Introduction

Remote sensing has revolutionized our understanding of forests worldwide. Since the first Earth observation missions were launched in the 1970s, remote sensing has contributed to quantifying and monitoring forest extent1,2, as well as forest functioning3. From tropical to boreal forests, these technologies have enabled us to track habitat loss4, ecosystem health5, and carbon storage6 at unprecedented spatial coverage and analytical depth, becoming an essential tool in monitoring key planetary boundaries related to land system change and climate regulation7,8.

Over the past decades, the field of forest remote sensing has made fast advancements in both methodological and computational capabilities. The field has shifted from simply mapping landscapes to actively monitoring dynamic processes at increasingly near real-time9. The ability to analyze the physiological and structural characteristics of forests has rapidly become both technically and operationally feasible. Light Detection and Ranging (LiDAR, also known as laser scanning) sensors mounted on aerial platforms have added a new dimension to these capabilities. By emitting laser pulses to measure distances, LiDAR technology provides precise and accurate, three-dimensional measurements of forest canopies. Although the first applications of airborne laser scanning (ALS) in forest studies began in the 1980s10,11, its widespread adoption did not gain momentum until after the 2000s12. Since then, LiDAR has become the remote sensing gold standard for quantifying forest structural characteristics such as canopy height and cover, and topographical features of the forest floor.

Among the remote sensing tools available to scientists, none provides a more accurate and detailed view of forest structure than terrestrial LiDAR technology, also known as terrestrial laser scanning (TLS). Unlike airborne systems, TLS instruments are positioned at ground level, allowing them to capture detailed measurements of both the forest understory and the upper canopy. Compared to other ground-based methods, such as mobile laser scanning, terrestrial photogrammetry, or traditional inventory techniques, TLS offers superior geometric accuracy and structural completeness, particularly for detailed modeling of individual trees and stand structure.

The adoption of TLS is much more recent in comparison to other remote sensing tools, increasing quickly from around 2010 onwards (Table 1). The first review articles on TLS applications in forests appeared in 2011, offering foundational insights into using TLS to assess forest structure, including tree height, stem diameter, and biomass13. Subsequent reviews highlighted TLS’s ability to improve plot-scale forest measurements and estimate tree metrics14, as well as its broader applications across forest science disciplines15. More recent reviews have summarized key developments and future challenges in TLS, reflecting the growing and diversifying TLS research community16,17. However, with the fast adoption of TLS for studying vegetation, substantial progress has been made beyond these latest assessments.

Although many factors have contributed to the uptake of TLS in forest studies, three aspects stand out: price, speed, and size. While high-end TLS instruments—typically characterized by high ranging accuracy and long effective range—remain prohibitively expensive for many research groups, the availability of more affordable devices has substantially increased in recent years, making the technology more accessible. Recent instruments are not only lighter but also offer significantly faster point acquisition rates. In addition, modern scanning protocols have become more efficient, as many systems no longer require fixed calibration targets for registration, reducing setup time and enabling faster fieldwork workflows18. These improvements allow researchers to cover greater areas more rapidly with multiple scan positions, which helps reduce occlusion and improve the completeness of forest structural data.

While increased accessibility and hardware improvements to TLS instruments have reduced or eliminated major data collection bottlenecks, extracting accurate information from the resulting point clouds remains a significant challenge. Recent algorithmic advances, including co-registration methods19, deep learning approaches for crown delineation20, and automated pipelines for large-scale tree extraction21, are streamlining TLS data processing and enabling more efficient analysis of complex point clouds. Together, these developments are expanding the scope of ecological research and transforming how we study forest structure and dynamics. This review provides a forward-looking assessment of the use of TLS technology in forest studies. We outline advancements made over the past years, explore emerging questions that TLS has the potential to address, and highlight the key challenges and bottlenecks that still limit its broader adoption and application.

TLS and increasing realism in modeling forests

The rapid increase in our ability to use TLS to capture extremely detailed 3D descriptions of tree and forest structure has led to an increasing interest in so-called “digital twin” or virtual forest approaches16,22. But what does this mean in practice: do digital twins represent a useful new conceptual framework, a rebadging of “a model” or somewhere in between? There is no doubt that the concept of digital twins is being used to underpin some very large initiatives in linking climate, observation, and modeling (https://destination-earth.eu/).

According to Batty23 the term digital twin was coined in the early 2000s by Michael Grieves[24](https://www.nature.com/articles/s41467-025-63946-6#ref-CR24 “Grieves, M. & Vickers, J. Digital twin: mitigating unpredictable, undesirable emergent behavior in complex systems. in Transdisciplinary Perspectives on Complex Systems 85–113. https://doi.org/10.1007/978-3-319-38756-7_4

(Springer International Publishing, 2017).“) and has subsequently been used in a range of contexts. In its original sense, a digital twin represents a digital mirror image of a physical process, designed to match it precisely in both space and time, with a bidirectional flow of data between the physical and digital counterparts. A model, on the other hand, is generally considered as an abstraction of a physical process, keeping the key elements we are interested in, but simplifying or even ignoring those we are not. A model is therefore a simplified representation of a physical system, whereas a digital twin implies a representation with the maximum detail we can provide23.

A good example of where TLS has facilitated this distinction between a model and a digital twin is in the process of representing radiative transfer in vegetation22,25. This is a crucial application for quantifying and understanding the processes affecting canopy photosynthesis, modeling the Earth’s radiation budget, and biophysical feedback between vegetation and climate26. A great deal of work has gone into developing simplified radiative transfer models of vegetation for successful global monitoring of vegetation properties27,28, using a range of approximations to represent, for example, leaf and soil scattering properties, leaf amount, and physical arrangements in space. However, an alternative approach has been to represent the 3D physical canopy as accurately as possible, including every leaf or needle, branch, and soil element in 3D, and then solving the radiative transfer problem using, for example, Monte Carlo ray tracing (MCRT)29,30.

A limitation in this high-detail MCRT approach, however, is that it requires the spectral properties of every canopy element, including leaves, bark, and soil, to be known. In practice, acquiring such detailed measurements for every forest is not feasible, which prevents the use of these models for inversion—that is, retrieving canopy properties from remotely sensed reflectance. Between simplified parametric models and full 3D reconstructions, intermediate approaches like voxel-based representations (used in models, e.g., DART31) offer a balance between structural realism and computational efficiency, and can be parameterized with TLS data. Thus, rather than serving in large-scale monitoring applications, radiative transfer models based on TLS data help create a detailed scientific understanding of how forest structure influences multi-angular scattering processes of forest canopies22. TLS data can also broaden the scope for modeling canopy scattering, such as estimating photon recollision probability within forest canopies32. The progression of canopy realism in radiative transfer is illustrated in Fig. 1.

Fig. 1: Progression of digital forest representation complexity, from simplified radiative transfer models to highly detailed reconstructions enabled by terrestrial laser scanning (TLS). On the left, parametric radiative transfer models depict forests as simple geometric shapes or approximations. In contrast, TLS-based reconstructions (right) provide high-resolution models of forest stands, including individual tree structures, detailed branching systems through quantitative structure models (QSMs), and foliage elements. TLS instruments capture dense 3D point clouds by scanning forests from multiple ground-based positions. These point clouds can then be processed to reconstruct tree geometry, crown shapes, and canopy surface details, supporting realistic structural inputs for modeling frameworks. The integration of TLS data into advanced modeling frameworks supports full forest reconstructions with applications in radiative transfer simulations, enhancing the realism of digital twins for forest ecosystems.

Another area where TLS is enabling this progression from simplified models to digital twins is in functional structural plant modeling (FSPM). The FSPM approach seeks to model the external structural expression of underlying genetic and phenotypic behavior33,34. FSPMs predict the 3D plant structure that arises from these underlying behaviors. The difficulty is testing and validating these structural predictions for real trees. There have been various attempts to couple 3D tree structure with FSPMs via manual measurements and even procedural models33,35. However, the advancements in TLS have opened the way to a much more effective parameterization of FSPMs, as well as direct tests of their predictions. O’Sullivan et al.34 suggested that TLS will contribute to FSPM development in two key areas: first, by parameterizing static FSPMs to simulate interactions between structure, environment, and physiology; and second, by enabling the testing and calibration of dynamic FSPM predictions to explore ecological and environmental hypotheses. Potapov et al.36, for example, developed a stochastic version of an existing FSPM (LIGNUM) for producing tree structures consistent with detailed TLS data. Sievänen et al.37 used TLS measurements of pine trees of different ages to construct a pseudo-time series of growth of a single tree. They used an FSPM to stochastically simulate crown development to match the TLS-derived crown development and suggested the resulting best-matching FSPM parameters represent the underlying crown development mechanisms. This ability to establish quantitative links between structure and function has enabled the development of a so-called structural economics spectrum, embedding tree size and structural diversity in the wider framework of plant resource use38.

Advances in capturing tree and forest structure via TLS are enabling the transition from simplified structural representations to digital twins, with very high levels of structural detail, using so-called quantitative structure models (QSMs), the algorithmic enclosure of point clouds in topologically-connected, closed volumes39,40. This, in turn, throws up some interesting challenges in terms of how best to use or interpret this detail. In radiative transfer modeling, for example, the challenge is no longer one of representing structure, but how to assign the underlying scattering properties of that structure—the leaves, branches, soil, etc. that make up the resulting scene model. This process will look very different across different wavelength domains, from the shortwave visible to thermal and microwave. In the case of FSPMs, a challenge will be how to feed back the phenotypic information expressed in observed structure to the underlying functional process representation. Challenges of course open further opportunities.

Returning to the question of whether digital twins represent something new and useful for forest monitoring, in the sense used here at least, digital twins are different from models and serve a different purpose. In essence, they allow us to move away from assumptions about tree and forest structure that have been imposed on us simply because of our inability to make the necessary measurements. TLS data are breaking this barrier down, which will benefit a wide range of ecological and environmental applications. The following sections explore these advancements at various levels. We discuss how TLS provides detailed insights into individual tree morphology, supports forest inventories, and aids in quantifying both the structural complexity of forest habitats and the impact of disturbances within these ecosystems.

Understanding the architecture of trees

Tree architecture, also known as tree structure or morphology, refers to the 3D size and arrangement of a tree’s fundamental components (e.g., trunks, twigs, branches, leaves, and needles). The aboveground arrangements determine the efficiency of light capture for photosynthesis, influence competition for resources, and affect ecological processes such as carbon storage, water, and nutrient cycling41. As a result, commonalities exist in the overall structure of different tree species, particularly in the stem and branching patterns, and in their functional roles within the ecosystem.

The architecture of a tree results from the interaction between genetic factors ultimately linked to the plant’s functional strategies (i.e., reproduction) that dictate morphological characteristics unique to each species and both long-term and short-term adaptations to the environment42. These adaptations are influenced by a variety of factors, including biotic pressures like competition for space and increases in liana abundance43, as well as abiotic elements such as light and water availability44,45. Additionally, wind (an abiotic factor) can influence tree architecture both by causing mechanical damage, particularly in structurally unstable trees, and by driving acclimation processes that shape tree form over time46. These effects can extend to neighboring trees and alter overall forest canopy structure46. Therefore, quantifying tree architecture can provide useful information for improving forest management strategies, assessing ecosystem productivity, and modeling carbon dynamics.

Architectural metrics, such as stem diameter, tree height, and crown area, can be easily measured from TLS point clouds47. However, capturing more complex metrics, such as branching patterns and woody volumes, requires more advanced modeling approaches to reconstruct the three-dimensional distribution of tree components16. QSMs of trees can capture the woody branching structure in detail, including the 3D topological branching pattern, as well as the diameters, lengths, surface areas, angles, and volumes of the stems and branches (Fig. 2). Typically, these models consist of a hierarchical collection of geometric primitives, mostly cylinders, which locally approximate the diameter and general geometry of the stem and branches. Collectively, these primitives provide an approximation of the entire woody structure, including the total woody volume48.

Fig. 2: Visualizing tree architecture using TLS-derived quantitative structure models (QSM). The QSM was generated using the TreeQSM algorithm and consists of cylinders approximating the tree’s woody component. Each branch is displayed in a different color. Four panels highlight quantitative structural information computed from the model. A The total branch length per branch order (excluding the stem), highlighting the distribution of smaller, higher-order branches. B Branch volume distribution across orders, indicating that larger branches are concentrated in lower orders. C Branch volume to height, showcasing how woody branch volume is distributed vertically within the tree. D The relationship between branch segment (excluding the stem) diameter and volume, emphasizing the contribution of smaller diameters to overall tree volume.

Several methods exist for generating QSMs from point cloud data, each with different assumptions and outputs. These include TreeQSM40, SimpleForest (formerly known as SimpleTree)48, which is part of the broader Computree platform[49](https://www.nature.com/articles/s41467-025-63946-6#ref-CR49 “Computree. Computree: collaborative platform for remote sensing data processing. https://computree.onf.fr/?page_id=589

(2024).“), 3D Forest50, and AdQSM51. More recent developments include TreeGraph[52](https://www.nature.com/articles/s41467-025-63946-6#ref-CR52 “Yang, W. et al. Treegraph: tree architecture from terrestrial laser scanning point clouds. Remote Sens. Ecol. Conserv. https://doi.org/10.1002/rse2.399

(2024).“) and L1-Tree53. Most methods follow a common workflow: segmenting the point cloud into stems and branches, followed by cylinder fitting to reconstruct geometry. However, some methods, such as Treegraph[52](https://www.nature.com/articles/s41467-025-63946-6#ref-CR52 “Yang, W. et al. Treegraph: tree architecture from terrestrial laser scanning point clouds. Remote Sens. Ecol. Conserv. https://doi.org/10.1002/rse2.399

(2024).“) and TreeQSM40, first derive a complete topology before addressing volume. Differences also arise in implementation: TreeQSM, for example, requires MATLAB, while some offer standalone or open-source solutions. These methods were also built for different purposes, such as targeting the volume or branching structure, and have been validated for different metrics. It is noteworthy that there is not a lot of validation data for many of the more complex metrics, such as individual branch diameters and lengths.

The processing of TLS point cloud data presents major challenges that impact the accuracy of QSM methods. The first is the need for a leaf-wood separation pre-processing step. This step is crucial because leaves can both obscure woody components and emulate them in the point cloud, potentially confusing QSM algorithms and leading to the creation of artificial branching structures. Another challenge is occlusion, where parts of the stem or branches may not be captured due to limited visibility from the scanner positions. Furthermore, accurate segmentation of individual trees from dense plot-level point clouds remains a critical bottleneck, especially in structurally complex forests. Recent advances in artificial intelligence are helping to overcome these challenges. Convolutional neural networks and point-based classifiers have achieved high accuracy in distinguishing foliage from woody elements in complex canopies54,55. To address occlusion, generative models and deep completion networks are being used to infer missing structural details and reconstruct plausible tree geometry from incomplete point clouds56. In the case of individual tree segmentation, 3D deep learning segmentation frameworks are now enabling automated, high-precision delineation of individual trees across a range of scanning platforms, reducing the reliance on manual input and improving scalability20,57.

Additionally, the accuracy of total tree volume estimates using QSMs is heavily influenced by the visibility of the tree structure in TLS data, making reliable volume and biomass estimates (assuming wood density is known) highly dependent on data quality. For many species, the total tree volume primarily consists of the stem and large branches, which are the most visible components in TLS data. Due to the centimeter-scale size of LiDAR laser beams upon contact with trees, point clouds often overestimate the diameters of small branches, leading to inflated volume estimates in QSMs58. One approach to address this overestimation is to apply filtering techniques or to incorporate actual twig diameter measurements[59](https://www.nature.com/articles/s41467-025-63946-6#ref-CR59 “Morales, A. & MacFarlane, D. W. Reducing tree volume overestimation in quantitative structure models using modeled branch topology and direct twig measurements. For. Int. J. For. Res. https://doi.org/10.1093/forestry/cpae046

(2024).“) to adjust the cylinder diameters for greater accuracy. More generally, estimates of the size and shape of small or distant branches, particularly those with diameters close to the TLS footprint, are likely to be unreliable without strong validation. These structural uncertainties propagate into biomass estimates, especially when combined with intra-tree and intra-species variability in wood density.

A new asset in forest inventories

Forest inventories have long been employed to assess and monitor the condition, composition, and changes in forests over time. These inventories provide important data for understanding forest resources, informing policy decisions, and assessing carbon stocks and biodiversity. For instance, National Forest Inventories (NFIs) are an essential tool for countries to report on their forest status to international organizations and agreements, such as the Food and Agriculture Organization or the United Nations Framework Convention on Climate Change. ALS has been adopted already for decades in operational and commercial forest inventories, particularly in Nordic countries, to enhance efficiency.

Research studies and reviews commonly agree on the technological readiness of TLS for operational inventories, in particular with respect to accurate geometric measurements at the centimeter to millimeter scales15,16,17. Recent benchmarking studies have shown that TLS can estimate tree attributes such as diameter at breast height (DBH) with errors typically below 2 cm and stem curve profiles with comparable accuracy in boreal forests60. In tropical agroforestry systems, TLS has also produced strong correlations with field-based measurements of canopy openness (r = 0.79) and tree height (r = 0.58)61. These findings confirm that TLS provides structural estimates comparable in accuracy to conventional methods, while also capturing three-dimensional complexity beyond what field inventories typically offer.

TLS therefore delivers not only standard inventory metrics, but also allows estimation of structurally detailed attributes that are typically unmeasured in the field, such as crown area and volume, foliage clumping (relevant for modeling light interception)62, and the space around a tree (for assessing growth potential)63 (Fig. 3). Given that the raw point clouds record rich 3D information, previously collected TLS data can be reprocessed to extract novel metrics as algorithms improve, even years after the original data collection. Moreover, the possibility of increasing sampling plot size with TLS compared to traditional inventories has been discussed to improve the representativity of samples and the link to airborne and satellite remote sensing data17,63.

Fig. 3: Overview of metrics derived from terrestrial laser scanning (TLS) for forest inventories. TLS instruments collect detailed 3D point clouds of forest plots, enabling accurate measurement of both tree- and stand-level structural attributes. At the tree level, TLS enables detailed measurements of tree height, crown dimensions, branching, and stem form, along with temporal monitoring. At the stand level, TLS provides metrics like basal area, competition indices, canopy height, and structural complexity, complementing traditional inventory methods with higher accuracy and expanded measurement capabilities.

Deriving additional information from inventory data, such as stem volume and tree biomass, strongly relies on estimates from allometric models. Ensuring the reliability of these models is therefore critical for accurate assessments. In some recent NFIs, differences in stem volume estimates have been observed when using models with varying predictors, such as two (DBH and height) versus three (DBH, height, and diameter at 6 m) variables. To address such discrepancies, projects utilizing TLS data have been undertaken to develop improved stem volume models64,65. For example, studies have shown that stem form can vary over time64 and across regions65,66, with changes most pronounced in the lower parts of tree stems. These findings highlight the importance of regionally calibrated allometric models and the potential of TLS data to capture variability in stem form across tree species and geographic areas.

Forest management, which includes activities such as thinning, clear-cutting, and selective harvesting, can have significant effects on tree growth, particularly in terms of stem form. For instance, management practices aimed at reducing competition for light and nutrients can influence how trees allocate resources to their trunks, resulting in changes in stem and branch diameters, height, and form. TLS has proven effective in assessing growth changes in trees after forest management interventions, capturing structural changes in tree morphology67,68,69. It has also been employed to generate competition indices using crown information, which improves upon traditional approaches relying solely on DBH or height70,71. Furthermore, point clouds derived from TLS enable a more comprehensive assessment of tree competition by quantifying the occupied space around trees, providing detailed insights into tree interactions and their surroundings72,73.

TLS data has been used in characterizing differences in stem form over both time and space64,74. For instance, bitemporal TLS data were used to capture stem growth dynamics, volumetric changes, and localized deformations, providing useful insights into tree responses to environmental factors and management interventions72,75. Bitemporal TLS data were also applied to assess changes in stem shape and identify relationships between stem deformation and drought stress, highlighting how environmental factors can influence tree morphology over time76. Interannual TLS data have further enhanced the understanding of tree dynamics. Seasonal radial growth has been detected in TLS point clouds collected before and after the growing season, although the study also highlighted important challenges in detecting millimeter-scale changes in stem diameter77. Furthermore, defoliation was assessed from TLS scans conducted during a single growing season and linked to independently measured growth losses69.

High temporal resolution TLS data of individual trees, e.g., once every 30–60 min, are also becoming increasingly accessible78,79. Multi-temporal approaches enable the monitoring of tree structural dynamics, such as movement and responses to environmental factors, at detailed temporal scales, offering deeper insights into plant physiology and interactions with their surroundings. Campos et al.[80](https://www.nature.com/articles/s41467-025-63946-6#ref-CR80 “Campos, M. B. et al. A long-term terrestrial laser scanning me