-2 data were downloaded from the European Space Agency (ESA) Copernicus Data Space Ecosystem. Level-2A products acquired in January 2025 and August 2025 were selected for the leafless period and leaf-on period, respectively. The data have a high spatial resolution and cloud cover of less than 5%, allowing for effective extraction of vegetation spectral characteristics and related indices. They are used to calculate vegetation parameters such as NDVI, FVC, and EVI, reflecting vegetation growth status, coverage, and physiological characteristics.

3.2.2. Sentinel-1 Satellite Data

This study selected Sentinel-1 IW SLC level data from the European Space Agency’s data sharing platform, dated 23 August 2025 and 4 September 2025, respectively, with VV/VH dual-polarization mode. The time span matches that of the Sentinel-2 data. These data are used to calculate the RVI, enhancing data reliability.

3.2.3. ICESat-2 Data

This study acquired ICESat-2 ATL08 LiDAR observation data spanning from February 2020 to October 2025 from the official data platform of the National Aeronautics and Space Administration (NASA), and further screened out datasets consistent with the seasonal phases of Sentinel-2 imagery. This product contains core parameters such as surface elevation, vegetation canopy height, and canopy openness [31], providing rich vegetation structure information and topographic height data. It is used to obtain spatial distribution data of vegetation canopy height and canopy openness in the study area, providing sample support for canopy height inversion.

3.2.4. DEM Data

This study downloaded the SRTM Digital Elevation Model from the National Aeronautics and Space Administration (NASA) data platform, with a spatial resolution of 30 m. Terrain features such as slope and aspect were further extracted based on this data to assist in the inversion of forest structure parameters.

3.3. Result Analysis

3.3.1. Comparison Results of Vegetation Parameters

Based on the preprocessed data and the preliminarily calculated vegetation parameters, the comparison of effective vegetation indices, such as NDVI, EVI, FVC, CIRE, CIG, NDRE, and RVI (Radar Vegetation Index), between the core study area and each comparison study area is presented in

Table 4. Calculation results of the effective vegetation indices.

	NDVI	EVI	FVC	CIRE	CIG	NDRE	RVI
Xiangshan Mining Area	0.56	0.65	0.59	1.25	1.99	0.41	0.81
Comparison Area 1	0.56	0.67	0.60	1.24	2.00	0.41	0.87
Comparison Area 2	0.55	0.65	0.58	1.24	1.96	0.41	0.90
Comparison Area 3	0.56	0.65	0.61	1.23	2.02	0.42	0.89

:

An examination of the analysis and comparison presented in Table 4 reveals that the vegetation indices across different regions are generally similar, suggesting that the overall vegetation condition in the Xiangshan mining area does not significantly deviate from that of the Comparison Areas.

NDVI serves as an indicator of vegetation growth status, with values closer to 1 signifying better growth. The NDVI values for each region hover around 0.56, indicating a fundamental consistency in vegetation growth conditions across these areas. EVI, by mitigating the influence of atmospheric aerosols and soil background, offers a more precise reflection of vegetation information in regions with high vegetation cover. The EVI value in the Xiangshan mining area closely resembles those in Comparison Areas 2 and 3, indicating no marked difference in vegetation density between the mining area and these comparison regions. In contrast, Comparison Area 1 exhibits more robust vegetation growth compared to the mining area. FVC represents the proportion of ground covered by vegetation. The minimal differences in FVC values among the regions suggest generally good and similar vegetation coverage across the areas.

CIRE, CIG, and NDRE are red-edge vegetation indices that are particularly sensitive to vegetation chlorophyll content, photosynthetic intensity, and stress conditions. In the Xiangshan mining area, the CIRE, CIG, and NDRE indices are 1.25, 1.99, and 0.41, respectively. The CIRE value is marginally higher than those in the Comparison Areas, while the CIG and NDRE values are at moderate levels. This indicates that the vegetation in the mining area has a relatively high chlorophyll content and is generally at a moderate level of vegetation cover, without showing obvious signs of heavy-metal stress. The vegetation health status is consistent with that of natural areas, and there are no significant differences in spectral characteristics based on the red-edge band.

The mean RVI value of the Xiangshan mining area is 0.81, which is remarkably lower than that of Comparison Area 1 at the same latitude (0.87), Comparison Area 2 at the same longitude (0.90), and Comparison Area 3 with high vegetation coverage (0.89). Previous studies have verified that RVI presents a linear correlation with vegetation water content. In the research on plot-scale soil moisture inversion in karst mountainous areas, Zhang [24] adopted RVI to estimate vegetation water content and took it as a characteristic factor for soil moisture inversion. The evident discrepancy in RVI values between the mining area and Comparison Areas indirectly indicates possible variations in vegetation structure as well as surface water and soil conditions within the mining zone.

In conclusion, when compared under the same latitude conditions, the vegetation growth status and coverage in the mining area do not show significant differences from those in natural forest land. Under the same longitude conditions, the NDVI, FVC, and CIG indices in the Xiangshan mining area are slightly higher than those in Comparison Area 2; it is speculated that this difference is related to the more rugged topography of Comparison Area 2. Compared with the natural ecological Comparison Area 3, all indices except RVI are at similar levels, with no statistically significant differences, indicating that the existence of the mining area has not had a substantial impact on basic vegetation growth indicators. However, further attention should be directed towards the differences in vegetation structure reflected by the RVI.

3.3.2. Canopy Height

Canopy height refers to the vertical distance from the top of the vegetation canopy to the ground. It directly reflects vegetation growth status and the complexity of canopy structure. By comparing canopy heights between the mining area and the Comparison Areas, the stability of the vegetation ecosystem can be assessed.

• leaf-on period

As illustrated in Figure 5a, the canopy height within the Xiangshan mining area predominantly falls within the range of 10 to 15 m. The regions with higher canopy heights are primarily situated at the periphery of the mining area, encompassing natural forest land, ecological restoration zones, and areas relatively distant from mining operations. In these locales, the vegetation remains undisturbed by mining activities, resulting in well-developed canopies. Conversely, the areas with lower canopy heights are mainly clustered in proximity to mining deposits, such as around the Youjiashan and Zoujia deposits. Here, the canopy heights are generally low, and the development of the vegetation canopy is stunted.

As depicted in Figure 5b,c, the spatial distribution of canopy height in Comparison Areas 1 and 2 is relatively even. Small patches of low-value areas are present, mainly influenced by topographic relief. These low-value areas are distributed in regions characterized by significant terrain variations, such as steep slopes and low-lying areas.

As presented in Figure 5d, Comparison Area 3, being a native area with high vegetation cover, generally exhibits higher canopy heights and the most uniform spatial distribution. It boasts the largest expanse of high-value areas and lacks any obvious concentrated low-value regions.

Table 5. Statistics of the canopy height retrieval results in the leaf-on period.

	Min (m)	Max (m)	Mean (m)
Xiangshan Mining Area	8.49	21.24	14.54
Comparison Area 1	7.42	20.81	15.86
Comparison Area 2	5.07	21.28	15.74
Comparison Area 3	10.51	21.62	18.22

reveals that the maximum canopy height in the Xiangshan mining area reaches 21.24 m, which is marginally higher than that of Comparison Area 1 yet lower than that observed in Comparison Areas 2 and 3. The minimum canopy height stands at 8.49 m, surpassing the values recorded in both Comparison Areas 1 and 2. The mean canopy height is calculated at 14.54 m, which is notably lower than the means of all three Comparison Areas.

It can be seen from

Table 6. Statistics of the area proportion of canopy height in the leaf-on period.

Canopy Height (m)	Xiangshan Mining Area (%)	Comparison Area 1 (%)	Comparison Area 2 (%)	Comparison Area 3 (%)
<10	0.00	0.02	0.02	0.00
10–15	0.62	0.24	0.29	0.02
15–20	0.37	0.74	0.69	0.91
>20	0.01	0.00	0.00	0.08

that in terms of area distribution, the canopy height of vegetation in the Xiangshan mining area is mostly concentrated in the range of 10 to 15 m, with the largest proportion, while the proportion of vegetation with canopy height ranging from 15 to 20 m is obviously lower than that in the three Comparison Areas. The results reveal that although the canopy height in the Xiangshan mining area has a wide distribution range, the overall vegetation canopy height is lower than that in the Comparison Areas, and the areas with low canopy height show obvious spatial aggregation characteristics.

ii.

leaf-off period

According to the spatial distribution maps of inverted canopy height in the leafless period (Figure 6), low-value areas of canopy height in the Xiangshan mining area are concentrated around mining deposits, which is consistent with the spatial distribution in the leaf-on period. It proves that the low canopy characteristics of vegetation in the mining area are not influenced by seasonal changes. The canopy height is evenly distributed in Comparison Area 1 and Comparison Area 2 without contiguous high-value or low-value regions. Comparison Area 3 has slight spatial differences in canopy height with an overall high canopy height level.

Table 7. Statistics of canopy height retrieval results in the leaf-off period.

	Min (m)	Max (m)	Mean (m)
Xiangshan Mining Area	7.48	21.41	14.64
Comparison Area 1	5.14	21.94	16.49
Comparison Area 2	4.47	22.11	15.95
Comparison Area 3	9.05	21.87	17.29

and

Table 8. Statistics of the area proportion of canopy height in the leaf-off period.

Canopy Height (m)	Xiangshan Mining Area (%)	Comparison Area 1 (%)	Comparison Area 2 (%)	Comparison Area 3 (%)
<10	0.00	0.01	0.01	0.00
10–15	0.61	0.15	0.26	0.04
15–20	0.38	0.83	0.72	0.94
>20	0.00	0.01	0.01	0.02

show the statistical values of inverted canopy height and the area proportion of canopy height at different intervals in each study area during the leafless period, and the results are generally consistent with those of the leaf-on period. Specifically, the mean canopy height of the Xiangshan mining area ranks the lowest among the four study areas in both leaf-on and leaf-off periods. Moreover, the mining area has the largest area proportion of canopy height within 10–15 m, while its proportion in the 15–20 m interval is lower than that of the three Comparison areas. The above results further demonstrate that the low-canopy structural features of vegetation in the mining area are independent of seasonal variation.

iii.

The Spatial Autocorrelation Analysis

Spatial autocorrelation analysis results (

Table 9. Spatial autocorrelation analysis of canopy height in various areas in the leaf-on period.

	Xiangshan Mining Area	Comparison Area 1	Comparison Area 2	Comparison Area 3
Moran’s I index	0.87	0.82	0.81	0.81
Z-score	146.80	140.43	138.04	137.19

and

Table 10. Spatial autocorrelation analysis of canopy height in various areas in the leaf-off period.

	Xiangshan Mining Area	Comparison Area 1	Comparison Area 2	Comparison Area 3
Moran’s I index	0.87	0.82	0.82	0.66
Z-score	147.90	139.49	139.61	112.60

) reveal that, except for the third Comparison Area covered by primary vegetation with obvious seasonal variations, the Moran’s I index and Z-score of other areas fluctuate slightly between the leaf-on and leaf-off periods, and their overall spatial aggregation characteristics remain stable. Statistical data show that Moran’s I index of the Xiangshan Uranium Mining Area steadily stays at 0.87 in both periods, and its Z-score ranks first among all study areas. This demonstrates that the vegetation canopy height in the mining area has significant positive spatial aggregation characteristics and forms a distinct spatial pattern with clear boundaries between high and low canopy vegetation. The first and second Comparison Areas also present stable spatial distribution rules. The Moran’s I index of the third Comparison Area is 0.81 in the leaf-on period and drops to 0.66 in the leaf-off period, which is the lowest among all areas in the corresponding period. Restricted by the inversion accuracy of the model, this study does not explore the internal mechanism of the seasonal decline of Moran’s I index in this area, and only adopts the data for horizontal comparison among different regions in the same period. In follow-up research, field observation data can be combined to further explain the causes of such numerical changes.

3.3.3. Canopy Openness

Canopy Openness is typically defined as the proportion of sky area not obscured by the vegetation canopy when observed upward from a point on the ground. It is a core indicator reflecting the degree of canopy closure and serves as an inverse indicator of forest canopy density. The lower the canopy openness, the denser and more intact the canopy structure; conversely, higher openness suggests structural damage to the canopy and lower density.

i.

leaf-on period

As shown in Figure 7a, the canopy openness values of vegetation in the Xiangshan mining area are mainly concentrated within the range of 0.3 to 0.4. It can be observed from the spatial distribution map that the clustered low-value areas of canopy openness overlap with the locations of mining deposits. High-value areas account for a small proportion within the mining area, while low-value areas are more widely distributed, showing a distinct difference from the Comparison Areas. Areas with high canopy openness are mostly distributed in the peripheral zones of the mining area, where the vegetation canopy structure is similar to that in the comparison regions.

The statistical data presented in

Table 11. Statistics of canopy openness retrieval results in the leaf-on period.

	Min	Max	Mean
Xiangshan Mining Area	0.22	0.58	0.37
Comparison Area 1	0.19	0.56	0.40
Comparison Area 2	0.14	0.60	0.41
Comparison Area 3	0.28	0.57	0.47

reveal that the maximum canopy openness in the Xiangshan mining area reaches 0.58, a value that is approximately on par with those observed in the three Comparison Areas. The minimum canopy openness in the mining area is 0.22, which is higher than the minima recorded in Comparison Areas 1 and 2 but lower than that in Comparison Area 3. The average canopy openness in the mining area stands at 0.37, which is lower than the averages of all three Comparison Areas.

Table 12. Statistics of the area proportion of canopy openness in the leaf-on period.

Canopy Openness	Xiangshan Mining Area (%)	Comparison Area 1 (%)	Comparison Area 2 (%)	Comparison Area 3 (%)
<0.3	0.07	0.04	0.05	0.00
0.3–0.4	0.66	0.35	0.36	0.04
0.4–0.5	0.26	0.60	0.57	0.76
0.5–0.6	0.01	0.01	0.03	0.20

indicates that within the Xiangshan mining area, canopy openness values in the range of 0.3–0.4 account for the largest proportion of the total area. In contrast, the proportion of the area with canopy openness values in the range of 0.4–0.5 is relatively low. This suggests that, compared to the three Comparison Areas, the Xiangshan mining area has a larger expanse of areas characterized by low-to-medium canopy openness values and a smaller area with high values. This reflects the fact that, relative to the Comparison Areas, the Xiangshan mining area boasts extensive regions with dense and well-structured vegetation canopies.

ii.

leaf-off period

As shown in Figure 8a, there are extensive low-value areas of canopy openness in the Xiangshan mining area during the leaf-off period, which is similar to the results in the leaf-on period.

Table 13. Statistics of canopy openness retrieval results in the leaf-off period.

	Min	Max	Mean
Xiangshan Mining Area	0.22	0.57	0.37
Comparison Area 1	0.13	0.59	0.42
Comparison Area 2	0.13	0.59	0.41
Comparison Area 3	0.24	0.59	0.45

and

Table 14. Statistics of the area proportion of canopy openness in the leaf-off period.

Canopy Openness	Xiangshan Mining Area (%)	Comparison Area 1 (%)	Comparison Area 2 (%)	Comparison Area 3 (%)
<0.3	0.04	0.02	0.02	0.00
0.3–0.4	0.66	0.23	0.31	0.06
0.4–0.5	0.29	0.74	0.65	0.88
0.5–0.6	0.01	0.02	0.02	0.06

show that the average canopy openness of the Xiangshan mining area is the lowest among all study areas. Within the mining area, regions with canopy openness ranging from 0.3 to 0.4 account for the largest proportion of the total area, while those in the 0.4–0.5 range occupy a relatively small proportion, which is consistent with the area statistics in the leaf-on period. It can be concluded that in the leaf-off period, the Xiangshan mining area features a wider range of areas with low and medium canopy openness and a smaller area with high canopy openness compared with the three Comparison Areas. This indicates that the mining area possesses a larger proportion of areas with compact vegetation structures, and such vegetation structural characteristics are not affected by seasonal changes.

iii.

The Spatial Autocorrelation Analysis

The Spatial autocorrelation analysis of canopy openness (

Table 15. Spatial autocorrelation analysis of canopy openness in the leaf-on period.

	Xiangshan Mining Area	Comparison Area 1	Comparison Area 2	Comparison Area 3
Moran’s I index	0.85	0.79	0.79	0.77
Z-score	144.44	135.31	133.73	131.51

and

Table 16. Spatial autocorrelation analysis of canopy openness in the leaf-off period.

	Xiangshan Mining Area	Comparison Area 1	Comparison Area 2	Comparison Area 3
Moran’s I index	0.76	0.70	0.73	0.47
Z-score	128.25	119.64	123.52	80.01

) indicated that in the leaf-on period, the Moran’s I index of the mining area was 0.85 with a Z-score of 144.44, while in the leafless period, its Moran’s I index was 0.76 and its Z-score was 128.25. Horizontal comparison with other Comparison Areas in the same period revealed that both indicators of the mining area were higher than those of other regions. This proves that the canopy openness in the mining area shows the strongest positive spatial aggregation among the four study areas. In comparison, Comparison Area 3 had a lower Moran’s I index than the other three areas during the same period, meaning its canopy openness is distributed more evenly in space, which is in line with the spatial distribution characteristics of canopy height.

3.4. Accuracy Evaluation of the Inversion Model

To verify the reliability of the vegetation canopy structure parameter inversion model based on multi-source remote sensing data, this study randomly divided the sample data into a training set and an independent test set at a ratio of 8:2. A quantitative accuracy evaluation (

Table 17. Model inversion accuracy.

	Leaf-On		Leaf-Off
	R2	RMSE	R2	RMSE
Canopy Height	0.301	4.931 m	0.328	5.105 m
Canopy Openness	0.225	0.154	0.240	0.158

) was performed on the inversion results of canopy height and canopy openness using the independent test set to ensure the objectivity of model evaluation, providing a reliable data support for the subsequent analysis of vegetation structural differences between the mining area and the Comparison Areas.

Model accuracy evaluation results show that for canopy height, the coefficient of determination R² on the independent test set is 0.301 for the leaf-on period and 0.328 for the leaf-off period, with corresponding RMSE values of 4.931 m and 5.105 m. For canopy openness, the R² is 0.225 for the leaf-on period and 0.240 for the leaf-off period, with corresponding RMSE values of 0.154 and 0.158. Existing studies have confirmed that this inversion method has limited predictive performance in areas with complex terrain and broad-leaved forest coverage. Accordingly, this study interprets experimental results based on spatial distribution trends rather than drawing absolute conclusions. In terms of spatial distribution patterns and relative regional differences, the model can effectively distinguish vegetation structural disparities between the mining area and Comparison Areas, and its reflected relative trends are reliable.

4. Discussion

4.1. Analysis of Error Sources and Limitations

Statistical comparison of inversion results shows that the Xiangshan area has a lower mean canopy height and canopy openness than the Comparison Areas, exhibiting an overall relatively short and dense canopy structure. However, the model achieves limited accuracy on the independent test set, which is consistent with the conclusions of existing studies. Tian et al. [32] conducted forest canopy height mapping by combining ICESat-2 and Landsat-8 data, with an average canopy height inversion accuracy of R² = 0.467 and RMSE = 2.848 m, confirming that differences in vegetation and terrain conditions lead to reduced accuracy. Zhang et al. [33] further verified that canopy height inversion in areas with vegetation coverage greater than 80% yields R² = 0.5 and RMSE = 4.6 m. Wang [26] systematically validated forest canopy height inversion accuracy, showing that steeper slopes and higher vegetation coverage result in larger errors. When the slope exceeds 20°, R² = 0.35 and RMSE = 6.52 m; when vegetation coverage is greater than 80%, R² = 0.51 and RMSE = 4.53 m. These studies collectively demonstrate that an R² value between 0.3 and 0.5 is a common outcome of current technical approaches for canopy structure inversion in regions with high vegetation coverage and complex terrain. The inversion accuracy of this study is therefore within a reasonable range, though it limits the results to spatial trend analysis rather than supporting precise quantitative assessment.

In addition, due to insufficient valid samples, this study employed Sentinel-2 data from January 2025 (leaf-off period) and August 2025 (leaf-on period), combined with ICESat-2 data from corresponding seasons between 2020 and 2025 for parameter inversion. Temporal mismatches between multi-period LiDAR data and optical imagery may introduce uncertainties into the inversion results. Future research can further improve inversion accuracy by incorporating data with higher temporal and spatial matching precision and integrating field sampling.

4.2. Complementary Analysis of Optical Vegetation Indices and Vegetation Structural Parameters

Three categories of vegetation indices with high research value for the study area were selected in this paper: visible-near-infrared vegetation indices (NDVI, EVI), red-edge vegetation indices (CIG, CIRE), and the microwave-based RVI. The aim was to comprehensively analyze the vegetation ecology of the study area from the perspectives of spectral characteristics, physiological status, and structural features. However, the experimental results show no significant differences in optical vegetation indices between the Xiangshan mining area and the Comparison Areas, indicating that the threshold ranges of optical indices alone cannot reveal vegetation structural differences, while RVI exhibits notable disparities. Therefore, two parameters reflecting canopy physical structure—canopy height and canopy openness—were introduced to verify structural variations in the study area.

Existing vegetation anomaly detection methods, such as the Vegetation Index considering Greenness and Shortwave infrared (VIGS), primarily rely on the spectral response of shortwave infrared bands to mineral alteration and vegetation stress to identify anomalies. The method adopted in this study complements such approaches: it performs quantitative inversion by integrating LiDAR data, while also incorporating RVI as supplementary information. Notably, when optical indices (e.g., NDVI, EVI) show no anomalies, structural parameters and RVI already exhibit remarkable differences. This suggests that monitoring schemes incorporating physical structural data may offer greater potential for early anomaly detection than those relying solely on optical indices. A systematic comparative validation of the two approaches can be explored in future research.

4.3. Relationship Between Vegetation Structural Anomalies and Mining Deposit Distribution

Combining the complementary advantages of optical vegetation indices and vegetation structural parameters, this study confirms that the vegetation in the Xiangshan mining area is characterized by low canopy height and dense canopy structure, which presents a spatially aggregated distribution rather than a random distribution. Despite the limited inversion accuracy, the low-value zones of canopy height and canopy openness maintain consistent spatial aggregation in both leaf-on and leaf-off periods, and such aggregation features are not observed in adjacent areas and primary forest zones.

Restricted by objective conditions, this study has not carried out field verification, and the analysis is mainly based on publicly available spaceborne remote sensing data. In follow-up research, interference from natural conditions, including terrain factors and human activities, should be further eliminated. Combined with field sampling and in situ measurement, the formation mechanism of vegetation ecological anomalies in the mining area can be deeply explored, so as to improve the reliability of applying vegetation ecological differences as an auxiliary indicator for concealed uranium deposit exploration.

5. Conclusions

The Xiangshan Uranium Mining Area in Jiangxi Province stands as a pivotal uranium resource hub in southern China. Nevertheless, there is a relative scarcity of multi-source remote sensing studies focusing on vegetation ecology within this mining area. This scarcity poses challenges in uncovering the potential influence of mining deposits on vegetation ecology and in precisely discerning the disparities in vegetation ecological characteristics between the mining area and typical vegetation zones. As a result, it hampers the ability to offer robust scientific backing for mineral exploration activities in the region.

In view of this, this study adopts multi-source remote sensing data fusion combined with the random forest machine learning algorithm. Firstly, multi-source remote sensing data are preprocessed to obtain valid data covering the entire study area. Afterward, various vegetation indices, spectral band texture features, and topographic factors are extracted to serve as characteristic variables and vegetation ecological evaluation indicators. Core characteristic variables are screened out through correlation analysis and a multicollinearity test. On this basis, a random forest inversion model is constructed to invert canopy structure parameters across the study area, and a comprehensive analysis is carried out on vegetation indicators and inversion results.

The research results demonstrate that there are no significant differences in optical vegetation indices, including NDVI, EVI, and FVC, between the Xiangshan Uranium Mining Area and Comparison Areas, and no obvious disparities are found in vegetation growth status and vegetation coverage. By contrast, RVI in the mining area is distinctly lower. Since RVI is sensitive to canopy structure and soil environment conditions, its low values may indicate damaged vegetation structural integrity or altered soil properties in mining areas, which need to be verified by field survey data in subsequent studies.

Further analysis on canopy structure indicates that after excluding the interference of climatic factors and within the acceptable range of model accuracy, canopy structure parameters present obvious spatial distribution differences between mining areas and Comparison areas, and such differences are consistent in both leaf-on and leaf-off periods. The inversion results of the Xiangshan mining area show an obvious positive spatial aggregation trend. The low-value zones of canopy height and canopy openness are spatially consistent with the distribution of the Youjiashan deposit, Zoujia deposit, and other mineral deposits, which indicates that vegetation in mining areas tends to grow shorter and denser. In addition, the mining area has a larger proportion of regions with medium and low canopy height and canopy openness values compared with ordinary vegetation areas, proving that identifiable ecological differences exist in canopy structure between vegetation around mining deposits and conventional vegetation zones.

To gain a better understanding of the practical significance of using vegetation in mining areas as an indicator for mineral exploration, future research can continue along this path by further improving data accuracy and conducting more precise analyses of vegetation ecological differences. At the same time, field sampling should be integrated to further examine specific changes in vegetation physiological characteristics and to delve into the underlying reasons for the impact of mining deposits on vegetation ecology.