This study proposes a method to improve the accuracy of ground elevation estimations by analyzing Global Ecosystem Dynamics Investigation (GEDI) waveforms. The impact on this new method on canopy-height and above-ground biomass (AGB) estimations was also evaluated. The method uses a deep learning model to derive ground elevations from GEDI waveform data. Geographic transferability was assessed by recalculating the accuracy of canopy-height and AGB estimates using the improved ground height while retaining the original GEDI formulas for relative height and AGB. GEDI waveform data with airborne laser scan (ALS) data from three regions in Japan were integrated, and transfer learning was applied to enhance the estimation accuracy in regions excluded from the training dataset. The results demonstrated that the deep learning-based ground elevation estimation both reduced the root mean squared error (RMSE) by >3 m compared to the previous GEDI L2A product and exhibited generalization performance. Further accuracy improvements were achieved through transfer learning and retraining with additional data, even with limited datasets. These findings demonstrate that improving the accuracy of ground elevation estimations can significantly optimize the GEDI global AGB estimation algorithm, achieving a canopy-height accuracy of approx. 1.5 m and AGB accuracy of approx. 70 t/ha. The method proposed herein is expected to improve the accuracy of GEDI ground elevation and AGB estimations outside of the study area, and it will be used for the Japanese space lidar mission MOLI scheduled for launch in 2027. The new method is expected to contribute to global and long-term high-precision AGB observations.

Forests play an important role as carbon dioxide sinks and are said to contribute greatly to mitigating climate change. In recent years, with the development of Light Detection and Ranging (LiDAR) technology, it has become common to observe forest height and structure with high accuracy. Global Ecosystem Dynamics Investigation (GEDI), a space-based LiDAR system for vegetation observation, has been observing forest structure at the global level since 2019. Methods based on deep learning have been reported for the analysis of GEDI-received waveforms, in addition to conventional methods based on peak fitting. In tropical forests in particular, a major problem is the lack of detailed, wide-area, three-dimensional structural data that can be used to teach or validate GEDI waveform analysis. In deep learning image processing, data augmentation has been used when the amount of training data is small. In this study, we use waveform simulation results using relatively easy-to-obtain point cloud data from Japan as training data to build a deep learning machine with ground surface estimation performance for tropical evergreen forests, which are completely different forest types, using data augmentation. Assuming that no training data for tropical evergreen forests can be obtained, we created a data augmentation method for satellite LiDAR waveforms based on the concept of image data augmentation, and analyzed whether this method is effective even when the amount of training data is very small. Our method significantly improved both the average error and the root mean square error (RMSE) of ground surface estimation, even in tropical evergreen forests in the central Amazon. Furthermore, even when the amount of training data was reduced to 1/100, the present data augmentation method achieved the same estimation accuracy. Our method can greatly improve the accuracy of ground height estimation from satellite LiDAR waveforms even when detailed point cloud data that serve as training data are not available. Hence, it is highly useful for analyzing the three-dimensional structure of forests.

Satellite light detection and ranging (LiDAR) technology enables efficient large-scale monitoring of forest canopy structure, which is crucial for understanding ecosystem dynamics and carbon cycling. The Global Ecosystem Dynamics Investigation (GEDI) mission provides unprecedented global coverage of forest canopy height measurements. However, GEDI-derived canopy height estimates and their spatial characteristics are not yet fully understood. In this study, we demonstrated that GEDI canopy height estimates exhibited significant spatial heterogeneity in their accuracy when compared to high-resolution airborne LiDAR data over the city of Nikko, Japan. We found that GEDI relative height metrics (RH98) correlated with airborne LiDAR canopy heights (r＝0.31), with mean absolute errors of 6.9 m. Importantly, we revealed substantial spatial variability in estimation errors using geographically weighted error metrics. Large overestimation errors were found in flat areas dominated by evergreen conifers, while underestimation occurred in steep terrain with deciduous conifers. Areas with deciduous broadleaf forests showed relatively small errors. These spatial patterns in accuracy were not captured by conventional global error metrics, highlighting the importance of local context when interpreting GEDI canopy height results. Our findings clarify how satellite LiDAR performs according to forest type and terrain, which provides crucial insights for improving global forest structure monitoring. This study introduces a novel spatial error assessment framework for satellite-derived forest metrics that can enhance our understanding of uncertainties in forest monitoring.

Forests play an important role in mitigating and adapting to global warming, and forest quality may be more important than the amount of forest. Therefore, there is global demand to create a method to monitor canopy height and forest vertical structure. Satellite Light Detection and Ranging (LiDAR) with satellite imagery can produce a map that describes the forest quality. High-accuracy mapping is expected to be achievable through deep learning. However, over-training may cause overfitting. In this study, wall-to-wall forest canopy height maps were created by fusing Global Ecosystem Dynamics Investigation Level 2A (GEDI L2A) data on canopy height (rh_095) and PlanetScope satellite images, to create data for use by a Convolutional Neural Network (CNN)-based deep learner. An early stopping method was introduced in this study to stop training before the learning process starts to produce large canopy height errors. In order to avoid inappropriate accuracy assessment due to bias in the validation sites, forest canopy height maps were validated for the entire area. The true value was taken as the result of satellite LiDAR waveform simulation using Airborne Laser Scanning (ALS) point cloud data. The tree height estimation error in the present study area in the Izu Peninsula had a Root Mean Square Error (RMSE) of 4.94 m; that is, the estimation error was significantly reduced by more than 1.7 m. Initially, 25% of the training data used in this study contained large height errors, however, our early stopping method successfully eliminated the pre-screening process of the error data . Furthermore, external reference data were no longer required during data screening. Our proposed method also automatically determines when to stop learning. This method could save a significant amount of the time and computational resources required to produce high-accuracy global tree height maps.