Abstract
Tree canopy height is crucial for policymaking, especially in climate change mitigation. Current global tree canopy height estimation research shows limited predictive accuracy due to insufficient LiDAR data as ground-truth. The lack of comprehensive and up-to-date LiDAR data obstructs accurate global canopy height estimations with training coupled with remote sensing images. Additionally, machine learning (ML) models struggle with domain gaps, performing well in one location but poorly in another. This research aims to use large-scale synthetic rendering to create imitation environments with accurate tree canopy height measurements. A semi-automated workflow simulates diverse environments by modifying tree type, height, leaf density, lighting, and seasonal variations. Generated mimicked satellite images, Digital Surface Models (DSM), and Digital Terrain Models (DTM) provide ground truth data to train ML algorithms, improving global canopy height estimations. This will enhance global carbon stock assessments, climate change mitigation, land use management, and conservation efforts.