The Global Ecosystem Dynamics Investigation (GEDI) is a geodetic-class laser altimeter that has provided over 20 billion high quality observations of vegetation vertical structure over the Earth’s temperate and tropical forests. These are the first set of spaceborne measurements from an instrument specifically designed to recover vegetation structure and they form the basis of critical reference data sets including canopy height, canopy cover, canopy vertical profile, bare earth topography, and biomass, among others. GEDI became operational in April 2019 on the International Space Station for what was originally planned as a 1-year mission but was extended to nearly 4 years. GEDI was shut down in March 2023 and is in hibernation for the next 13-18 months at which point it is expected to become operational again in Fall of 2024 to start a new era of forest structure measurements that may last through 2030. This pause in data collection provides an ideal opportunity to assess the mission’s impact during its first observing epoch. In this paper we take a comprehensive look at the mission’s accomplishments and the science its data has enabled. We first provide a brief history of the mission and summarize the status of its data products. Next, we examine some results from the mission, both in terms of its observational archive, but also with respect to the wide-ranging applications of its data, from biomass to biodiversity to fusion and beyond. We then describe our activities during the hibernation period that include improved algorithms as well as the creation of new, mission data products. Lastly, we look ahead to GEDI’s next observational epoch and explore the exciting possibilities that a 10-year time-series of GEDI data could enable.

Tropical savanna vegetation is inherently challenging to model. Terrestrial LiDAR (TLS), in combination with quantitative structure modelling (QSM), can help us better understand structural dynamics; however, the cost point of current best practice hardware is prohibitive for many users outside elite research institutes. In this study, we investigate the utility of an entry-level TLS scanner and reduced model complexity to encourage wider adoption of TLS. In tropical savannas, frequent disturbance leads to a high carbon turnover and strong spatiotemporal heterogeneity. At the individual tree scale, structural heterogeneity destabilises allometric relationships and exacerbates uncertainty in field inventories. However, the high acquisition cost of alternative high-grade TLS scanners commonly used in ecological surveys and the high computing power required to apply QSMs to large numbers of trees present a barrier to their adoption in applied spaces such as land management or carbon accounting. We tested the utility of a Leica BLK360 scanner and simpler mathematical models for the targeted extraction of ecologically meaningful tree metrics. Using all trees within a 1 ha plot, we first confirmed the accuracy of BLK360 measurements by testing against field-based diameter at 1.3 m (DBH). We then identified an increasing heterogeneity in crown size with increasing tree height as a likely source of the uncertainty in DBH-based allometric models. We proposed individual plant-occupied space (POS) as an alternative metric able to account for crown structure and investigated a link between changes in POS and stand-scale carbon flux. We identified voxel models as an efficient alternative for calculating POS compared to QSM and enabled upscaling to airborne LiDAR using allometry based on tree height and crown area, dramatically increasing the area that can be surveyed. Our findings highlight how LiDAR technology for mapping, monitoring and modelling savanna vegetation and carbon sequestration can be accessible to a wider audience.

Forest stand characteristics influence species composition and trophic relationships. However, the understanding of this dependence is still limited. Recent advances in the field of 3D remote sensing allow a detailed characterization of complex forest structures. In conjunction with image-based technologies, the assessment of the relative contribution of remotely sensed structural and physiological forest parameters to biodiversity composition and functioning could be achieved. This study aims to evaluate the use of different close-range remote sensing platforms, lidar and image-based, and different measurement setups for each platform for the estimation of forest parameters, primarily focusing on structural parameters.

The measurements are conducted on three neighboring patches of 15x15m on the Zürichberg in Zürich, Switzerland; one dominated by conifer trees, a mixed, and a deciduous patch. For each patch, terrestrial and UAV-based data is acquired under leaf-off and leaf-on condition. The terrestrial sensors are a handheld mobile laser scanner (GeoSLAM ZebHorizon) and a Leica BLK 360 terrestrial laser scanner, UAV-based sensors are a DJI Mavic 3M (image-based) and a DJI Zenmuse L1 (lidar). Within this study, only the RGB information from the multispectral Mavic 3M is used and processed to a 3D point cloud. A combination of the BLK 360 and the Zenmuse L1 acquisition is defined as the reference dataset as it represents the 3D forest structure the best. All other acquisitions and combinations are thereafter evaluated regarding their completeness against this reference dataset. It is further evaluated how the completeness may affect forest structure.

Preliminary results show that a combination of the BLK 360 and the structure from motion point cloud provide a good representation of the 3D forest structure. In a next step, the best sensor or sensor combination as found in this study will be analyzed regarding their suitability for AGB estimation and the potential to derive light availability.

3D radiative transfer models serve as vital tools for comprehending the interplay between solar radiation and forest canopies. However, obtaining the intricate canopy structural data necessary for these models is a challenging task. In this study, an alpha-shape based approach for canopy modeling was proposed, using airborne laser scanning (ALS) data to reconstruct 3D forest scenes. The proposed approach characterizes crowns with their intricate boundaries extracted through the alpha-shape algorithm and represents leaf clusters as a turbid medium, with leaf volume density being estimated from pulse intensity information. We assessed the accuracy and feasibility of the reconstructed scenes for radiative transfer simulations by utilizing the ALS point cloud simulated with the large-scale remote sensing and image simulation framework (LESS, http://lessrt.org/) and RAMI forest scenes. Specifically, we selected three RAMI scenes: a summer birch forest (HET09), a deciduous forest (HET51), and a savanna forest (HET50), for the ALS point cloud simulation and the scene reconstruction. We compared the images simulated with the original RAMI scenes to those reconstructed in this study. Furthermore, we included voxel-based and ellipsoid-based scene reconstructions for each scene. The results indicated that, for relatively sparse canopies, where individual trees are more easily identifiable (HET09 and HET50), the three approach have comparable accuracies with an R2 > .89. In contrast, for dense and continuous canopies (HET51), the ellipsoid approach has the worst accuracy, with an R2 value < .4. Of the three approaches, the voxel approach is particularly sensitive to voxel size. Larger voxels lead to decreased accuracy, limiting its usefulness in representing large-scale forest scenes. Conversely, the alpha-shape based approach proposed here strikes a better balance between simulation accuracy and computation efficiency. Furthermore, a GUI-based tool has been developed to facilitate the use of this approach. It can be accessed through the LESS model.

Airborne laser scanning (ALS) technology has revolutionised data acquisition and resource quantification in forestry. The widespread adoption of ALS-based technologies has generated an important need for processing software and scripts. In-house algorithms designed for specific user needs gradually gave way to the development of specialised, computationally efficient software, which contributed greatly to the uptake of the technology in practice. In a research context, however, the often closed-source nature of commercial software brings two important limitations when used for research purposes. First, the use of pre-established routines and algorithms limits the ability of researchers to explore, compare or even implement different methods of interest. Second, the use of closed-source algorithms implies the inclusion of ‘black boxes’ in researchers’ workflows.

The lidR package for R was first and foremost developed to overcome these issues in the specific context of a Ph.D. project (Roussel et al. 2017; 2018). Since then, the package has acquired a large community of users internationally. A full description of the main features and development philosophy is available in Roussel et al. (2020). Here, we look back on the development of the lidR package over the past five years from the perspective of the team that supervised the work. We present the lessons learned from this endeavour and make a call for researchers to keep contributing to this collective effort.

Despite the production of shared code being promoted and encouraged, source code often remains difficult or impossible to access in forestry studies. We advocate for a more systematic and comprehensive effort to sharing code, routines and workflows alongside the published results of ALS studies in forestry. Whereas shared code should ideally be efficient, implementable and well commented, even imperfect code is much preferable to ‘in-house-only’ algorithms. In our experience, such sharing can yield unforeseen benefits.