Hyperspectral Imaging for Fine Tree Species Identification and Leaf Area Index Retrieval

Mangroves are among the most important ecological barriers in coastal environments, and their health directly affects shoreline stability, biodiversity, and carbon storage. To understand the condition of mangrove ecosystems, two indicators are especially important: tree species distribution and leaf area index (LAI). In recent years, hyperspectral mangrove ecological monitoring has become an increasingly effective approach for capturing both types of information with high precision and high efficiency.

In a study conducted in the Zhangjiang Estuary Mangrove National Nature Reserve, UAV-based hyperspectral imaging was used to support fine tree species identification with a hyperspectral camera and leaf area index retrieval using hyperspectral imaging. The results showed that hyperspectral remote sensing can accurately distinguish dominant mangrove species and provide reliable LAI estimation for ecological assessment, restoration planning, and vegetation monitoring across complex intertidal environments.

Why Hyperspectral Mangrove Ecological Monitoring Matters

Mangrove reserves often contain multiple species growing in fragmented and overlapping patterns. At the same time, field investigation in coastal wetlands is usually limited by tides, muddy terrain, and low accessibility. This makes traditional vegetation surveys slow, labor-intensive, and difficult to scale.

The Need for Fine Tree Species Identification

For mangrove management, species-level information is essential. Different species have different ecological functions, growth responses, and restoration values. Accurate mapping of species distribution helps researchers and reserve managers understand habitat structure and make better conservation decisions.

This is why fine tree species identification with a hyperspectral camera is such an important application. Compared with conventional imaging methods, hyperspectral imaging can capture subtle spectral differences among visually similar species. That makes it particularly valuable in mixed mangrove environments where accurate classification is difficult to achieve through RGB or standard multispectral imagery alone.

The Importance of Leaf Area Index Retrieval

Leaf area index retrieval with a hyperspectral camera is another key component of ecological monitoring. LAI reflects canopy density, vegetation growth status, and photosynthetic capacity. It is widely used to evaluate ecosystem productivity, vegetation health, and restoration performance.

Traditional LAI measurement methods are often time-consuming and unsuitable for large-scale wetland monitoring. By contrast, hyperspectral leaf area index retrieval provides a faster and more scalable way to estimate canopy parameters across broad areas while maintaining strong accuracy.

Hyperspectral Camera Advantages for Mangrove Monitoring

The study demonstrates how hyperspectral imaging can support both fine tree species identification and leaf area index retrieval in complex coastal ecosystems.

High Spectral Sensitivity for Species Differentiation

A hyperspectral camera can record detailed reflectance information across a continuous spectral range, making it possible to detect small differences in leaf and canopy reflectance among species. This capability is especially important for hyperspectral camera-based tree species identification, where classification depends on subtle but stable spectral variation.

In the Zhangjiang Estuary reserve, dominant mangrove species such as Kandelia obovata, Aegiceras corniculatum, and Avicennia marina occur in a complex mosaic distribution. Their spectral differences can be effectively captured through hyperspectral imaging, supporting more accurate classification than traditional survey approaches.

High Spatial Resolution for Canopy Structure Analysis

Besides spectral richness, UAV hyperspectral imaging also offers high spatial detail. At a flight altitude of 100 meters, the imagery achieved centimeter-level ground resolution, allowing clear observation of canopy structure, crown texture, and vegetation boundaries.

This is particularly important for hyperspectral mangrove ecological monitoring, because species classification and LAI retrieval both benefit from detailed structural information at the crown and canopy level.

Reliable Data for Ecological Applications

Stable image acquisition is critical for any ecological monitoring workflow. High-quality hyperspectral data improve the reliability of spectral analysis, feature extraction, classification models, and vegetation parameter inversion. In this case, the hyperspectral imaging workflow provided a strong data foundation for both hyperspectral camera fine species recognition and leaf area index retrieval using hyperspectral imaging.

Data Acquisition and Preprocessing for Hyperspectral Mangrove Ecological Monitoring

A robust workflow begins with appropriate data acquisition planning and preprocessing. In coastal wetlands, timing and environmental conditions can significantly affect the quality of remote sensing data.

UAV Flight Planning and Survey Design

The aerial survey was carried out in December 2024 during low tide, reducing the influence of tidal water on image acquisition. The study area was divided into six sections to improve survey efficiency and data consistency. The UAV flight parameters were set to ensure both broad coverage and fine spatial detail.

The hyperspectral imaging system used in this study was Ramoton RT-FS60. During data collection, the platform achieved a ground resolution of approximately 4.71 cm under a flight altitude of 100 m, with a flight speed of 5 m/s and a main flight line angle of 87°.

Radiometric Correction and Image Generation

The raw hyperspectral data were processed in ENVI 5.3 for radiometric correction in order to reduce the influence of atmospheric effects, illumination conditions, and solar angle variation. After that, geometric registration and strip mosaicking were completed to generate a complete hyperspectral image of the mangrove reserve.

Mangrove Area Extraction

To isolate mangrove vegetation from non-target surfaces, the researchers applied an NDVI threshold method using a range of 0.53 to 1.00. This process masked water bodies, saline marshes, and artificial surfaces. The result was then refined through manual interpretation, ensuring accurate extraction of the mangrove area for subsequent analysis.

This preprocessing workflow was essential for improving the performance of hyperspectral camera tree species identification and hyperspectral leaf area index retrieval.

Fine Tree Species Identification with a Hyperspectral Camera

After preprocessing, the research focused on fine tree species identification with a hyperspectral camera, which is one of the core goals of the study.

Multi-Scale Segmentation and Feature Construction

The processed hyperspectral imagery was analyzed using an object-oriented multi-scale segmentation method. The optimal segmentation scale was determined to be 25, with a merging scale of 80, effectively reducing the risk of over-segmentation or under-segmentation.

A total of 323 features were constructed from the data, including:

Original spectral band features
Vegetation indices
Texture features
Shape features
LiDAR-derived height features
LiDAR-derived intensity features

This high-dimensional feature set provided a strong basis for hyperspectral camera-based fine tree species identification in mixed mangrove stands.

Feature Selection and Deep Learning Classification

After correlation analysis removed 273 highly redundant features, the remaining variables were screened using:

RF-RFE
XGBoost
LightGBM

The intersection of the selected results generated a 14-dimensional optimal feature set, including sensitive spectral, structural, and canopy-related variables.

These features were then input into the Swin-UperNet deep learning model to classify the dominant mangrove species. This workflow enabled accurate discrimination among Kandelia obovata, Aegiceras corniculatum, and Avicennia marina, showing the practical value of hyperspectral camera fine species recognition in ecological applications.

Tree Species Identification Results

The overall classification accuracy reached 91.18%, with a Kappa coefficient of 0.85, indicating strong agreement between model predictions and field conditions.

The producer’s accuracy for each species was:

Aegiceras corniculatum: 92.08%
Avicennia marina: 91.38%
Kandelia obovata: 90.66%

These results confirm that fine tree species identification with a hyperspectral camera can provide accurate and actionable species maps for mangrove ecosystem monitoring.

Leaf Area Index Retrieval Using Hyperspectral Imaging

Based on the species classification results, the next step was leaf area index retrieval using hyperspectral imaging, which is another central focus of this study.

Species-Specific LAI Retrieval Strategy

Rather than applying a single model to all mangrove vegetation, the researchers selected sensitive features separately for each species. This species-specific strategy improved the accuracy of hyperspectral camera leaf area index retrieval, because canopy structure and spectral response vary among different mangrove species.

Model Development and Comparison

The selected features were incorporated into three LAI estimation algorithms:

SVR
PLSR
BPNN

By combining different feature selection methods with these algorithms, the research team built nine hybrid cross-models and compared their retrieval performance.

The optimal models were identified as follows:

RF-RFE + SVR for Kandelia obovata
RF-RFE + SVR for Avicennia marina
XGBoost + BPNN for Aegiceras corniculatum

This approach significantly improved hyperspectral leaf area index retrieval performance at the species level.

LAI Retrieval Results

The retrieved LAI values were highly consistent with field measurements.

Estimated LAI ranges were:

Kandelia obovata: 0.897–3.001, mean 1.416
Aegiceras corniculatum: 0.934–1.919, mean 1.355
Avicennia marina: 0.913–2.221, mean 1.443

Among the models, the result for Avicennia marina was especially strong:

R² = 0.87
MAE = 0.02
RMSE = 0.04

For Kandelia obovata, the model achieved:

R² = 0.81
MAE = 0.09
RMSE = 0.05

These results demonstrate that leaf area index retrieval with a hyperspectral camera can produce reliable and precise canopy information for ecological analysis and restoration monitoring.

airborne hyperspectral camera for vegetation monitoring

UAV hyperspectral imaging for mangrove monitoring

tree species classification using hyperspectral imaging

Ecological Value of Hyperspectral Mangrove Ecological Monitoring

The significance of this study goes beyond technical accuracy. It also shows how hyperspectral mangrove ecological monitoring can support practical ecosystem management and scientific decision-making.

Revealing Spatial Patterns in Mangrove Condition

The study revealed a clear spatial trend in mangrove LAI across the reserve, showing a pattern of decreasing from north to south, then increasing, and decreasing again. Such spatial information provides valuable insight into vegetation condition, habitat variation, and ecological function.

Improving Ecological Interpretation with Multi-Source Features

Another important finding was the combined contribution of height features and intensity features to LAI estimation. This confirms that integrating spectral information with structural characteristics can enhance the performance of hyperspectral leaf area index retrieval and improve the interpretation of mangrove canopy dynamics.

Supporting Conservation and Restoration

By combining fine tree species identification with a hyperspectral camera and leaf area index retrieval using hyperspectral imaging, researchers and conservation managers can obtain more accurate information for:

Mangrove conservation planning
Restoration effectiveness assessment
Vegetation health monitoring
Wetland resource management
Long-term ecosystem observation

This makes hyperspectral remote sensing a highly valuable tool for coastal ecological protection.

Product Information for Airborne Vegetation Monitoring

For airborne vegetation monitoring applications, the system used in this study was Ramoton RT-FS60. Its core specifications include:

Spectral range: 400–1000 nm
Spectral resolution: 2.5 nm
Spectral bands: 1200
Spatial pixels: 1920

These technical features make hyperspectral imaging highly suitable for:

Fine tree species identification
Leaf area index retrieval
Mangrove ecological monitoring
Wetland vegetation analysis
Forest resource surveys
Crop growth assessment

Conclusion

This study in the Zhangjiang Estuary Mangrove National Nature Reserve shows that hyperspectral remote sensing is highly effective for fine tree species identification with a hyperspectral camera, leaf area index retrieval with a hyperspectral camera, and hyperspectral mangrove ecological monitoring.

With an overall species classification accuracy of 91.18% and LAI retrieval error as low as 0.04, the workflow demonstrates the strong potential of UAV hyperspectral imaging in complex coastal ecosystems. It provides precise and efficient data support for mangrove conservation, restoration, and management, while also offering broad application potential in forestry, wetlands, and vegetation monitoring.