An improved forest canopy height mapping method by combining GAN-enhanced optical imagery, SAR, and GEDI data

Forest canopy height (FCH) is a crucial parameter for characterizing forest structure and ecosystem dynamics. NASA’s Global Ecosystem Dynamics Investigation (GEDI) has greatly facilitated tree height estimation by providing accurate vertical structure measurements. However, the sparse spatial distribution of GEDI observations remains a significant constraint for large-scale FCH mapping. Although optical remote sensing imagery can compensate for such spatial discontinuities, its accuracy is often hindered by the presence of clouds, haze, and sensor interference. Additionally, in densely vegetated areas, optical data are susceptible to saturation effects, frequently resulting in the underestimation of canopy height. In this study, an improved FCH mapping method by combining GAN-enhanced optical Imagery, SAR, and GEDI data is proposed, which consists of three main steps: (1) cloud removal in optical imagery using a Generative Adversarial Network (GAN), which reduces the impact of cloud and fog on FCH estimation; (2) feature extraction from enhanced optical imagery, SAR, topographic and GEDI data; (3) FCH mapping using machine learning, and accuracy assessment. The proposed method maintained high accuracy in estimating FCH in regions with frequent cloud cover, particularly where continuous cloud-free imagery is difficult to obtain. Using Washington State, USA, as the study area, the method improved the footprint-level performance compared to optical data with approximately 20 % cloud cover, increasing R² from 0.60 to 0.73 and reducing RMSE from 5.45 m to 4.53 m. At the mapping stage, model validation against UAV-derived canopy height models (CHM) yielded an R² of 0.61 and an RMSE of 5.31 m.