In our recent research, we have explored the application of UAV drone remote sensing technology for monitoring geomorphological landscapes, particularly in mining environments. The unique capabilities of UAV drone platforms, when combined with multispectral sensors, provide a powerful tool for capturing high-resolution spatial and spectral data. This approach overcomes many limitations of traditional ground-based surveys and satellite remote sensing, offering timely, accurate, and cost-effective monitoring solutions. In this article, we present our findings based on a case study of an ecological restoration area in a mining district, where we employed a UAV drone equipped with a multispectral camera to assess land cover changes and vegetation dynamics over a two-year period.
Our work demonstrates that UAV drone remote sensing enables centimeter-level spatial resolution, minimal atmospheric interference, and flexible data acquisition schedules. By utilizing multispectral imagery, we can derive quantitative vegetation indices, classify land use types, and detect subtle changes in surface features. The technology has proven invaluable for mine landscape monitoring, reducing the enormous human and material resources traditionally required for such tasks. Below, we detail the technical principles, system components, methodology, and application results from our study.
Technical Principles of UAV Drone Multispectral Remote Sensing
The core of UAV drone multispectral remote sensing lies in capturing the reflected or emitted electromagnetic radiation from surface objects across multiple narrow spectral bands. Each band is selected to highlight specific biophysical or chemical properties of the target. For instance, healthy vegetation strongly absorbs red light and reflects near-infrared (NIR) light, while bare soil and water exhibit distinct spectral signatures. This spectral fingerprint forms the basis for classification and change detection.
The typical processing chain for UAV drone data includes radiometric calibration (correcting for sensor noise and illumination variations), geometric correction (orthorectification), and the generation of reflectance maps. Vegetation indices such as the Normalized Difference Vegetation Index (NDVI) are then computed using:
$$NDVI = \frac{\rho_{NIR} – \rho_{Red}}{\rho_{NIR} + \rho_{Red}}$$
where ρNIR and ρRed are the reflectances in the near-infrared and red bands, respectively. Other indices include the Normalized Difference Water Index (NDWI) and the Soil Adjusted Vegetation Index (SAVI), which help discriminate water bodies and sparsely vegetated areas. Table 1 summarizes common vegetation indices used in our analysis.
| Index | Formula | Application |
|---|---|---|
| NDVI | $$\frac{NIR – Red}{NIR + Red}$$ | Vegetation density and health |
| NDWI | $$\frac{Green – NIR}{Green + NIR}$$ | Water body detection |
| SAVI | $$\frac{(NIR – Red)(1 + L)}{NIR + Red + L}$$ (L = 0.5) | Vegetation in areas with high soil exposure |
| EVI | $$2.5 \times \frac{NIR – Red}{NIR + 6 \times Red – 7.5 \times Blue + 1}$$ | Enhanced vegetation index in high biomass regions |
UAV Drone System Components
Our monitoring system consisted of three main components: an autonomous UAV drone platform, a multispectral sensor payload, and post-processing software. We selected the ZhiHang SF3300, a vertical takeoff and landing (VTOL) fixed-wing UAV drone, for its long endurance (up to 90 minutes) and ability to cover large areas efficiently. The MS600 multispectral camera, with six configurable spectral channels, was mounted on the UAV drone. This sensor supports 17 optional wavelengths and includes a downwelling irradiance sensor for real-time radiometric correction. Table 2 lists the key specifications of our system.
| Parameter | Value |
|---|---|
| UAV drone model | ZhiHang SF3300 (VTOL fixed-wing) |
| Multispectral camera | MS600 (6 channels) |
| Typical flight altitude | 430 m above ground |
| Ground sampling distance | 30 cm |
| Endurance | 80 min (for the mission) |
| Spectral bands used | Blue (450 nm), Green (560 nm), Red (650 nm), Red-edge (730 nm), NIR (840 nm), another NIR (900 nm) |
The figure below shows an example of our UAV drone platform during a field mission.
Data Acquisition and Processing Workflow
Our data acquisition plan involved a single flight over the study area of approximately 7.04 km². The flight parameters were optimized to achieve 80% forward overlap and 70% side overlap to ensure high-quality three-dimensional reconstruction and seamless image mosaicking. Table 3 summarizes the flight parameters used.
| Parameter | Value |
|---|---|
| Flight platform | ZhiHang SF3300 UAV drone |
| Multispectral camera | MS600 |
| Flight altitude (AGL) | 430 m |
| Ground speed | 6.8 m/s |
| Forward overlap | 80% |
| Side overlap | 70% |
| Number of flights | 1 |
| Total flight time | 80 min |
| Ground sampling distance | 30 cm |
After the flight, the raw images were transferred to a processing workstation. We used DPGrid software for aerial triangulation (AT) and digital elevation model (DEM) generation. The orthophoto mosaic was created after applying radiometric correction using the irradiance sensor data. The final digital orthophoto map (DOM) had a resolution of 0.3 m and was georeferenced to the Chinese Geodetic Coordinate System 2000 (CGCS2000). For change detection, we compared this DOM with a historical orthophoto from the previous year using polynomial registration.
Land Cover Classification and Change Detection
We classified land cover into ten categories based on the national standard GB/T 21010-2017 and field observations: high-coverage grassland, shrubland, dryland cropland, bare land, built-up land, other woodland, reservoir/pond, forestland, medium-coverage grassland, and river/canal. Spectral indices like NDVI and NDWI were used to generate initial classification layers, followed by object-oriented segmentation and supervised classification using the maximum likelihood algorithm. Accuracy assessment was performed using 64 validation points randomly selected from the field survey. The overall classification accuracies for 2022 and 2023 were 91.94% and 90.62%, with Kappa coefficients of 0.899 and 0.863, respectively. Table 4 shows the confusion matrix for the 2022 classification.
| Actual \ Predicted | High grassland | Shrubland | Dryland | Bare land | Built-up | Other woodland | Reservoir | Forestland | Medium grassland | River | Total |
|---|---|---|---|---|---|---|---|---|---|---|---|
| High grassland | 60 | 1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 62 |
| Shrubland | 0 | 15 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 0 | 16 |
| Dryland | 0 | 0 | 38 | 1 | 0 | 0 | 0 | 0 | 1 | 0 | 40 |
| Bare land | 0 | 0 | 1 | 12 | 0 | 0 | 0 | 0 | 0 | 0 | 13 |
| Built-up | 0 | 0 | 0 | 0 | 10 | 0 | 0 | 0 | 0 | 0 | 10 |
| Other woodland | 0 | 0 | 0 | 0 | 0 | 45 | 0 | 0 | 0 | 0 | 45 |
| Reservoir | 0 | 0 | 0 | 0 | 0 | 0 | 4 | 0 | 0 | 0 | 4 |
| Forestland | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 12 | 0 | 0 | 12 |
| Medium grassland | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 0 | 18 | 0 | 20 |
| River | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 2 | 2 |
| Total | 61 | 16 | 41 | 13 | 10 | 45 | 4 | 12 | 20 | 2 | 224 |
To quantify land cover transitions, we constructed a transition matrix between 2022 and 2023 (Table 5). The analysis revealed that dryland cropland increased by 24.62 ha (a 40.39% increase), primarily converted from high-coverage grassland (11.23 ha), medium-coverage grassland (8.29 ha), and bare land (4.99 ha). Conversely, medium-coverage grassland decreased by 17.21 ha (38.79% decrease), with significant losses to high-coverage grassland (5.78 ha), dryland (8.29 ha), and bare land (3.59 ha). Shrubland declined dramatically by 48.12% (17.12 ha), mostly converting to high-coverage grassland (12.94 ha).
| 2022 \ 2023 | High grassland | Shrubland | Dryland | Bare land | Built-up | Other woodland | Reservoir | Forestland | Medium grassland | River | Total 2022 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| High grassland | 413.10 | 0 | 11.63 | 2.93 | 0.03 | 0.06 | 0 | 0 | 0.23 | 0 | 427.98 |
| Shrubland | 12.94 | 18.46 | 0 | 0.28 | 0 | 3.24 | 0 | 0 | 0.66 | 0 | 35.58 |
| Dryland | 0.40 | 0 | 58.36 | 0.39 | 0 | 0.53 | 0 | 0 | 1.28 | 0 | 60.96 |
| Bare land | 0.40 | 0 | 5.38 | 17.75 | 0 | 0 | 0 | 0 | 0.77 | 0 | 24.30 |
| Built-up | 0 | 0 | 0 | 0 | 19.65 | 0 | 0 | 0 | 0 | 0 | 19.65 |
| Other woodland | 1.58 | 0 | 0.29 | 0.36 | 0 | 65.89 | 0 | 0 | 0 | 0 | 68.12 |
| Reservoir | 0 | 0 | 0 | 0 | 0 | 0 | 4.86 | 0 | 0 | 0 | 4.86 |
| Forestland | 0.05 | 0 | 0 | 0.24 | 0 | 0 | 0 | 17.47 | 0 | 0 | 17.76 |
| Medium grassland | 6.01 | 0 | 9.57 | 4.36 | 0.18 | 0 | 0.03 | 0 | 24.22 | 0 | 44.37 |
| River | 0 | 0 | 0.35 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0.35 |
| Total 2023 | 434.48 | 18.46 | 85.58 | 26.31 | 19.86 | 69.72 | 4.89 | 17.47 | 27.16 | 0 | 703.93 |
These changes reflect the ongoing ecological restoration activities in the mining area. The increase in dryland and high-coverage grassland indicates successful revegetation on reclaimed lands, while the loss of shrubland and medium-coverage grassland may be due to land conversion for agricultural purposes or natural succession dynamics. The UAV drone remote sensing data allowed us to pinpoint these transitions with high spatial accuracy.
Discussion and Advantages of UAV Drone Technology
Our study confirms that UAV drone remote sensing offers distinct advantages over traditional methods. Compared to manual surveys using total stations or GPS-RTK, UAV drone surveys cover the entire 7 km² area in a single flight of 80 minutes, whereas ground-based work would require weeks. Compared to satellite imagery, UAV drone data provide 30 cm spatial resolution free from cloud cover and with flexible revisit times. The multispectral capability enables quantitative monitoring of vegetation health through indices like NDVI. Table 6 compares the performance of UAV drone remote sensing with traditional approaches based on our experience.
| Feature | Manual Survey (Total Station/GNSS) | Satellite Remote Sensing | UAV Drone Remote Sensing |
|---|---|---|---|
| Coverage area | Limited (points or small areas) | Large (hundreds of km²) | Medium (tens of km² per flight) |
| Spatial resolution | Point accuracy (cm) | 0.5–30 m (often >10 m) | 1–30 cm |
| Temporal resolution | Days to weeks | Days to weeks (revisit cycle) | Hours to days (on-demand) |
| Data dimensionality | Geometric only (x,y,z) | Multispectral/hyperspectral | Multispectral + 3D point cloud |
| Cost per km² | High (labor-intensive) | Moderate (image purchase) | Low (equipment + operator) |
| Weather limitations | Works in most weather | Cloud cover problematic | Wind and rain limitations |
| Vegetation index capability | No | Yes | Yes (high resolution) |
The high spatial resolution of UAV drone data allowed us to detect small-scale features such as individual reclaimed patches, erosion gullies, and vegetation stress zones that would be invisible in satellite imagery. The ability to generate accurate digital surface models (DSM) from the same flight provided additional topography information for geomorphological analysis. In our project, the UAV drone-derived NDVI maps clearly distinguished between different vegetation cover classes, as shown in Table 7, where we classified NDVI ranges into five categories.
| NDVI Range | Vegetation Coverage | Description |
|---|---|---|
| −1 to −0.5 | Severely non-vegetated | Bare soil, buildings, water |
| −0.5 to 0 | Non-vegetated | Almost no vegetation |
| 0 to 0.3 | Low vegetation cover | Sparse vegetation |
| 0.3 to 0.6 | Moderate vegetation cover | Active but not dense vegetation |
| 0.6 to 1 | High vegetation cover | Dense, healthy vegetation |
One of the key contributions of our work is the generation of quantitative change maps that directly inform mine restoration managers. For instance, the expansion of dryland cropland by 24.62 ha suggests successful soil reconstruction and agricultural reclamation. However, the reduction of shrubland by 17.12 ha might indicate a need for targeted shrub reintroduction programs. The UAV drone technology enables such fine-scale assessment annually or even seasonally.
Future Outlook
We believe that the integration of UAV drone remote sensing with artificial intelligence will further enhance its capabilities. Autonomous flight planning systems that adapt to real-time conditions, along with deep learning models for automatic feature extraction and change detection, are already emerging. Multi-modal data fusion, combining UAV drone multispectral with LiDAR or thermal infrared sensors, will provide a more complete picture of landscape health. As sensor miniaturization continues and battery technology improves, the operational efficiency and data quality of UAV drone systems will only increase, making them indispensable for geomorphological monitoring worldwide. Our study serves as a concrete example of how UAV drone technology can transform traditional environmental monitoring into a precise, efficient, and intelligent process.