The research problem is formulated around one key question: how can we combine satellite imagery, GEDI LiDAR, super-resolution, and deep learning to generate more accurate forest-structure information, especially canopy height, which is essential for carbon stock estimation?
Our framework integrates multi-source data, including Sentinel-1, Sentinel-2, GEDI LiDAR, and UAV/drone imagery. A DSen2-based super-resolution model improves Sentinel-2 spatial quality, while a GEDI-guided deep learning model estimates dense canopy height maps at 10 m resolution. The system also supports biodiversity mapping by connecting remote sensing data with vegetation and habitat information.
The main objectives are to improve satellite data quality, enhance canopy height mapping, support carbon-relevant estimation, and move toward a national environmental intelligence platform for Vietnam.
The achieved results show clear improvements in super-resolution quality and canopy height estimation. The proposed model reduces prediction errors compared with previous methods and demonstrates the value of direct GEDI guidance. Overall, this research contributes to smarter, data-driven forest monitoring for carbon management, biodiversity assessment, and sustainable environmental planning in Vietnam.