Terrestrial laser scanning (TLS) has been demonstrated to capture in-situ forest structure with enough detail to potentially reflect 3D structural changes from disturbances (e.g. fire, logging, drought). However, the ability to utilize these spatially explicit data to analyze the interactions between forest structure and disturbance in the context of global change is still underdeveloped. Quantitative structure models (QSMs) have been instrumental in furthering the ability of LiDAR data to accurately and easily quantify tree structure and aboveground biomass. Despite this, multiple algorithms and versions for QSM reconstruction exist with no standardized procedure for benchmarking or validation.
Here, we utilized the co-incident TLS and destructive harvesting of trees dataset from multiple temperate species in Belgium to compare and benchmark multiple different QSM reconstruction algorithms and versions. The objectives of this work are to validate current and past research utilizing different versions of the software TreeQSM, provide direction for future development of QSMs, and establish a standard for future validation. The dataset consists of 65 adult trees of 4 different species (2 conifer and 2 broadleaf) across 5 different sites throughout Belgium. Diameter at breast height and tree height were obtained from the TLS point cloud, and stem volume and total volume were obtained using TreeQSM v2.0, v2.3, and v2.4. These values were then compared to each other and to destructive sampling values. Benchmarking of different QSM algorithms will lead to increased data-interoperability of these relatively new 3D structural measurements.

Creating accurate 3D models of trees, from points clouds generated with laser scanning, is becoming fundamental to advancing our understanding of tree structure and function, and quantification of forest ecosystem services. Quantitative Structure Models (QSM) can estimate the volume of the woody parts of trees, by fitting cylinders representing different parts of the woody skeleton identified in the point cloud. While QSMs, reconstructed with methods such as TreeQSM, are becoming a standard approach there are also other approaches for generating tree volume or biomass from a point cloud, including the Outer Hull Model (OHM). With multiple approaches and limited high-quality reference data, there is still a lot of uncertainty regarding how accurate these various approaches are. Here, we examine leaf-off scans of deciduous trees of two different species (Acer rubrum and Quercus rubra) paired with highly-detailed destructive sampling data to determine differences in volume and biomass estimation for the oldest (2.3.0) and newest versions (2.4.1) of TreeQSM and compare them to estimates from the OHM. The results reveal large differences in branch volume and between the two different versions of Tree QSM and suggest that the newer version exacerbates branch volume estimation errors, which appear endemic to any a cylinder-fitting approach, especially for small branches. The OHM provided superior estimation of tree volume and biomass for branches and similar accuracy for main stems, then TreeQSM, but the OHM method lacked the ability to describe detailed topology of tree branching architecture. Also discussed are causes of volume differences between TreeQSM versions and OHM and implications for corrective measures to eliminate bias in QSMs.

Quantitative structure models (QSMs), derived from point clouds generated by terrestrial laser scanning (TLS), are used to generate topologically accurate and detailed models of individual trees, comprised of cylinders representing their woody parts. Occlusion, beam divergence, and software registration errors are technical limitations inherent to current TLS technology that prevent fine details, such as smaller branches and twigs, from being accurately resolved. These technical limitations contribute to overestimation of the size of smaller branches and twigs when QSMs attempt to fit cylinders to them. This can cause large overestimations of tree branch volume or surface area, especially for smaller trees. Many attempts have been made to correct such problems in QSMs, e.g., by including some type of “allometric” correction, where anomalous cylinders are adjusted based on some type of volume or taper model fitted to data generated by the QSM. However, since QSMs generally contain few or no accurately modeled small branches (they might all be too big), such model-based corrections are necessarily biased. Here, we demonstrate a new algorithm (RealTwig) that corrects overestimated and poorly fit cylinders in the popular TreeQSM modeling software, by including external information describing realistic twig sizes for the tree species being modeled. We measured the twig diameters of common North American tree species to determine what the corresponding cylinder diameters of twigs in a QSM should be, and then modeled the taper of every path in the tree’s branching network, forcing the model intercept to be the species’ minimum twig diameter. To test the accuracy of our algorithm, we converted QSM volumes to mass using reference wood density values and compared the estimated mass against high-quality reference destructive sampling data. We found that our algorithm dramatically improved small branch and twig mass estimation and generally eliminated bias, compared to currently available versions of TreeQSM.

Diameter at breast height (DBH) serves as a key attribute for a qualitative and quantitative evaluation of a tree and a forest in general. Thus, DBH belongs to the most frequently derived tree attributes in forestry and is included to measurement protocols of every national forest inventory (NFI). Recently, terrestrial LiDAR data has been actively investigated for extracting various forest features including DBH. So far, various approaches have been developed. However, they often require an extensive pre-processing and filtering of the point cloud and rely on an expert-based iterative parameter optimisation. This results in a low level of automation and high time consumption. Recent advances in the field of computer vision show that deep learning methods are often highly efficient in extracting meaningful features from complex real-world data.

We propose a fully automated deep-learning-based approach for a fast and accurate tree position and DBH retrieval from dense terrestrial LiDAR data. The approach relies on plot-level point clouds and requires very little data pre-processing. The backbone of the approach is an existing deep learning model (YOLOv5) retrained on our data. In addition to real-world terrestrial (TLS) and mobile (MLS) laser scanning data, we enhanced our training dataset with synthetic point clouds of simulated forest stands. The final model was tested on a “blind” dataset including TLS and MLS point clouds with reference data available (DBH and tree position measured by experts in the field using calipers and a total station).

First results reveal high performance of our approach both in speed and in accuracy. The processing time for a forest plot was reduced to several minutes and a high tree detection rate was achieved. Thus, our approach can be used as an efficient tool for tree position and tree diameter retrieval from terrestrial LiDAR data.

Descriptions of vertical canopy structure are widely used in global climatic, ecological, hydrologic and biodiversity studies and required for input to dynamic global vegetation models. Two spaceborne laser altimetry missions - GEDI and ICESat-2 - have been in operation between 2018 and 2023, however of studies published between 2021 and 2022 that compared their canopy measurement performance, none separated the impact of geolocation accuracy and measurement error. GEDI implemented a pointing knowledge system that met its geolocation requirement of 10m (1𝜎) and while this level of geolocation accuracy is not a source of error for the GEDI 1-km gridded products, colocation of footprint level measurements with reference data is necessary since canopy structure may vary substantially over short distances. Direct comparison of GEDI and ICESat-2 is also limited by differences in canopy height definition, footprint size and shape, and measurement approach. Here we present a global, spatially explicit, comparison of GEDI and ICESat-2 canopy height measurement performance from 2019 to 2023. Measurement errors were quantified through colocation of individual beams with equivalent reference values simulated from a global database of community sourced ALS surveys, and uncertainty related to the orbital sampling design were quantified with an estimator that accounts for the cluster-based sampling. We found that ICESat-2 alone will not meet GCOS ECV biomass product coverage or height accuracy requirements - for example 68% of 1-km cells across the Amazon Basin were within 2-m standard error of the mean canopy height compared to 6% from ICESat-2 - however does provide observations that extend the applicability of current spaceborne laser altimetry measurements to savanna as well as boreal ecosystems. We will summarize the implications of this finding for the fusion of GEDI and ICESat-2 measurements in large area mapping and monitoring and considerations for a future spaceborne laser altimetry mission.