Forest mapping is essential for sustainable forest management and climate adaptation, enabling the assessment of forest composition and condition to prevent degradation. This study developed a U-Net-based deep learning framework for forest type classification using Sentinel-2 MSI satellite imagery and vegetation indices that capture seasonal canopy properties. A two-step approach was adopted, first delineating forested areas and then classifying forest types into needleleaf, broadleaf, and mixed forests. The forest area classification model achieved an overall accuracy of 0.958 (Kappa = 0.916), confirming reliable separation of forest and non-forest areas. For forest type classification, incorporating multi-seasonal imagery consistently enhanced performance, with the NDVI-based model achieving the highest overall accuracy of 0.831 (Kappa = 0.698). These results highlight the importance of integrating multi-seasonal spectral information to capture canopy variability and improve classification accuracy. The resulting reproducible framework thus supports ecosystem monitoring, hazard assessment, and adaptive forest management, offering foundational data for near real-time resource management under changing climatic conditions.
Chang et al. (Mon,) studied this question.