Research Analysis of the Joint Use of Sentinel-2 and ALOS-2 Data in Fine Classification of Tropical Natural Forests

Qiqi Xie, Wenxue Fu, Weijing Yan, Jiankang Shi, Chengzhi Hao, Hui Li, Sheng Xu, Xinwu LiForests 2025 · DOI

TL;DR

This study demonstrates that combining Sentinel-2 optical data with ALOS-2 SAR data, along with elevation information, significantly improves the fine classification of tropical natural forests. The XGBoost machine learning model achieved the highest accuracy (0.89 overall, 0.82 Kappa), highlighting the power of integrating diverse remote sensing data for ecological monitoring.

What Problem Does This Paper Address?

Tropical natural forests are vital for global climate regulation and ecosystem health, but they face increasing threats from human activities and climate change. Accurately mapping and classifying these forests at a fine scale is crucial for effective conservation and management strategies. Traditional field surveys are labor-intensive and impractical for large-scale monitoring. While remote sensing offers a solution, single data sources (like optical or SAR) have limitations: optical data is hampered by cloud cover and limited spectral information, while SAR data can be noisy and difficult to interpret. This paper addresses the challenge of achieving accurate and detailed classification of diverse tropical forest types by exploring the synergistic use of multi-source remote sensing data and advanced machine learning techniques.

Methods

The researchers employed a multi-stage approach:

  1. Data Preprocessing: Sentinel-2 optical data underwent atmospheric correction and cloud masking. ALOS-2 PALSAR data was terrain-corrected and radiometrically calibrated. SRTM DEM data was resampled to match the resolution of the other datasets. All data was aligned to a 10m spatial resolution.
  2. Feature Construction: A comprehensive set of features was engineered, including:
  3. Spectral Features: Reflectance values from Sentinel-2's visible, near-infrared, and short-wave infrared bands.
  4. Vegetation Indices: Standard indices like NDVI, NDWI, EVI, and others like RENDVI and DVI were calculated.
  5. Texture Features: Derived from optical and SAR data using the Gray-Level Co-occurrence Matrix (GLCM) to capture spatial patterns.
  6. Backscattering Features: HH and HV polarization backscatter coefficients from ALOS-2, along with derived Radar Vegetation Index (RVI) and polarization difference.
  7. Topographic Features: Elevation, slope, and aspect derived from SRTM DEM.
  8. Feature Selection: Spearman's correlation was used to identify and remove highly correlated features, reducing redundancy and improving model efficiency. Feature importance was then assessed using a Random Forest model.
  9. Model Training and Evaluation: Four machine learning algorithms were trained and compared: Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), and Extreme Gradient Boosting (XGBoost). A stratified random sampling approach was used to divide the data into training (80%) and validation (20%) sets. Model performance was evaluated using metrics like overall accuracy, F1 score, recall, and Kappa coefficient, along with confusion matrices.

Datasets

  • Sentinel-2 MSI: Multispectral optical satellite data (10m resolution) providing spectral information.
  • ALOS-2 PALSAR: L-band Synthetic Aperture Radar data (10m x 10m resolution) providing backscattering and structural information, less affected by weather.
  • SRTM DEM: Shuttle Radar Topography Mission Digital Elevation Model (30m resolution) providing topographic information.
  • Field Survey Data: 180 ground-truth sample points collected on Hainan Island, covering various natural forest types, used for training and validation.

Key Insights and Ideas

  • Synergy of Optical and SAR is Crucial: The study strongly supports the idea that combining optical and SAR data significantly enhances classification accuracy and boundary delineation compared to using optical data alone. SAR's ability to penetrate canopy and its weather independence complements optical data's spectral richness.
  • XGBoost Dominates: Among the tested algorithms, XGBoost consistently outperformed RF, SVM, and LR, demonstrating its superior ability to model complex, non-linear relationships between features in challenging tropical forest environments.
  • Topography is King (for this region): Elevation emerged as the most critical feature for classifying most tropical forest types, underscoring its significant influence on hydrothermal conditions and, consequently, forest distribution.
  • Red-Edge Bands Matter: The red-edge spectral bands (specifically B5) were found to be important for distinguishing evergreen broadleaf forests, suggesting their sensitivity to specific vegetation characteristics.
  • Mangroves are Distinct: Mangrove forests showed exceptionally high classification accuracy across all models, indicating their unique spectral and structural characteristics make them easily distinguishable from other forest types.

Key Results

  • Highest Accuracy: The XGBoost model achieved an overall accuracy of 0.89 and a Kappa coefficient of 0.82.
  • Feature Importance: Elevation was the most important feature, followed by red-edge spectral bands (B5) and short-wave infrared bands (B12).
  • Forest Distribution Patterns: The study revealed distinct elevational distributions: mangroves at the lowest elevations, tropical rainforests at middle and low elevations, and tropical coniferous forests tending towards middle to high elevations.
  • Model Performance Ranking: XGBoost > RF > SVM > LR in terms of overall classification accuracy.
  • Joint Classification Improvement: The integration of optical and SAR data led to improved boundary clarity and reduced misclassification for various forest types, particularly in detailed plot analysis.

Limitations and Future Work

  • Limited Machine Learning Models: The study used four classical machine learning models. The authors acknowledge that deep learning models (like CNNs) could potentially further improve accuracy, especially with high-resolution imagery.
  • Data Scale: The current sample data is at a point scale. Deep learning semantic segmentation typically requires high-precision, manually delineated training samples, which are not readily available for tropical natural forests.
  • Future Directions:
    • Incorporate deep learning techniques (e.g., CNNs) with high-resolution imagery.
  • Develop new individual or hybrid models to leverage the strengths of different algorithms.
  • Expand the study to other tropical regions and explore the integration of seasonal dynamics and time-series data for enhanced temporal generalization.
  • Investigate the use of hyperspectral data.

Summary updated

Abstract

Tropical natural forests play a crucial role in regulating the climate and maintaining global ecosystem functions. However, they face significant challenges due to human activities and climate change. Accurate classification of these forests can help reveal their spatial distribution patterns and support conservation efforts. This study employed four machine learning algorithms—random forest (RF), support vector machine (SVM), Logistic Regression (LR), and Extreme Gradient Boosting (XGBoost)—to classify tropical rainforests, tropical monsoon rainforests, tropical coniferous forests, broadleaf evergreen forests, and mangrove forests on Hainan Island using optical and synthetic aperture radar (SAR) multi-source remote sensing data. Among these, the XGBoost model achieved the best performance, with an overall accuracy of 0.89 and a Kappa coefficient of 0.82. Elevation and red-edge spectral bands were identified as the most important features for classification. Spatial distribution analysis revealed distinct patterns, such as mangrove forests occurring at the lowest elevations and tropical rainforests occupying middle and low elevations. The integration of optical and SAR data significantly enhanced classification accuracy and boundary recognition compared to using optical data alone. These findings underscore the effectiveness of machine learning and multi-source data for tropical forest classification, providing a valuable reference for ecological monitoring and sustainable management.
BibTeX
@inproceedings{xie2025research,
  author = {Qiqi Xie and Wenxue Fu and Weijing Yan and Jiankang Shi and Chengzhi Hao and Hui Li and Sheng Xu and Xinwu Li},
  title = {{Research Analysis of the Joint Use of Sentinel-2 and ALOS-2 Data in Fine Classification of Tropical Natural Forests}},
  booktitle = {Forests},
  year = {2025},
  doi = {10.3390/f16081302},
}
Figure