Vegetation canopy height retrieval in complex mountainous regions based on data calibration and CNN model

Vegetation canopy height is a vital metric for quantifying aboveground biomass, evaluating ecosystem integrity, and informing forest management strategies, playing a pivotal role in climate change mitigation and sustainable forestry. However, accurate estimation of canopy height faces three major challenges: topographic complexity in mountainous regions, landscape spatial heterogeneity, and inherent limitations of spaceborne LiDAR systems (e.g., GEDI) in detecting low-stature vegetation. This study proposes a regional calibration approach for GEDI-derived height labels and integrates the calibrated labels with multi-source remote sensing data to retrieve canopy height at 10 m resolution. Key findings demonstrate that: (1) The calibration model that couples Random Forest Regression and Multiple Linear Regression can effectively mitigate GEDI’s overestimation of low-stature vegetation and provide reliable label data for subsequent canopy height mapping; (2) Among all feature combinations, optical imagery emerged as the most influential contributor to model performance, significantly improving predictive accuracy when included; (3) The optimized model achieves a mean absolute error (MAE) of 2.26 m against field measurements, showing strong consistency with airborne LiDAR validation (MAE = 2.63 m). Qualitative evaluations highlight the model’s heightened sensitivity to diverse land cover types, enabling more precise delineation of vegetation boundaries. The calibrated height labels further improve the characterization of canopy height across forest, shrub, and grassland areas. These results demonstrate the potential of GEDI label calibration combined with multi-source remote sensing data for large-scale canopy height mapping in complex mountainous regions.