Canopy Height Mapping in Complex Temperate Forests Using Spaceborne GEDI LiDAR and Multitemporal Sentinel-2 Data

Forests with complex canopy structures pose significant challenges for accurate canopy height estimation, which is essential for carbon monitoring, conservation planning, and sustainable forest management. This study integrates multitemporal Sentinel-2 imagery and spaceborne GEDI LiDAR data using a random forest regression model to generate a high-resolution canopy height map for over two million hectares of structurally complex temperate forest. Model performance was evaluated using 2,543 national forest inventory plots with corrected geolocation to address spatial misalignment. Results show that multi-temporal Sentinel-2 composites significantly improved GEDI relative heights predictions compared to single-season inputs, increasing R² from 0.12–0.22 to 0.26–0.28. ALSO, GEDI quality filtering increased R² of multi-temporal models to 0.53–0.56. Spring and summer imagery contributed most to model accuracy due to enhanced vegetation activity and stronger spectral contrast during peak growing seasons. A novel geolocation correction approach based on a high-resolution canopy height model derived from a foundation model (Meta CHM) was implemented to adjust field plot locations in the absence of airborne LiDAR data. This method, based on a search radius to identify the best match between field-measured heights and Meta CHM values, substantially improved model performance, increasing R² from 0.12–0.21 to 0.21–0.45. Error analysis revealed that the model tends to underestimate higher canopy height values, likely due to spectral saturation in dense and highly stratified stands. Comparisons with existing global canopy height products showed that the locally calibrated model better captured structural variability, particularly in complex and multi-layered forests. By integrating seasonal spectral information and improving spatial alignment of reference data, this approach provides a practical and scalable framework for canopy height estimation in data-scarce regions, particularly where airborne LiDAR data are not available.