10 m Forest Height Map of China

The product was generated using a terrain-insensitive forest height mapping framework that integrates slope-adaptive relative height metrics, Sentinel-1/2 predictors, and a U-Net deep learning model. To reduce terrain-induced biases in GEDI relative height metrics over mountainous areas, slope-adaptive relative height metrics were derived from GEDI waveforms and used as improved training labels.

Region-specific high-greenness Sentinel-1/2 composites were constructed to reduce phenological inconsistencies across different natural regions of China. The resulting forest height product represents canopy height in meters and is designed to provide spatially continuous and terrain-insensitive forest structure information across China. Independent airborne laser scanning data covering more than 300 km² were used to evaluate the dataset, demonstrating improved accuracy and reduced slope-dependent errors compared with products trained using original GEDI RH metrics.