Potential Assessment of SAR and Optical Data with Machine Learning to Monitor Temporal Changes in Tall Vegetation

The global decline in tall vegetation (TV) due to environmental degradation necessitates effective detection methods for conservation and sustainable development goals. This study presents a novel machine learning (ML) approach for mapping TV cover changes at a pixel level (10 m) using information fusion of Synthetic Aperture Radar (SAR) and multispectral optical data. Demonstrated across the Roorkee-Dehradun Bypass in India, the approach integrates Sentinel-1 SAR and Sentinel-2 multispectral data, validated with field observations. Our method, which utilizes a total of fourteen features extracted from both data sources, was rigorously tested with six ML models. Random Forest notably outperformed other models, achieving the highest classification accuracy of 96.5% when fusing SAR and optical data. This performance is closely followed by the XGBoost and LightGBM models. The integration of both data types leverages complementary features, significantly enhancing model performance compared to using each data type individually. The analysis revealed a consistent decrease in TV from 2016 to 2018, aligning with observed changes in the study regions. By effectively identifying and classifying TV changes due to highway construction and urbanization, our approach provides a valuable tool for monitoring environmental degradation. This work highlights the potential of integrating SAR and optical data through information fusion to achieve more accurate and reliable change detection, offering a significant advancement in fine-resolution environmental monitoring and conservation management.