Recently, the cultivation of fast-growing tree (FGT) plantations has gained importance due to the growing energy and climate crisis. FGT plantations have the potential to reduce carbon footprints and lower greenhouse gas emissions by utilization of local renewable energy sources. Effective monitoring of above-ground biomass (AGB) is crucial for the successful management of these plantations. Standard methods for estimating AGB rely on easily measurable parameters, such as Diameter at Breast Height (DBH) and tree height, which are highly correlated with AGB. Traditional methods for measuring DBH include measuring tapes and calipers; however, these techniques can be labor-intensive, time-consuming, and limited when assessing large areas. Innovative approaches, such as photogrammetry, terrestrial laser scanning (TLS), mobile laser scanning (MLS), and iPhone LiDAR scanning, can complement these traditional methods by generating point clouds that can be used for extracting dendrometric parameters. This study evaluates the effectiveness of TLS (RIEGL VZ-1000), MLS (Stonex X120 GO), iPhone LiDAR (iPhone 13 Pro MAX), and terrestrial photogrammetry (iPhone 13 Pro MAX) for estimating DBH in a Paulownia plantation. Each technology has limitations: while TLS offers high accuracy, it is also expensive and time-consuming. Similarly, MLS is relatively costly. On the other hand, iPhone LiDAR and terrestrial photogrammetry are more affordable alternatives; however, the iPhone LiDAR has a limited scanning range, and photogrammetry requires considerable time and expertise for data collection and processing. The primary objective of this study was to evaluate these technologies based on their accuracy in DBH estimation, ease of use, data collection, processing time, and cost within the ideal conditions of a Paulownia plantation (characterized by the absence of understory, level ground, and uniform tree shape and spacing). The aim was to determine whether traditional methods could be replaced with more efficient, quicker, easier, and cost-effective alternatives. Results indicated that TLS, MLS, and photogrammetry provided similar DBH estimation accuracies, with root mean square error (RMSE) values between 0.7 and 0.72 cm and relative RMSE values between 2.87 % and 2.95 %. In contrast, the iPhone LiDAR was the least accurate, with an RMSE of 1.7 cm and an rRMSE of 6.96 %. This study demonstrates that all evaluated technologies offer sufficient accuracy for DBH estimation, although TLS and MLS capture additional parameters at a higher cost. Therefore, TLS is impractical for DBH estimation in plantation environments due to its high cost, time, and labor demands. While less expensive, terrestrial photogrammetry also requires significant time investment and operator expertise. Despite its cost, MLS achieved the best results among all the evaluated technologies and proved to be the fastest and relatively simple. If cost is a concern, the best solution for DBH estimation in an FGT plantation environment would be iPhone LiDAR scanning. It represents the most affordable option with satisfactory accuracy and ease of use.
This abstract is based upon work from COST Action 3DForEcoTech, CA20118, supported by COST (European Cooperation in Science and Technology) and APVV 20 0391 Monitoring of forest stands in three-dimensional space and time by innovative close-range approaches.
How to cite: Skladan, M., Chudá, J., Singh, A., Masný, M., Lieskovský, M., Pástor, M., Mokroš, M., and Vybošťok, J.: Choosing the right close-range technology for measuring DBH in fast-growing trees plantations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-2134, https://doi.org/10.5194/egusphere-egu25-2134, 2025.
Advances in 3D reconstruction techniques have recently democratized point cloud generation methods in various domains, including forestry and general tree mapping. This in turn has been driving the development of the virtual forest concept. Despite its potential, this concept remains only vaguely defined within the forestry domain, often varying in meaning depending on its implementation. Recognizing this ambiguity, our work seeks to unify the diverse interpretations of virtual forests by proposing a standardized definition from a geomatics perspective. Two main ambiguities may be identified in the literature: first, which sensor to use in which case during the data acquisition phase of virtual forests. Second, how to represent the data in the virtual world. In an attempt to bridge these gaps, we introduce two critical concepts: the sensor-oriented Level of Scale (LoS) and the data-centered Level of Detail (LoD). The LoS concept aims to help 3D technology users in choosing which sensor is best suited for their purposes by using the scale of the scene and its complexity as the determining factors. This presents a very useful tool during the project planning phase, where balance between data quality and project budget is an important aspect. The LoD concept on the other hand, draws inspiration from established definitions in CityGML to represent trees in different complexities. In this study, the proposed LoD also incorporates an additional dimension to account for variations in data formats (e.g., mesh, point clouds, parametric models, etc.). These frameworks aim to clarify and structure the representation of virtual forests, addressing inconsistencies in their application across different contexts. A numerical analysis was also conducted to further highlight the practical implications of these concepts in improving the precision and utility of 3D vegetation mapping techniques. Although the findings of this study do not aim to establish an official standard—achieving this would require further collaborative efforts across disciplines—they provide a foundational framework for advancing standardization efforts. By offering a structured approach to defining and representing virtual forests, we hope to contribute to the broader development of practical, scalable guidelines that can be applied within forestry and related fields. This initiative marks a step forward in aligning geomatics with the needs of modern forestry applications.
How to cite: Murtiyoso, A., Overney, N., and Suwardhi, D.: Sensor selection and 3D data representation in virtual forests, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21636, https://doi.org/10.5194/egusphere-egu25-21636, 2025.
In the context of the COST Action 3DForEcoTech and an ISPRS scientific initiative, a benchmarking activity was conducted to evaluate the performance of 13 software solutions designed for automated forest inventory using ground-based point clouds. These tools, which serve as digital analogs to traditional forest inventories, were tested on 12 datasets from four distinct forest plots featuring diverse forest types and acquisition methods, including two different Terrestrial Laser Scanners (TLS) and a handheld laser scanner. The experiments, carried at TU Wien (Vienna, Austria) in September 2023, with 15 researchers working on identical computing environments, assessed each software’s ability to detect trees, extract the positions, and estimate DBH, along with computational efficiency. Results showed that while most solutions achieved good performance, a few significantly outperformed the rest in accuracy and processing time, whereas others struggled with larger point clouds, highlighting important differences in scalability and robustness. This study provides valuable insights into the current capabilities and limitations of automated forest inventory tools, guiding both researchers and practitioners in selecting the most suitable software for their needs.
Keywords: LiDAR, TLS, Forest Inventory, DBH Estimation, Software Benchmarking, Point Cloud Processing, 3DForEcoTech
This contribution is based upon work from COST Action CA20118, supported by COST (European Cooperation in Science and Technology).
How to cite: Cabo, C., Rehush, N., Wild, B., Murtiyoso, A., Iglseder, A., Hollaus, M., and Mokros, M.: Benchmarking Software Solutions for Forest Inventory from Ground-Based Point Clouds, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21885, https://doi.org/10.5194/egusphere-egu25-21885, 2025.
Forests are critical natural resources for human life and wildlife, as they sustain and protect biodiversity, and supply multiple ecosystem services. However, these ecosystems are vulnerable to human-driven climate change, necessitating automated systems to monitor structural changes at the individual tree level and assess forest responses to climate anomalies. Forest inventories, that contain accurate and detailed measurements of forest structure, are essential to improving our knowledge of ecosystem services and functions. UAV and LiDAR-based tree canopy detection is valuable for estimating essential ecosystem variables (EEVs). In this research, we have validated tree structural properties primarily diameter at breast height (DBH) and tree height obtained from UAV imagery and airborne LiDAR point cloud data using field measured data. We developed Drone4Tree, a user-friendly platform built on Streamlit and Flask that provides an end-to-end solution for processing UAV imagery. The platform processes UAV-acquired data to generate orthomosaics using OpenDroneMap, delineate tree crowns using U-Net based segmentation, and derive tree attributes such as tree height, canopy area etc.
The LiDAR data was processed using forest structural complexity tool (FSCT). This tool applies sensor agnostic semantic segmentation on the point cloud to obtain individual trees, stems and their structural properties. The LiDAR and UAV derived properties were joined with the field obtained parameters. Comparative analysis shows strong agreement between field DBH and LiDAR-derived DBH (R2 = 0.97), indicating reliable DBH estimation from LiDAR data. For tree height, the LiDAR-based measurements correlated well with field measured tree heights (R2 = 0.73), though comparisons with the UAV-based tree height (R2 = 0.97) obtained from canopy height models (CHM) revealed a lower correlation (R2 = 0.66). UAV-based tree height measurements show statistically significant relation with field measured height (R2 = 0.57). These results indicate that LiDAR and UAV data complement each other, with UAVs offering efficient monitoring capabilities while LiDAR providing additional precision.
These findings underscore the potential of integrating UAV and LiDAR technologies for accurate and efficient forest monitoring, enabling improved assessment of ecosystem functions and responses to climate change. By combining these complementary methods, platforms like Drone4Tree can support sustainable forest management and contribute to addressing the monitoring of global environmental changes.
How to cite: Gupta, S. K., Schulze, F., Mallast, U., Gründling, R., Brede, B., Kleidon-Hildebrandt, A., Rebmann, C., Dienstbach, L., and Schmidt, P.: Validating tree inventory: Analysing tree structural properties from high-density airborne LiDAR point clouds and UAV imagery, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18078, https://doi.org/10.5194/egusphere-egu25-18078, 2025.
Persistent homology (PH) is a well-established mathematical approach that has been increasingly used to measure plant morphology. Stemming from topological data analysis, PH was developed as a mathematical framework to characterize topological relationships between data points. Structures are found by tracking topological features that persist across scales, making it resistant to noise and invariant to orientation and size. The great advantage of PH lies in its ability to integrate several morphological features into a single metric value. In this way, it captures multiple and comprehensive measurements better than uni- or multivariate systems. Consequently, PH enables an accurate quantification of phenological variations, quantifying the complete morphology of a plant, including growing branching structures.
Most studies that use PH for plant morphology quantification use 2D images for the task. This is despite the fact that plants are essentially three-dimensional objects and should be analysed in that space. Studies that have explored PH in 3D focus on classification of man-made objects (e.g., toys or furniture). However, point clouds of trees that were acquired in their natural environment present a bigger challenge. There, the collected data is unevenly distributed, includes occlusions and highly depends on the season (leafing stage). All of these can vastly influence the topological analysis, and lead to incorrect structures. Not only that, but also the platform used to acquire the data might greatly affect the quantification. This is due to the point of view (i.e., from the air or terrestrially), which documents different parts of the tree.
In this work, we test the applicability of PH for tree morphology characterization. We show how such an analysis enables us to describe various branching topologies. We use PH on individual trees that were acquired by different laser scanning platforms (i.e., UAV-borne and terrestrial), with and without leaves. This enables us to evaluate the potential of PH for 3D tree morphology characterization, test its limits, and explore its application in tree species classification.
How to cite: Arav, R.: Introducing persistence homology in 3D point cloud processing for tree morphology characterization, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18085, https://doi.org/10.5194/egusphere-egu25-18085, 2025.
Estimation of vegetation water content (VWC) in forest ecosystems is essential for understanding and monitoring forest health, transpiration, and responses to environmental changes. Currently there exists no remote sensing method capable of observing VWC changes in forests.
The BorealScat-2 radar tower, constructed in the Svartberget Experimental Forest in northern Sweden, aims to address this challenge by investigating the relationship between radar backscatter and VWC in a boreal forest. The tower provides tomographic radar images at P- (435 MHz), UHF- (600 MHz), and L- (1270 MHz) across all polarization combinations, offering high precision measurements at 30-minute intervals. The high precision is made possible by moving the antenna frame along a 4-m horizontal baseline, followed by incoherent averaging of the tomograms. Complementary in situ measurements, including sap flow sensors, trunk moisture sensors, dendrometers, and an eddy covariance flux tower, enable detailed analysis of water dynamics across the soil-plant-atmosphere continuum within the radar footprint.
Time series results show that P- and UHF-band backscatter covaries (positively correlated) with VWC, with UHF-band backscatter capturing VWC trends over timescales from hours to months. L-band radar observations, however, reveal an unexpected diurnal backscatter pattern, where canopy backscatter increases during periods of decreasing VWC, suggesting a complex interplay of scattering and attenuation effects. This behaviour contrasts with the expectation that an increase in VWC leads to an increase in backscatter. We propose an inversion model, accounting for attenuation effects, for estimating VWC from L-band backscatter.
Results show that the model successfully estimates changes in VWC from canopy backscatter and attenuation measured by the tower over timescales of hours to weeks, demonstrating the possibility of using radar observations for estimating VWC changes in forests. The findings underscore the relevance of tower-based radar observations for refining remote sensing algorithms and for presenting new applications for upcoming L-band synthetic aperture radar missions for global forest monitoring.
How to cite: Leistner, T., Monteith, A. R., Ulander, L. M. H., Bennet, P. J., and Gutierrez Lopez, J.: Towards Estimating Vegetation Water Content in Boreal Forests Using Tower-Based Radar Observations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10626, https://doi.org/10.5194/egusphere-egu25-10626, 2025.
Tree growth is a key indicator of forest health and development, especially in the context of a changing climate. Interactions between abiotic and biotic factors influencing tree growth are highly complex, with their full magnitude still unknown. Even trees of the same species and within the same forest can present high variability in their growth, as they are affected by various macro- and micro-scale factors. In order to detect and quantify tree growth at tree level, close-range monitoring with high spatial and temporal resolutions is required. For this purpose, LiDAR (Light Detection and Ranging) data is widely recognized for its ability to produce high-resolution point clouds, which enable studying intricate changes in trees.
The goal of our study is to explore the potential of daily LiDAR time-series for detecting the onset of tree height growth and quantifying the total growth in tree height, to help understand the biotic and abiotic factors contributing to height growth variability in Scots pine (Pinus sylvestris) trees. Here, we studied 97 Scots pine trees during the growing season of 2021 with dense spatiotemporal point cloud time series collected with LiDAR Phenology Station (LiPhe) in Hyytiälä forest research station, Finland. We developed a semi-automatic framework to extract individual tree height time-series, which includes point cloud registration, point cloud segmentation, and tree height estimation. Based on extracted height time-series derived from LiPhe, we detected the onset of tree height growth using a change point detection algorithm.
We found up to 28 days of variability in the onset of height growth within the studied Scots pine trees. To investigate the factors influencing the variability in the onset of height growth, we used tree size, neighborhood characteristics, and topography as explanatory variables in a linear mixed-effects model. These variables were also estimated from LiDAR data. The best performing model for modelling the onset of growth combined Plant Area Index (PAI), Vertical Complexity Index (VCI), and Topographic Wetness Index (TWI), as fixed-effect terms.
Our results suggest that higher density and complexity of neighboring trees leads to earlier onset of tree height growth, which can suggest competition for light and microclimate variability. Meanwhile, lower TWI led to earlier onset of tree height growth, indicating that trees located on a slightly higher slope with less water availability grew earlier in height. Lower areas may have a cooler microclimate, since they often retain more soil moisture and are less exposed to wind, which can lead to later growth onset.
We conclude that daily LiDAR time-series enables measurements that are challenging to achieve using other techniques, such as detecting the onset of height growth. Our study suggests that the onset of height growth may be mainly influenced by light competition and microclimate, demonstrating the potential of tree-level LiDAR-derived metrics in studying how microclimatic changes affect forest adaptability in the context of a changing climate. We will further continue to study the influence of the timing of growth on the amount of growth during the growing season.
How to cite: Korpelainen, T., Campos, M., Yrttimaa, T., Junttila, S., Wang, Y., and Puttonen, E.: Investigating within-forest variability in the onset of tree height growth in a boreal Scots pine forest, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8679, https://doi.org/10.5194/egusphere-egu25-8679, 2025.
Accurate forest inventories are crucial for sustainable management, but processing ground-based LiDAR and photogrammetric point clouds remains complex and inaccessible to many users. 3DFin addresses this gap as a fully automated, open-source software for extracting key tree metrics—tree height, diameter at breast height (DBH), and tree location—from Terrestrial Laser Scanning (TLS), Mobile Laser Scanning (MLS), and photogrammetry-derived point clouds. Unlike existing tools, 3DFin is designed for complete ease of use, requiring no technical expertise, allowing any user—regardless of experience—to generate forest inventory data with just two clicks. Integrated as a plugin in CloudCompare, the most widely used free software for 3D point cloud processing, 3DFin seamlessly fits into existing workflows without requiring additional installations or programming. Tested on publicly available datasets across diverse forest conditions, it achieves near-perfect tree detection rates and DBH estimations with RMSE <2 cm, all while processing plots in just 2–7 minutes. By bringing cutting-edge point cloud analysis to a broader audience, 3DFin makes advanced forest inventory processing accessible to all, bridging the gap between research and real-world application. In addition, we also introduce here 3DFos, a soon-coming open-source plugin in CloudCompare which automatically segments forest point clouds in vegetation classes such as stems, branches and leaves, understory and ground.
Keywords: CloudCompare, LiDAR, TLS, MLS, Photogrammetry, Tree Metrics, Open-Source, Forest Inventory
How to cite: Santin, C., Laiño, D., Ordoñez, C., Sánchez-López, N., and Cabo, C.: 3DFin and 3DFos: Open-Source CloudCompare Plugins for Automated 3D Forest Inventory and Classification, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-21887, https://doi.org/10.5194/egusphere-egu25-21887, 2025.
In recent years, numerous methods have been developed to automate tree species classification using Shallow and Deep Learning techniques. Traditional approaches often utilize scanner-measured data (e.g., intensity) and 3D geometric features to compute statistical descriptors, which are used to train algorithms like Random Forests or Support Vector Machines. Deep Learning approaches, such as convolutional neural networks, process 2D raster images of point clouds but may lose critical 3D geometric details. Graph-based methods and architectures directly processing unstructured 3D data have shown promise but are often computationally intensive and less practical for industry.
To address these challenges, we developed a method that combines 2D raster and 3D point cloud features, achieving over 90% average classification accuracy. Our approach leverages well-established techniques and integrates them into TerraScan software for industrial use. Data augmentation, including SMOTE, addresses class imbalances, while features extracted from multiple raster viewpoints enhance dataset diversity.
Using TerraScan, users can efficiently preprocess data, augment training examples, and train models for over 10,000 trees in under 40 minutes on a GeForce RTX 4080. The system provides confidence scores with predictions, enabling manual evaluation of low-confidence results. This versatile method shows potential for broader object classification tasks beyond tree species identification.
How to cite: Shcherbacheva, A., Puttonen,, A., and Soininen, A.: AI-aided forest inventory with TerraScan: combining 3D and 2D features for tree species classification, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1985, https://doi.org/10.5194/egusphere-egu25-1985, 2025.
Accurate and scalable tree species identification remains a critical challenge for global forest monitoring and management. Despite the increasing availability of remotely sensed data, the lack of standardized, high-quality ground truth datasets limits the potential of supervised machine learning models in capturing the tree diversity of forest ecosystems across different environmental and geographic contexts. Prior studies have highlighted the need for global-scale, high-resolution datasets to develop robust algorithms capable of capturing the diversity of forest ecosystems.
Towards a benchmark dataset for tree species identification in high-resolution aerial imagery. To address this critical gap, we introduce the TreeAI database, an open-access dataset designed to support advanced research in tree species identification and forest dynamics. The database comprises 53 datasets (47 publicly available) from 32 countries, representing 61,158 annotated trees across 5,000 ha of forest ecosystems, and it is still growing.
The TreeAI database provides annotations paired with high-resolution imagery (RGB and near-infrared bands at 1–10 cm spatial resolution, with an average of 3.5 cm). The database offers three key advancements. First, its global representation spans diverse ecosystems, climates, and species, enhancing its applicability across regions. Second, including centimetre-scale orthophotos ensures sufficient detail for identifying subtle differences between species. Finally, its community-driven design fosters ongoing contributions and ensures a dynamic dataset that evolves with the field's needs.
Preliminary tree species identification analysis using deep learning algorithms conducted for Switzerland, with very heterogeneous forest ecosystems and challenging topography, yielded promising results. The average F1-score for nine common species was 0.72, with Larix spp., Picea abies, and Tilia spp. exceeding 0.80. The mean average precision (mAP) across all the species was 0.76. These findings underscore the potential of the TreeAI. To further harness TreeAI’s potential, a scientific competition will be launched in 2025, challenging participants to develop deep-learning algorithms that maximize tree species identification accuracy across a broad range of forest ecosystems.
The impact of a global database for tree species annotations. The TreeAI database serves as a benchmark dataset for advancing artificial intelligence models, enabling automated forest inventory systems. This capability allows for the creation of high-resolution maps detailing tree species distributions, which can be used by researchers and practitioners for applications such as forest management, biodiversity monitoring, and ecosystem conservation. Moreover, the dataset complements existing National Forest Inventory (NFI) data, providing additional resources for point-based regional studies and enhancing ecological research at finer scales. Furthermore, the database promotes the refinement of AI models for practical forestry applications, fostering innovation in open science and collaborative research.
Further needs and collaboration potential: i.) expanding its geographic and tree species coverage, such as tropical forests, which remain inadequately sampled in existing datasets. ii.) integrating TreeAI with Earth observation platforms, such as Planet Scope, Sentinel-2, and GEDI. iii.) exploring methods to enhance data accessibility and interoperability, ensuring that the database meets the evolving needs of its users. Feedback from the broader forestry community will be instrumental in shaping these developments, emphasising addressing challenges related to data standardization, processing efficiency, and algorithm performance.
This contribution is based upon work from COST Action CA20118, supported by COST (European Cooperation in Science and Technology).
How to cite: Mokros, M., Xia, Z., Cheng, Y., Gessler, A., Kattenborn, T., Liang, X., Mosig, C., Puliti, S., Rehush, N., T. Waser, L., C. Griess, V., and Beloiu Schwenke, M.: TreeAI: a global database for tree species annotations and high-resolution aerial imagery, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18117, https://doi.org/10.5194/egusphere-egu25-18117, 2025.
Accurate classification of forest types and tree species is an important aspect of forest monitoring but it requires vast amounts of spatio-temporal data. Remotely sensed data provides a viable solution for acquiring the necessary global-scale information. Historically multispectral data has been used for forest monitoring, but the limited number of spectrally-broad bands often do not provide sufficient differentiation between similar tree species, lowering classification accuracy. Hyperspectral data offers improved spectral resolution which enables to differentiate similar tree species. Nevertheless, the quality of ground truth data used for classification remains a challenge, as it is often limited and noisy.
This study presents a hierarchical, three-stage classification approach utilizing hyperspectral data, cascaded machine learning models and spectral unmixing algorithms to classify forest types and individual tree species. The approach integrates coarse level dataset for broad level classification and finer resolution hyperspectral imagery for fine-scale spectral and structural variability. Furthermore, to address the possibility of low quality ground truth labels we propose a semi-supervised training framework leveraging pseudo-labeling.
The cascaded three-stage architecture sequentially processes the data, with each stage consisting of an XGBoost model trained to address specific challenges. The first stage is a coarse classifier, classifying forest into three broad categories: Evergreen, Deciduous, and Mixed. This model is trained on coarse resolution 60m GSD EMIT data and supervised labels generated using the National Land Cover Database. The second stage further refines the three classes into 28 different forest group types labels as defined by the USDA Forest Service's Forest Inventory and Analysis (FIA). The third and the final stage classifies each of the forest pixels by its dominant tree species, leveraging the outputs from the previous stage and AVIRIS-NG high resolution 4m GSD hyperspectral data as additional input features. Non-dominant tree species are identified using Vertex Component Analysis based spectral unmixing and classified into pure tree species spectras using spectral similarity metrics. The abundances of dominant and non-dominant spectras are then mapped using the Fully Constrained Least Squares approach.
This method was tested over two regions: Shasta-Trinity National Forest, California, USA and Grand Mesa National Forest, Colorado, USA. The resulting tree distribution mapped 10 different individual tree species and were validated against USDA’s Treemap product . For these test regions the resulting overall accuracy from the entire 3-stage model is 80%. The individual stage accuracies for stage 1, stage 2, and stage 3 classification, were 94%, 92%, and 92% respectively.
Despite these promising results, the approach is constrained by the availability of high-quality ground-truth data for supervised training. To address this, a pseudo-labeling technique that generates additional training data by iteratively assigning labels to unlabeled samples with high model confidence was explored. The preliminary results indicate that the inclusion of pseudo-labeled data training can enhance the classification accuracy of the proposed hierarchical cascaded approach for forest applications.
How to cite: Ambudkar, S., Kravitz, J., and Yeggina, S.: Hierarchical Classification of Forest Types and Tree Species Using Multi-Resolution Hyperspectral Data and Pseudo-Labeling for Enhanced Model Training, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17277, https://doi.org/10.5194/egusphere-egu25-17277, 2025.
Accurate tree inventories are critical for monitoring forest resources and assessing ecosystem services, particularly carbon storage. This study presents the first airborne tree inventory for Cyprus, a climate change hotspot lacking a comprehensive national forest inventory. Using high-resolution orthophotos, we developed a novel method for tree segmentation and individual-level carbon stock estimation.
Tree identification and segmentation were performed using a published state-of-the-art convolutional neural network (CNN) model, previously applied in Denmark and Finland, which was completely re-tuned using local annotations to account for Cyprus’s specific conditions. This approach achieved 90% accuracy in tree crown delineation. Given the absence of suitable allometric equations for Cyprus' tree species, we developed novel, locally tailored allometric equations for above-ground biomass estimation, achieving 92% accuracy. These equations, derived from crown dimensions and height extracted through CNN models applied to canopy height maps (CHMs), enable accurate carbon stock estimation for individual trees.
The integration of orthophotos and CHMs proved highly effective in capturing detailed structural data across diverse forest landscapes. Our methodology is scalable, cost-effective, and robust, offering a valuable tool for forest management, climate change mitigation, and policy development in Cyprus. This project establishes a comprehensive baseline for Cyprus' forest resources and demonstrates the potential of combining remote sensing and AI technologies for national-scale environmental monitoring, including urban trees.
How to cite: Zenonos, A., Li, S., Brandt, M., Sciare, J., and Ciais, P.: Development of the First Airborne Tree Inventory for Cyprus and Novel Allometries for Carbon Stock Estimation Using AI Models and High-Resolution Remote Sensing Data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4730, https://doi.org/10.5194/egusphere-egu25-4730, 2025.
Abstract:
Data assimilation (DA) has been applied for several decades in areas such as meteorology and robotics, to predict the state of systems that evolve over time, by integrating model-based forecasting with repeated observations. Recently, DA has gained attention in forest inventory applications. For instance, study by Nyström et al. (2015) not only demonstrated the theoretical potential of employing dense time series of remotely sensed (RS) data but also identified several obstacles that must be overcome before the methodology can be practically adopted. Within the SmartForest project, we are further exploring the usefulness of DA techniques for forest inventory and mapping of forest attributes.
Recent studies have shown that DA has a potential to maintain the accuracy of plot and stand level information, obtained from accurate but expensive surveys, such as airborne laser scanning (ALS), by making use of inexpensive optical satellite data and DA throughout several subsequent years. However, with ever-increasing amounts of RS data, it is important to evaluate not only how to make assessments and growth updates through DA, but also how to best utilize huge amounts of RS data from within single years. For example, the European Space Agency’s Sentinel-2 satellites currently provide new data across boreal forests every second week.
In a study initiated within the Norwegian SmartForest programme, we evaluate whether building separate models for each RS dataset and applying composite estimation or merging all data into a single model through principles of partial least squares regression and random forest non-parametric regression, yields the best results in terms of prediction accuracy.
Our investigation was conducted within the Våler municipality of Norway and focused on growing stock volume as our primary target variable. The RS data were acquired in 2022 and included ALS point clouds, digital aerial photogrammetric point clouds, and Sentinel-2 spectral data. Alongside comparing prediction accuracies, we conducted a qualitative assessment to discern the practical advantages and disadvantage of each method in integrating them into a multi-temporal data DA system.
Reference:
Nyström, M., Lindgren, N., Wallerman, J., Grafström, A., Muszta, A., Nyström, K., Bohlin, J., Willén, E., Fransson, J.E., Ehlers, S. and Olsson, H., 2015. Data assimilation in forest inventory: first empirical results. Forests, 6(12), pp.4540-4557.
How to cite: Saarela, S., Gobakken, T., Ørka, H. O., Bollandsås, O. M., Næsset, E., and Ståhl, G.: Handling single-year big data in multi-temporal forest inventory and mapping systems, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6584, https://doi.org/10.5194/egusphere-egu25-6584, 2025.
Our objective was to train a nationwide ICESat-2 model for the estimation of above-ground biomass (AGB) and its uncertainty for the entire country of Finland. The model was trained using data from eight forest inventory areas from different parts of the country. The inventory areas had airborne laser scanning, Sentinel-2 data, and field plots publicly available, and these data were used to construct proxy models that were employed to predict AGB values for the ICESat-2 tracks overlapping with the inventory areas. The final ICESat-2 AGB model was based on n = 11676 track segments (90 x 15 m) from the eight training areas. Both day and night data were used in the construction of ICESat-2 model, but all data with snow or cloud cover were omitted.
The ICESat-2 model was applied to all forested ICESat-2 segments (n = 288391) obtained from Finland in year 2021. The total AGB for Finland and its uncertainty were estimated using a hierarchical hybrid approach that only used this sample of ICESat-2 tracks without wall-to-wall mapping. The uncertainty estimation considered tree biomass models, proxy models, the nationwide model, and sampling as error components. The final biomass estimate was compared with the official statistic from the Finnish National Forest Inventory (NFI).
The total AGB estimated for Finland was 1063.0±114.9 million tons, while the reference value from NFI was 1308 million tons. Thus, our method resulted in clear underestimation of AGB. Probable reasons for the observed underestimation include averaging of large biomass values due to the long model chains, and misclassification of sparser canopies as noise. Nevertheless, our result shows that ICESat-2 is feasible for AGB estimation in large areas, but more research is needed to reduce the underestimation.
How to cite: Korhonen, L., Saarela, S., Maltamo, M., Packalen, P., Popescu, S. C., and Varvia, P.: Nationwide estimation of boreal forest above-ground biomass using ICESat-2 data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5030, https://doi.org/10.5194/egusphere-egu25-5030, 2025.
Forest canopy structural complexity (CSC), the intricate arrangement and occupation of canopy elements in three-dimensional space, plays a critical role in shaping forest ecosystem productivity and stability by regulating light and water distribution within the canopy. However, the relationship between forest CSC and forest ecosystem productivity and stability remains controversial in current regional-scale studies, necessitating further investigation at broader spatial scales. Here, we introduce a novel entropy-based metric, canopy entropy, to quantify forest CSC from light detection and ranging (lidar) data. This metric effectively captures forest CSC variations arising from both horizontal and vertical arrangements and occupations of canopy elements. Notably, canopy entropy estimates from multiplatform lidar data demonstrate strong agreement, establishing its suitability for large-scale applications. Leveraging these advantages, as well as airborne lidar data from 4,000 forest plots worldwide and spaceborne lidar data from the Global Ecosystem Dynamics Investigation, we map the global distribution of forest CSC and investigate its relationships with forest ecosystem productivity and stability. We find climatic factors, especially water availability, play a critical role in driving the global distribution of forest CSC, while biotic factors exhibit a strong coupling impact with climatic and edaphic factors. From a global perspective, forest CSC predominantly enhances productivity and stability, although substantial variations are observed among forest ecoregions. The effects of forest CSC on productivity and stability are the balanced results of biodiversity and resource availability. These results offer valuable insights into understanding controversies in regional-scale studies. Furthermore, we found that managed forests generally exhibit lower CSC compared to intact forests but demonstrate stronger positive effects of CSC on ecosystem productivity and stability, underscoring the urgent need to incorporate CSC into forest management strategies to enhance climate change mitigation efforts.
How to cite: Liu, X., Feng, Y., Hu, T., Luo, Y., Zhao, X., Wu, J., E. Maeda, E., Ju, W., Liu, L., Guo, Q., and Su, Y.: Global Mapping of Forest Canopy Structural Complexity and Its Links to Ecosystem Productivity and Stability, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5107, https://doi.org/10.5194/egusphere-egu25-5107, 2025.
Forest change detection and monitoring is a key part of the Copernicus Land Monitoring Service (CLMS) (https://land.copernicus.eu/). Various methodologies already implement near real-time (NRT) forest monitoring in tropical regions (e.g. Reiche et al., 2021) with the focus on timely detection of deforestation activities. However, there is not yet an operational pan-European product tracking forest dynamics at such temporal frequency, which has moreover the capability to separate also subtle disturbances of the tree canopy from signal noise. This kind of product is under demand by the user community, hence a new CLMS prototype on “Continuous Forest Monitoring”, with the goal to capture natural and human-induced forest disturbances by detecting tree cover vitality loss on a monthly basis is tested and implemented within the Horizon Europe project Evolution of the Copernicus Land Service portfolio (EvoLand). In a second instance, the feasibility to detect disturbance agents, i.e., (i) windthrow/storm damage, (ii) wildfire, (iii) insect infestations, as well as (iv) human-induced disturbances (e.g., forest clearing, clear-cutting, and thinning activities) is tested.
Dense time series from Sentinel-2 serve as main input for both prototypes, supported by forest masks from the CLMS High Resolution Vegetated Land Cover Characteristics (HRL VLCC) and ancillary data on forest disturbance locations and agents. From a benchmarking of various tools, the Exponentially Weighted Moving Average (EWMA) – proposed by Brooks et al. (2014) for Landsat time series data and implemented as part of the JRC-NRT tool (https://github.com/ec-jrc/nrt) – yielded the most promising results, especially considering the balance between accuracy, NRT capability, and computational effort. It is an unsupervised data-driven approach using univariate input indices to detect location and timing of disturbances. A supervised classification to derive the disturbance agents is added on top.
This study describes the implementation and results of this prototype and compares the detected forest disturbance locations and dates to the radar-based Tree Cover Disturbance Monitoring (TCDM) product and the 3-yearly VLCC forest change product. Two large EvoLand European sites were chosen for a first phase implementation: one in Germany (analysis years 2019-2021) and another in Spain (analysis years 2020-2022). The evaluation is carried by disturbance agent, concluding to different effects on either the physical structure of the trees and/or the spectral signal of the canopy, and hence also on the suitability of a method of detection. Products are delivered at pixel level (10m spatial resolution), improving the 20m resolution of the currently available CLMS forest change products, while increasing the change frequency from 3-yearly or yearly to monthly.
The resulting information can be utilized to enhance forest management and planning, aid forest-related decision-making or contribute to reporting on forest-related EU policies. These two prototypes are proposed within EvoLand to enhance the CLMS forest portfolio and to meet or go beyond users' requirements and demands.
How to cite: Moser, L., Grabatin-Homolka, A., Langner, A., Jahangir, F., Berndt, F., Wegscheider, S., Menini Matosak, B., Stumpf, A., Ruiz, I., Puhm, M., and Deutscher, J.: Large-scale Monitoring of Forest Disturbances – a Future CLMS Prototype, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18497, https://doi.org/10.5194/egusphere-egu25-18497, 2025.
Wind is among one of the most frequent drivers of forest disturbance around the world. Wind disturbance (blowdown, windthrow) results from particular meteorological conditions, where wind gust speed is a key factor. Blowdown is conditioned also by wind direction together with local topography, and influence the soil disturbance due to tree uprooting.
We mapped and analyzed the large-scale 2020 blowdown in the spruce-fir subalpine forest on the western slope of the Front Range, Rocky Mountains (Colorado, US). The area of interest (AOI) is a 9 x 29 km rectangle (39.80° N, 105.77° W and 40.06° N, 105.67° W) located south of the Rocky Mountain National Park and north of the Berthoud Pass. The mapping focused on developing and automating the workflow based on Sentinel 2 data and Change Vector Analysis (CVA) and comparing its output with the Global Forest Change (GFC) data. The CVA mapping is based on 1) the difference image computed using post- and pre-event images, and 2) the parameters calculated using two bands of the difference image: magnitude (mgt) representing the amount of change, and direction (drct) referring to the type of change. To create the CVA output, we used bands 11 and 12, together with 40° < drct < 47° and mgt > 0.1. Both CVA output and GFC data have a true positive rate (TPR) of 66-67%, with a false positive rate (FPR) of 0.9% and 3%, respectively. The CVA can be adjusted to achieve TPR up to 75.5%, of which FPR was 5.8%. Our approach is based on an unsupervised method, and open-source data, and is fully automated using R. Using CVA, the blowdown area was estimated to 1379.7 ha. The comparison between GFC data and CVA output revealed the higher efficiency of CVA for small patches with intensive damage. GFC data were better for indicating the location of patches with lower damage intensity.
We also aimed to capture different environmental insights related to the meteorological conditions causing the blowdown, soil disturbance patterns, and the impact of topography. Large-scale blowdowns are infrequent in the Rockies and are usually associated with the occurrence of unusual meteorological conditions. The blowdown was caused by strong easterly winds (gusts of 30 m•s⁻¹) blowing on September 7th - 9th, 2020, associated with the passage of a cold front causing exceptionally early late-summer cooling. The blowdown patches distribution generally followed the run of the valleys and ridges (SE-NW), with large patches in southern and central parts, and smaller ones in the northern part. The blowdown caused soil disturbance, with root plate volumes of 0.1 – 0.8 m3. The bearings of uprooted tree stems followed the direction of the main wind currents reported in the climate time series. Our approach can be valuable for research on blowdown mapping and triggering factors, GFC data assessment, soil disturbance, and interplays with relief.
The study has been supported by the Polish National Science Centre (project no. 2019/35/O/ST10/00032) and by the Polish National Agency for Academic Exchange (agreement no. PPN/STA/2021/1/00081/U/00001).
How to cite: Godziek, J., Pawlik, Ł., and Buma, B.: The 2020 windstorm forest damage in the Colorado Rocky Mts. - satellite-based mapping automation and environmental insights, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4582, https://doi.org/10.5194/egusphere-egu25-4582, 2025.
Extreme weather events pose substantial risks to forest ecosystems and forestry operations. Synthetic Aperture Radar (SAR) can address these challenges due to its ability to operate in all weather conditions and penetrate cloud cover. This study demonstrates, to the best of our knowledge, for the first time the potential of Sentinel-1 (S1) interferometric SAR (InSAR) coherence time series for rapid detection of windthrow-induced forest damage.
The study focuses on a severe storm near Otok (45°09′N, 18°53′E), Croatia, on 19 July 2023. We analyzed 84 forest plots, categorized into five damage classes: A – no damage (0–10%, 13 plots), B – minor (10–20%, 18), C – moderate (20–50%, 19), D – significant (50–80%, 22), and E – severe damage (80–100%, 12). Each plot represented a 50-meter radius area (~0.8 hectares).
Coherence magnitudes in VV and VH polarizations were calculated from consecutive image pairs for three S1 relative orbits (51, 73, 175). The pre-storm (25 June–18 July) and post-storm (1–24 Aug) periods were analyzed, each spanning 24 days and six S1 images (2 per orbit). Image pairs with second images from 19–31 July were excluded to avoid interference from the storm. Data were grouped by damage class, and statistical differences were assessed using the Mann-Whitney U test.
Post-storm, intra-group median VV coherence magnitudes differed significantly between no-damage and heavy-damage groups (e.g., A vs. D, and A vs. E). However, the coherence signal was near noise levels, reflecting the subtlety of the damage signature (Table 1). No significant differences were observed during the pre-storm period, underscoring VV coherence's sensitivity to storm-induced structural damage. VH coherence and VV and VH backscatter were not sensitive to windthrow.
This study highlights the potential of Sentinel-1 InSAR coherence in forest monitoring frameworks, supporting operational planning in forestry. The inclusion of Sentinel-1C will reduce the temporal baseline (from 12 to 6 days) further mitigating temporal decorrelation and enabling denser time series.
|
Table 1. Inter-group comparison (Mann-Whitney U test) and intra-group statistics. |
|||||||
|
|
Post-storm |
|
|
|
Pre-storm |
|
|
|
P-value (Significance ≤ 0.001) |
A |
D |
E |
|
A |
D |
E |
|
D |
0.0003 |
- |
|
|
0.5799 |
- |
|
|
E |
7.0e-06 |
0.0903 |
- |
|
0.5410 |
0.7479 |
- |
|
Group size |
78 |
132 |
72 |
|
78 |
132 |
72 |
|
Median coherence |
0.145 |
0.196 |
0.227 |
|
0.127 |
0.129 |
0.136 |
|
IQR of coherence |
0.096 |
0.132 |
0.141 |
|
0.103 |
0.094 |
0.108 |
How to cite: Tampuu, T., Buslavičiūtė, E., Gašparović, M., Pilaš, I., and Klobučar, D.: Detecting forest storm damage with multi-temporal Sentinel-1 InSAR coherence time series, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8125, https://doi.org/10.5194/egusphere-egu25-8125, 2025.
In the wake of extreme heat and drought events, excess tree mortality is increasing globally. While forest inventories provide valuable data for geolocating tree mortality, they are sparse and do not identify individual tree mortality. Aerial data captured by drones and airplanes provide precise centimeter-scale imagery that can be used to map individual tree mortality and fractional forest cover. The deadtrees.earth platform provides a comprehensive archive of annotated high-resolution orthoimages captured around the globe for different ecosystems and biomes. By using the imagery from deadtrees.earth, it is possible to detect and predict individual tree mortality and forest cover using high-resolution RGB orthoimagery at the regional scale. Here we present a methodology that allows generating global maps of tree mortality and fractional cover using satellite imagery from high-resolution aerial orthoimagery.
The Sentinel-2 satellite fleet, equipped with the MultiSpectral Instrument (MSI), covers the entire Earth within five days at spatial resolutions ranging from 10 m to 60 m. The Sentinel-1 satellite fleet offers global temporally continuous radar coverage that penetrates clouds. Tree mortality and forest cover reference data in diverse ecosystems is obtained by using multiple segmentation models on the globally distributed high-resolution aerial orthoimagery database deadtrees.earth. Spatio-temporal signatures of Sentinel 1/2 satellites are then translated into forest properties by using novel Transformer architectures. In this study, we highlight how to map the share of standing deadwood and forest cover at 10 m resolution annually, generalizing to all ecosystems. This will enable us to map tree mortality and forest cover at a global scale at a new level of precision.
How to cite: Mosig, C., Mahecha, M., Cheng, Y., Jehle, J., Junttila, S., Hartmann, H., Horion, S., Montero, D., Schwenke, M. B., and Kattenborn, T.: Mapping Fractional Tree Mortality and Tree Cover at Global Scale Using Sentinel-1 and 2, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18232, https://doi.org/10.5194/egusphere-egu25-18232, 2025.
Tree mortality rates are rising across many regions of the world. These are driven by the complex interplay of abiotic and biotic factors, including global warming, climate extremes, pests, pathogens, and other environmental stressors. Despite the urgency of understanding these dynamics, critical gaps remain in our ability to determine where trees are dying, why they are dying, and to predict future mortality hotspots. These knowledge gaps are primarily caused by missing data on tree mortality events. Ground-based observations, such as national forest inventories, are often sparse, inconsistent, and lack the spatial precision needed for comprehensive analysis. By contrast, satellite observations have a high temporal resolution, but their spatial resolution is often too coarse to identify individual trees. Earth observations combining drones and satellites, using machine learning, offer a promising avenue for mapping standing dead trees and uncovering the underlying drivers of tree mortality.
Here we introduce deadtrees.earth, an initiative focusing on multi-scale remote sensing of tree mortality across scales. At its core, deadtrees.earth curates the largest archive of centimeter-scale RGB aerial imagery of forests, with over 2,000 orthoimages representing diverse forest biomes across continents and major forest types. Using extensive annotations of dead canopies, we develop computer vision models capable of automated semantic and instance segmentation of dead tree canopies in RGB orthoimages. These model variants demonstrate robustness across varying resolutions, biomes, and forest types, and can be applied to any orthoimage imagery submitted to the platform, enabling users to exploit these tools for their analysis.
These local-scale predictions derived from drone and airplane imagery form the foundation for training satellite-based AI models to monitor tree mortality and forest cover on a global scale. We showcase recent advancements in spatiotemporal transformer models utilizing Sentinel-1 and Sentinel-2 data to produce global-scale, annual maps of forest cover and standing deadwood fractions at 10-meter resolution.
A recent key functionality of the deadtrees.earth platform is its web-based annotation tools, which allow users to contribute additional training data or provide feedback on existing predictions. This crowdsourcing functionality promotes community engagement, facilitating continuous improvement and fostering trust in the provided aerial image-based and satellite-based models and products.
Future work will also include expanding the coverage of aerial imagery, particularly in underrepresented regions such as Asia and Africa, which remains a cardinal priority to ensure the inclusivity and representativeness of the platform’s global-scale analyses. Moreover, we aim to apply the data products in a range of use-cases, ranging from attribution and forecasting of mortality to calibrating mortality in dynamic vegetation models. By bridging local and global scales, this work offers a critical tool for monitoring forest mortality trends, contributing to climate change impact assessments, and enhancing predictive capabilities for ecosystem resilience.
How to cite: Kattenborn, T., Mosig, C., Vajna-Jehle, J., Cheng, Y., Hartmann, H., Montero, D., Juntilla, S., Horion, S., Beloiu Schwenke, M., and Mahecha, M.: deadtrees.earth: tree mortality monitoring from local to global scales with AI and remote sensing, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19718, https://doi.org/10.5194/egusphere-egu25-19718, 2025.
Posters on site: Tue, 29 Apr, 10:45–12:30 | Hall X1
Virtual laser scanning (VLS) [1] has been used intensively for method development and machine learning in forestry, e.g., for quantification of leaf angle distribution [2], aboveground biomass prediction [3], or leaf-wood segmentation [4]. So far, these applications have been limited to mono-temporal VLS acquisitions where scenes were simplified to being completely static. However, forests are inherently dynamic environments with processes occurring at different timescales and rhythms, such as wind-induced movement, response to varying water potential, seasonal changes, stress-induced changes or growth.
Given the increasing availability of multi- and hyper-temporal point cloud data [5] as well as the potential of cloud-to-cloud data fusion [6], we propose virtual laser scanning of dynamic scenes (VLS-4D) [7] to develop methods for monitoring vegetation movement, tree health, and forest growth. Unlike real-world data collection, which is limited by time or equipment to one or a few scenarios, VLS-4D allows the creation of many different scenarios. This is achieved by combining different scene compositions and dynamics, acquisition modes and sensor settings. Furthermore, VLS-4D includes perfect reference data of the underlying scene, including semantic labels, geometry and changes (e.g., as deformation/movement values or change labels). Such data is usually difficult, time-consuming or impossible to obtain when working with real point clouds, or it is associated with considerable errors or unknown ambiguities. The scenario building capabilities, together with the availability of reference data, make VLS-4D a promising data generation tool for the ever-growing pool of deep learning methods for the analysis of forest point clouds and point cloud time series.
We distinguish three concepts of how dynamic scenes can be implemented in LiDAR simulation [7]:
a) Few static representations of the forest scene at different epochs, e.g., one before and one after a forest disturbance event.
b) Many static snapshots sampled from an animated scene, e.g., daily snapshots to simulate a permanent laser scanning setup.
c) Animation within the scene, e.g., vegetation moving in the wind during a single terrestrial laser scan.
We will present simulation workflows for each of these concepts using the open-source software HELIOS++ [8], from animated 3D scene generation in Blender to final simulated point clouds and point cloud time series. With these simulation examples, we illustrate the research gaps that can be filled by such virtual experiments, address strategies and challenges in implementing VLS-4D, and discuss future directions. We expect VLS-4D data to play an essential role in the development of innovative methods for forest monitoring, complementing the still limited and typically unlabelled real-world multitemporal datasets.
References:
[1] Winiwarter, L., et al. (2022): DOI: https://doi.org/10.1016/j.rse.2021.112772
[2] Liu, J. et al. (2019): DOI: https://doi.org/10.1016/j.isprsjprs.2019.01.005
[3] Schäfer, J. et al. (2023): DOI: https://doi.org/10.1093/forestry/cpad061
[4] Esmorís, A. et al. (2024): DOI: https://doi.org/10.1016/j.isprsjprs.2024.06.018
[5] Eitel, J.U.H. et al. (2016): DOI: https://doi.org/10.1016/j.rse.2016.08.018
[6] Balestra, M. et al. (2024): DOI: https://doi.org/10.1007/s40725-024-00223-7
[7] Weiser, H. & Höfle, B. (2024): DOI: https://doi.org/10.31223/X51Q5V
[8] HELIOS++: https://github.com/3dgeo-heidelberg/helios
How to cite: Weiser, H., Albert, W., Tabernig, R., and Höfle, B.: Virtual Laser Scanning of Dynamic Scenes (VLS-4D): A Novel Opportunity for Advancing 3D Forest Monitoring, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-1560, https://doi.org/10.5194/egusphere-egu25-1560, 2025.
This study leverages deep learning to enhance the identification of trees, shrubs, grasses, and other vegetation as distinct plant growth forms, which are essential for deriving vegetation structural properties. These properties are crucial for improving the identification of Fractional Vegetation Cover (FVC) components and the structural elements contributing to FVC, such as separating non-photosynthetic vegetation (NPV) on the ground (litter and coarse woody debris) from standing wood or distinguishing grasses, shrubs, and tree canopies by incorporating 3D data. Fractional Vegetation Cover (FVC)—comprising green vegetation (GV), non-photosynthetic vegetation (NPV), and bare earth (BE)—is particularly challenging to discriminate and map in centimetre-scale Unoccupied Aerial Systems (UAS) imagery due to spectral similarities and environmental variability.
To address these challenges, this study combines LiDAR voxel data, transformed into 2D raster representations, with multispectral imagery. Each raster channel encodes aggregated attributes such as mean voxel height, point density, maximum height, and intensity. These inputs serve as the foundation for a 2D U-net deep learning model trained using reference datasets from Calperum Station in a semi-arid ecosystem in South Australia. By incorporating canopy and ground elements, such as NPV (e.g., litter and coarse woody debris), this approach aims to enhance the model’s capacity for accurate FVC classification.
Initial experiments yielded promising results. Site-specific models achieved high overall accuracies exceeding 89% and F1 scores above 0.9, but their performance declined to approximately 69% in dense vegetation. For the generic model, accuracy dropped further to 28.48%, highlighting significant challenges in generalisation across diverse vegetation types. These findings underscore the limitations posed by complex environments, limited reference data, and the low frequency of NPV as a minority class. To address these issues, further advancements are proposed, including integrating additional LiDAR data, expanding training datasets, and employing data augmentation techniques. Data augmentation, in particular, can address environmental and illumination variability, improving the model’s ability to learn underrepresented classes and increasing robustness across diverse ecosystems.
The anticipated outcomes include improved identification of plant growth forms, with the potential for more reliable estimates of vegetation structural metrics. These advancements support derived estimates of aboveground biomass, enhanced water content assessment, and the evaluation of other critical ecosystem services. This framework leverages voxel-projected features to support vegetation analysis and improve classification performance. Additionally, it aims to enable high-resolution mapping of FVC components under plant growth form, bridging the gap between fine-scale UAS observations and regional-scale satellite imagery to support ecosystem monitoring.
How to cite: Sotomayor, L., Lucieer, A., Turner, D., Lewis, M., Levick, S., and Kattenborn, T.: Deep learning for identification of 3D plant growth forms in Fractional Vegetation Cover, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3984, https://doi.org/10.5194/egusphere-egu25-3984, 2025.
Advancements in remote sensing technologies have enabled comprehensive and detailed forest mapping, as demonstrated by recent initiatives in Slovenia. Such maps are essential for sustainable forest management, biodiversity conservation, and monitoring changes in forest structure and composition over time. A forest stand map was generated for the year 2023 in central Slovenia, encompassing 7,005 km², 62% of which is forested. We developed and calibrated two distinct models based on Sentinel-1 SAR data to map growing stock and the proportion of coniferous and broadleaved species in growing stock. In addition, we used the LiDAR-based canopy height model (CHM) to map forest stand height and canopy cover.
LiDAR data acquisition occurred in spring 2023, coinciding with varying levels of leaf development across deciduous forests. This led to heterogeneity in the point cloud data, affecting CHM-based estimations of forest stand height and canopy cover. Tree-top CHM heights were relatively unaffected, but the crown shapes of deciduous trees were heavily influenced by the state of leaf development. To mitigate these effects, the CHM’s horizontal resolution was reduced by aggregating the highest point within each 10-meter pixel, downsampled from the original 50 cm CHM. Forest stand height was calculated as the mean height of all 10-meter CHM pixels within forest stand polygons. Canopy cover was derived as the percentage of 10-meter pixels exceeding a height threshold of 20 meters.
Growing stock and the proportion of species were estimated using random forest models trained on field-measured forest stand data, and Sentinel-1 imagery. Field data were provided by the Slovenian Forestry Service and included forest stand data for quasi-randomly distributed forest management units. Vegetation indices were derived from Sentinel-1 daily data, including the Radar Vegetation Index (RVI = 4 × VV / (VV + VH)), the Normalized Radar Vegetation Index (NRVI = (VV - VH) / (VV + VH)), and the Radar Forest Degradation Index (RFDI = VV - VH), where VV and VH represent Sentinel-1 polarization modes. Monthly composites of these indices, spanning January 2022 to December 2023, were generated and smoothed using a 3 × 3 low-pass filter.
The random forest models, consisting of 100 regression trees each, were optimized based on R2 performance on unseen test data generated during the cross-validation process. The optimal tree depths were 20 and 15 for the growing stock and the proportion of species models, respectively, yielding R2 values of 0.34 and 0.57. Final model-based predictions were aggregated to forest stand polygons, providing spatially explicit estimates of growing stock and species composition.
Forest stands were delineated using the Segment Mean Shift image segmentation tool in ESRI ArcGIS Pro, applied to a set of Z-score standardized raster maps representing forest stand height, canopy cover, and the proportion of coniferous and broadleaved species. Further work is planned to include lidar data and optical data into the models.
How to cite: Jevšenak, J., Skudnik, M., and Kobler, A.: Mapping Forest Stand Characteristics Using Aerial LiDAR and Sentinel-1 Data: A Case Study from Slovenia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5526, https://doi.org/10.5194/egusphere-egu25-5526, 2025.
Accurate modeling of subcanopy solar radiation is vital for ecological modeling, forest management, and urban planning, as it influences vegetation growth, energy balance, and environmental dynamics. This study provides a comprehensive evaluation of two solar radiation models: PCSRT and r.sun, leveraging LiDAR datasets from terrestrial (TLS), unmanned aerial (ULS), and airborne (ALS) scanning. The results demonstrate that the choice of modeling approach and data source substantially impacts the accuracy of solar radiation estimates, particularly in complex forested environments. PCSRT, with its voxel-based 3D modeling, excels in capturing intricate subcanopy radiation dynamics, especially when combined with high-density LiDAR data such as TLS and ULS. In contrast, the raster-based r.sun model, while computationally efficient and scalable, is better suited for broader regional analyses, particularly in less heterogeneous environments such as urban areas. This research underscores the critical role of LiDAR data density in determining model accuracy, with ULS providing the most reliable results, TLS capturing detailed local variations but facing coverage limitations, and ALS offering scalability but with reduced precision in dense canopy structures. Practical implications of this study include tailored recommendations for selecting modeling tools and LiDAR datasets based on the objectives and spatial scale of the study. PCSRT is recommended for high-resolution ecological studies requiring detailed subcanopy analysis, whereas r.sun is preferable for large-scale applications where computational efficiency is prioritized. However, limitations of each approach are acknowledged, including the computational intensity of PCSRT and the lower precision of r.sun in capturing canopy interactions.
Keywords: subcanopy solar radiation, solar radiation models, LiDAR, r.sun, PCSRT
How to cite: Buchalová, D., Hofierka, J., Šupinský, J., and Kaňuk, J.: Comparative modeling of subcanopy solar radiation: Evaluating PCSRT and r.sun with multi-source LiDAR data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5816, https://doi.org/10.5194/egusphere-egu25-5816, 2025.
Analyzing forest conversions and unraveling their drivers is a significant challenge for ecological research. The forests of northeastern China, including major temperate forest types in East Asia, constitute key ecological function zones. In this study, we investigated the changes in fractional forest cover (FFC) in Northeast China over the past 37 years using Landsat time series data and explored the underlying mechanism of FFC response to environmental changes with two methods: the partial correlation analysis between different time series and a SHAP-based explainable CatBoost algorithm at the point level. First, a probability-based random forest model was developed to classify forest and non-forest areas, achieving an average overall accuracy of 85% from 1987 to 2023. Pixel-by-pixel tracking of forest conversions over ten-year intervals revealed a significant net increase of more than 16×107 forest pixels. Notably, a declining trend in forest expansion (lost) was observed along altitude gradients above 200 m during four epochs. In the analysis of driving forces, our results indicated that FFC was positively correlated with temperature in 65% but negatively correlated with precipitation in 63% of the study region. Moreover, the impacts of temperature on increasing and decreasing FFC were contrasting: at altitudes above 200 m, increasing FFC exhibited a positive partial correlation with temperature, while decreasing FFC showed a nearly negative correlation with temperature. Furthermore, the predictive CatBoost model explained 58% of the increase in FFC, attributing it to variability in meteorology (mean annual temperature and precipitation), mean annual soil moisture, population density, and elevation. In comparison, these environmental factors accounted for 38% of the decrease in FFC. For total FFC changes (both increases and decreases), the optimized model achieved a precision of 40%. Among the driving factors, mean annual temperature played a predominant role in accounting for both total FFC changes and FFC increases. Meanwhile, mean annual precipitation was the most critical indicator triggering FFC decreases. Our findings provided valuable evidence and insights into the relationships between environmental factors and FFC dynamics in the context of sustainable development.
How to cite: Wang, X., Huang, F., and Gao, L.: Exploring spatiotemporal pattern of fractional forest cover in Northeast China from 1987 to 2023: Unraveling its change-driven factors, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6074, https://doi.org/10.5194/egusphere-egu25-6074, 2025.
Land Use/Land Cover (LULC) monitoring is essential for understanding Earth's surface dynamics, particularly in assessing the impact of vegetation and land changes on hydrological systems, vulnerability to extreme climatic events, and forest health. In recent years, increasing forest dieback caused by climate change, pests, and diseases has raised global concerns about ecosystem stability and biodiversity. The rapid spread of tree mortality and the need to accurately capture its temporal evolution highlight the necessity of precise detection and monitoring of dead tree areas, which are crucial for effective forest management and mitigating related environmental impacts.
To address this, our study aimed to map LULC and identify areas of dead trees from 2018 to 2023 in the Harz Mountains and its surrounding regions in Germany, an area severely affected by bark beetle infestation. For this purpose, we trained a multi-modal U-Net architecture, a supervised learning model with an encoder-decoder structure designed to capture contextual features across multiple scales. The training applied multi-temporal optical (Sentinel-2) and radar (Sentinel-1) imagery acquired during the growing season (May-August) of 2020-2021 as the training dataset, with ESA 2020/2021 data, tree species distributions from the Thünen Institute of Forest Ecosystems, and manually annotated dead trees as the reference dataset. Annual LULC maps for 2018-2023 were generated by processing each image using the trained model and subsequently combining the predictions per image using a majority voting approach, considering seven LULC classes: cropland, grassland, built-up areas, water bodies, coniferous, deciduous, and dead trees. Furthermore, a change analysis was performed on the predicted maps from 2018 to 2023.
Accuracy assessment demonstrated the model’s robust performance, with an overall accuracy of 0.88. Additionally, a comparison between a European LULC map -ELC10- and our predicted LULC map for 2018 resulted in an overall accuracy of 0.86, further highlighting the reliability of this method. Among the classes, cropland achieved the highest F1-score (0.97), likely due to the higher number of training samples available (40% of the total training samples). In contrast, the dead tree class demonstrated the lowest F1-score (0.60), attributed to its limited sample size (1% of the total training samples) and confusion with coniferous trees. The model effectively mapped the other classes, with F1-scores exceeding 0.70. The analysis revealed an increase in dead trees and grassland areas, primarily at the cost of coniferous trees, which can be linked to tree mortality caused by bark beetle infestation and prolonged drought, particularly from 2018 to 2022. It also revealed deforestation patterns between 2018 and 2023, with dead tree areas initially concentrated near Brocken in the Harz Mountains. Over time, these areas steadily expanded from the southeast towards the western and central parts of the study area.
These findings, based on freely accessible satellite data, can support forest managers in monitoring landscapes and tree mortality and help identify effective control measures.
How to cite: Saba, F., Achanccaray, P., and Gerke, M.: Land use/Land cover mapping to monitor dead tree areas using multi-modal, multi-temporal Remote Sensing imagery and a deep learning model, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6990, https://doi.org/10.5194/egusphere-egu25-6990, 2025.
As one of the resilient and largest land biomes on Earth, boreal forest (comprising 30% of the global forest area) provides ecosystem services that benefit society at levels ranging from local to global including climate regulations. Nonetheless, stoked by arctic warming, wildfires are remaking the boreal forest. In recent decades’ boreal wildfires' extent and severity increased and reached record levels. In the last two decades (2001-2020: 31.4 million acres) wildfires in Alaska have burned more than twice as many acres than the previous two decades (1981-2000: 14.1 million acres). Severe fires lead to the most extensive regrowth of broadleaf shrubs and trees. Previous studies predict that broadleaf stands, which accounted for less than half of interior Alaska’s forests in 2001, will expand to cover two-thirds of the forested area. Some studies predict forests being permanently replaced by shrubland and grassland. Whatever the new shape of the forest, the change will ripple through wildlife. Given consensus on the value of boreal forest to the climate system, biodiversity, and society, data and policy-driven improvement in forest protection and management is needed. Current and reliable map products and up to date post-fire forest demography will be valuable to new policy formulation to sustain forest cover, and reduce fire risk. Given that fire frequency and severity are expected to increase the prevalence of early-successional broadleaf species, it is particularly important to monitor forest demography to better understand how changing climate conditions and wildfires are affecting overall forest health, resilience, and carbon drawdown. Post-fire forest cover changes from one type to another need to be mapped and documented every 2-5 years to support effective forest protection and management efforts. Satellite imaging provides a consistent, enduring record of the landscape, and repeated imaging has potential to map forest recovery and demography post-fire. In this study, we investigate the post-fire forest demography within select historic burn scars from the 1980s using satellite remote sensing. Our goal is to gain novel insights on post-fire forest recovery and composition i.e. post-fire what percentage of a burn scar is conifer vs broadleaf and how do their composition evolve with time? For a select historic fires from the 1980s, we will employ a time-series analysis of post-fire vegetation recovery by species at 2-5 years interval. We will use peak growing season spectral indices (NDVI, NBR) along with spring (leaf off) images for mapping vegetation by spices. We will use Random Forest image classifier to generate the final vegetation maps and composition statistics. The research will offer novel insights on post-fire forest recovery and composition, and its findings will provide a locally relevant record of forest change by (i) being spatially explicit, (ii) quantifying gross forest loss and gain, and (iii) quantifying trends in forest demography. The derived map products and statistics will empower the U.S. Forest Service, Alaska DNR, and private landowners to take measures for effective management of forest land and resources to sustain ecosystem services benefiting society and climate regulations.
How to cite: Panda, S.: Post-Fire (1983-2024) Boreal Forest Demography, Interior Alaska, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14242, https://doi.org/10.5194/egusphere-egu25-14242, 2025.
Infrastructure development and environmental conservation are critical challenges in modern engineering. Although pipeline refurbishment projects are essential for maintaining the safety and reliability of energy transportation, can often lead to considerable environmental impacts. With our approach in Pieve di Soligo, Salgareda natural gas pipeline in the Veneto region (Italy) we adopted a sustainable approach to balancing industrial needs with ecological preservation. The pipeline route spans over 37 km across 12 municipalities. The pipeline company implemented a vegetative restoration plan across all the area, focusing on diverse interventions to ensure effective environmental recovery, employing grass planting techniques (hydraulic seeding and manual sowing) as well as reforestation, to reestablish forest or herbaceous cover. For restoration interventions, they planted tree and shrub species matching those present before the project began, placing protective measures such as anti-wild boar wire mesh, to mitigate external stressors. Moreover, they established a five-year cultivation care program, including irrigation, fertilization and pruning to support long-term success. To assess the effectiveness of these interventions, we employed the mobile laser scanner (MLS) FARO Orbis, equipped with SLAM (Simultaneous Localization and Mapping) technology, to capture high-resolution 3D data at four monitoring stations, mapping a total of 2300 m2 in approximately 25 minutes with a loop-close path. We conducted the surveys in July and October, obtaining a multi-temporal dataset which allows us for comparative analysis of trees growth over four months. The monitored areas included varied landscapes, from flat agricultural fields dominated by vineyards to moderately sloping wooded regions. We process the data using both FARO Connect and CloudCompare software. We aligned the two point clouds, acquired in different periods from the same area, by using the align command and then we isolated the monitored vegetation using the segment command. By using the CSF Filter plugin, we created the terrain meshes and from them we normalized the point clouds, obtaining the vegetation heights. We analyzed the tree growth patterns by measuring the differences in tree heights in the two survey periods, obtaining an index of the restoration’s effectiveness. Although we achieved vegetation-height assessments, it was not possible to extract variations in diameter at breast height (DBH) because the protective barriers around the trunks obstructed the LiDAR beams. Thanks to the MLS surveys, performed in different time of the year, we can effectively monitor the recovering process and understand if the choices made in the field are giving the expected results. This capability can also facilitate rapid corrective actions, where and if necessary. This study underscores the importance of integrating ecological principles with modern technological methods in infrastructure projects, which allow to extract measurements or observations that are difficult to obtain using traditional surveying techniques. Results from the analysis demonstrate effective recovery of vegetation, offering valuable insights into the long-term sustainability of restoration interventions.
How to cite: Pierdicca, R., Balestra, M., Moschini, D., and Bonacoscia, M.: Integrating ecological restoration and multi-temporal mobile laser scanning in a natural gas pipeline refurbishment: a case study in Veneto, Italy., EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17980, https://doi.org/10.5194/egusphere-egu25-17980, 2025.
Above Ground Biomass (AGB), the total dry biomass found above the ground, plays a vital role in understanding the global carbon cycle and biodiversity. Recognized by international organizations as an Essential Climate Variable, AGB is a key component for carbon accounting and climate modeling. Despite its importance, accurately estimating AGB remains a challenge.
Allometric models have long been a central focus of research due to their critical importance in estimating variables such as AGB based on the relatively easy-to-measure single-tree parameters such as Diameter at Breast Height (DBH) and Tree Height (TH). This led to the development of numerous species- and biome-specific allometries. Many of these models are accessible through dedicated online platforms or published scientific studies. However, their derivation is resource-intensive, and they exhibit significant variability across different species and ecosystems, both limiting their broader applicability.
Terrestrial Laser Scanning (TLS), provides a non-destructive and highly accurate method for estimating AGB through volume calculation. TLS-generated point clouds can be processed into Quantitative Structure Models (QSMs) by fitting a hierarchy of cylinders to the 3D data, enabling precise AGB estimation. Additionally, these QSM-derived tree volumes can be used to optimize parameters for allometric models.
In this contribution, we explore the application of a novel toolbox to derive allometric models for diverse forest environments and species. The toolbox was employed to generate highly accurate single-tree volume measurements, which were combined with traditional measurements of DBH and TH to develop finely tuned allometric models. A key focus of the research is the investigation of an integrated workflow for enhancing traditional forest inventory practices. This workflow combines TLS-derived QSMs with in-situ measurements of DBH and TH, which, as demonstrated in various studies, can also be increasingly reliable obtained using smartphones. This approach introduces new possibilities for studying and monitoring AGB in forests with greater efficiency and broader accessibility.
How to cite: Wild, B., Özkan, T., Pöppl, F., Milenković, M., Hofhansl, F., Lamprecht, J., Pfeifer, N., and Hollaus, M.: Efficient derivation of allometric models using laser scanning for improved AGB estimations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18889, https://doi.org/10.5194/egusphere-egu25-18889, 2025.
The increasing global availability of dense point clouds provides the potential to better capture complex environments and their changes, e.g., forests state and growth, river erosion processes, city planing, etc. An important process for turning point clouds into useful datasets is their classification. The quality of classified point clouds relies on four critical processing steps: neighborhood definition, feature extraction and selection, quality of training data and classification model. Determining the optimal neighborhood for each point is essential for capturing local information, fast calculation, enhancing feature richness, and improving the quality of downstream processes.
We propose a novel method for constructing neighborhoods for geometric feature calculation using a kd-tree-based region-growing approach. We construct neighborhoods by selectively adding points guided by local point connectivity, normal orientations, and distance from the seed point. In particular, the local connectivity is determined by a nearest-neighbor graph, parameterized to connect only points belonging to the same object. Following this graph, points are added iteratively to the neighborhood if the angular difference between their normal orientations lies below a locally derived tolerance threshold. The growing process is limited by the distance from the seed point. The new neighborhoods maximize information gain while minimizing boundary crossings between classes, e.g., ground to wall (normal orientation) or branches to buildings (connectivity constraint). Our results demonstrate that this approach outperforms the classical spherical neighborhood and good classification results are more resilient to changes of the neighborhood size.
Our analysis focus on the classification of urban areas, including ground, building, vegetation and other classes. We evaluate performance using datasets from various platforms, including airborne, mobile, and UAV systems and across different areas such as Berlin, Potsdam, and Paris. The effects of sensor characteristics and point-cloud densities are investigated as well as the improvement of individual features.
How to cite: Hess, M., Rheinwalt, A., and Bookhagen, B.: Optimizing point-cloud neighborhood calculation for classification, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-15790, https://doi.org/10.5194/egusphere-egu25-15790, 2025.
Forest and wood procurement planning requires detailed information about the quantities and characteristics of individual trees within forest stands. Stand-specific forest inventory data is increasingly being produced using remote sensing combined with accurately measured sample plots. In recent years, close-range sensing techniques such as terrestrial laser scanning (TLS) have been recognized as effective tools for providing precise measurements of tree characteristics—including features that cannot be directly measured nondestructively. To replace conventional field inventory methods using calipers and clinometers, there is a growing need for agile techniques that enable rapid and accurate measurements of all trees within inventoried sample plots. Mobile laser scanning (MLS) offers enhanced data acquisition speed by enabling detailed point cloud reconstructions of the surrounding forest environment on the move, making it an attractive technology for operational forest surveying, particularly for measuring forest sample plots. Previous studies have demonstrated the potential of MLS for individual tree characterization, but there remains a gap in understanding how its tree-level performance translates to plot-level accuracy under varying boreal forest conditions, where the presence of evergreen foliage often poses challenges for tree characterization.
The aim of this study was to evaluate how accurately MLS can measure forest stand attributes such as mean basal area (BA), tree density (number of trees per hectare; TPH), and basal area-weighted mean tree diameter and -height (Dg, Hg). Additionally, we investigated the scanning setups required to achieve accurate measurements of stand attributes across different forest types. The study was conducted in Heinävesi, Finland, where 50 plots (typically 30 m x 40 m in size) were measured tree by tree (n = 5227) in the field during the autumn of 2023. MLS data from these sample plots were collected using the Faro Orbis scanner in the summer of 2024. Trees were identified and their dimensions extracted from the point clouds, with plot-level forest stand attributes aggregated and compared to those measured using traditional caliper and clinometer methods.
Experiences from the data acquisition campaign highlighted the ease of MLS-based forest surveying, enabling agile data collection. Sample plots ranging from 370 to 2000 m² were captured within an average of 21 minutes, although more complex forest structures and walk paths increased the required time. Preliminary results indicate that, using semi-automatic tree detection methods, approximately 99.6% of trees with diameters greater than 5 cm were successfully identified. Diameter at breast height (DBH) and tree height were measured with RMSEs of 15.7% (2.5 cm) and 12.03% (2.9 m), respectively. At the plot level, these measurements provided unbiased estimates of basal area (G) and trees per hectare (TPH), while slightly overestimating Dg and Hg in more complex forests. These findings underscore the potential of MLS for operational forest inventory measurements.
How to cite: Yrttimaa, T., Liikonen, L., Erkkilä, A., Paakkari, J., Kotivuori, E., and Vastaranta, M.: Measuring Forest Inventory Attributes Using Faro Orbis Mobile Laser Scanner in Managed Boreal Forests, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18099, https://doi.org/10.5194/egusphere-egu25-18099, 2025.
Non-stand-replacing forest disturbances are increasingly threatening Europe’s forests under climate change. Monitoring and mapping of these disturbances remain a challenge in remote sensing due to the small magnitude of change signals. We present a detection method for satellite image time series analysis based on the Kalman filter and the Neyman-Pearson lemma. The method (1) amplifies the spectral change signals from abrupt forest disturbances in time series data, and (2) compares the amplified change signal to a prior expectation. Through these improvements, detection performance is greatly improved, with initial results from six study areas across Finland showing an F1-score of 0.7 for non-stand-replacing disturbances. Stand-replacing disturbances are detected by this method at an equal rate as the European Forest Disturbance Atlas and the Stochastic Continuous Change Detection methods. We demonstrate the theory behind this detection method along with initial results, sensitivity to different priors and potential for further improvement.
How to cite: Schraik, D., Seppänen, A., and Packalen, P.: Bayesian detection of non-stand-replacing forest disturbances in satellite image time series, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9165, https://doi.org/10.5194/egusphere-egu25-9165, 2025.
Forests play a vital role in providing diverse ecosystem services, including recreational opportunities for the local population, climate regulation, and timber production, as well as by supporting biodiversity. In parts of Central Europe, the primary function of these forests is to provide a stable and sustainable income for foresters and forest owners. Regardless of the focus, the forests’ ability to provide these services are increasingly threatened by the effects of climate change through extreme events like droughts or floods and biological calamities caused by pests like the European spruce bark beetle. To ensure continuing forest health forest personal is required to maintain dense monitoring in the field in order to act against these dangers. However, such close monitoring using conventional methods can be very time consuming and difficult from the ground. To combat this, this work attempts to get a detailed overview of the economic value of a forest based on individual trees based on UAV remote sensing. Our previous work has proven the effectiveness of UAV LiDAR remote sensing for the delineation of tree stems and their diameter under ideal UAV flight parameters. Building on this, we present a framework combining UAV LiDAR and multispectral data to estimate individual tree value based on diameter, straightness of stem, tree height, and species. Further, the difficulty of harvesting can also be accounted for using information about terrain, density of understory vegetation, and distance to forest and logging roads. This method can further be used to analyse possible areas of increased economic risk for biological pests or extreme events. Overall, this would substantially reduce the amount of fieldwork necessary by foresters and allow for a much more accurate and less tedious method of ensuring continued economic and ecological prosperity.
How to cite: Eisenschink, P., Obermeier, W., Zerres, V., Suerbaum, A., and Lehnert, L.: A Framework for Assessing Tree Value and Forest Vulnerability Using UAV Remote Sensing , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10908, https://doi.org/10.5194/egusphere-egu25-10908, 2025.
The mission of this study is to improve the accuracy of leaf area index (LAI) using ground-based forest inventory 'bottom-up' LAI with satellite-derived 'top-down' LAI estimates. Specifically, we compared LAI values obtained using allometric equations applied to over 30,000 trees in the Austrian National Forest Inventory (NFI) with satellite-derived LAI estimates from MODIS (Moderate Resolution Imaging Spectroradiometer) and Sentinel data sets (Sentinel-3 TOC reflectance and PROBA-V). Our results indicate that satellite-derived LAI estimates often underestimate the actual LAI observed in terrestrial data. This discrepancy is mainly due to the inability of remote sensing technologies to account for the Crown Competition Factor (CCF), which significantly influences canopy structure. As LAI is a critical parameter in ecosystem modelling, accurate LAI estimates are essential for reliable model outputs. To address this issue, we developed a logistic correction function by incorporating bottom-up and top-down LAI to improve the accuracy of LAI estimates for a sustainable forest management.
How to cite: Sinan, M. and Hasenauer, H.: Improving leaf area index (LAI) estimation by integrating forest inventory and remote sensing, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-4928, https://doi.org/10.5194/egusphere-egu25-4928, 2025.
The AI consultancy of the KIDA project is a collaboration between seven institutions within the area of the German Federal Ministry of Food and Agriculture. The members of the AI team solve service requests from the participating institutions on a cross-institutional basis. An advisory project is presented in which the AI team has trained a recurrent neural network (LSTM) to classify different tree species (e. g. deciduous trees, conifers) and tree species groups (e. g. oak, fir, ...) on time series of satellite images of sentinel-2. Several challenges as for example cloud covers had to be overcome. Finally, an accuracy of 97.9% was achieved for the classification of tree species, 98.4% for conifers and 91.6% for deciduous trees. The results show, how promising it is to carry out corresponding data collections in the future with the help of satellite data and AI to be able to recognise changes in the tree population quickly and efficiently in order to be able to react to them. In particular, large areas with relatively small sections (10m x 10m) could be monitored automatically. This opens up new opportunities in a rapidly changing world.
How to cite: Schneider, M.: The KIDA AI consultancy: Tree detection with satellite data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16720, https://doi.org/10.5194/egusphere-egu25-16720, 2025.
Biodiversity monitoring is a critical global issue, requiring reliable and precise information on forest and tree attributes to ensure sustainable management and biodiversity conservation. Remote sensing (RS) technology has emerged as a powerful tool for forest monitoring, offering significant advantages over traditional methods. The advent of advanced LiDAR technologies has revolutionized the field, enabling high-resolution 3D data collection and capturing intricate forest structures. Despite advancements, efficiently monitoring dense and complex forest environments in three dimensions remains a challenging task. This study in processing and analysing Simultaneous Localization and Mapping (SLAM) and Terrestrial Laser Scanning (TLS) LiDAR datasets to estimate biodiversity relevant attributes in Greek Natura 2000 (N2K) forested areas. The study is implemented as part of the hELlenic BIOodiversity Information System (EL-BIOS). The EL-BIOS is the first national-scale EODC infrastructure, with the aim of advancing EO data and products use for biodiversity management and conservation over Greece. This research encompasses three 0.1 ha plots, distributed in two distinctive protected areas: the Kotychi–Strofilia National Park in south Greece and the Northern Pindos National Park in north Greece. Open-source tools such as LAStools and 3D-Forest were utilized for individual tree segmentation and the calculation of key forestry parameters. Optimal algorithm configurations and functional tools were explored to compute structural attributes such as tree height, diameter at breast height (DBH), and crown metrics. To evaluate the performance and accuracy of the SLAM and TLS datasets, the automatically derived parameters were compared against traditionally collected in-situ data using classification metrics, accuracy statistics, and precision measures. The findings indicate that both SLAM and TLS effectively captured detailed 3D point cloud data of the forested plots, albeit with differences in accuracy, resolution, and acquisition time. TLS consistently delivered higher-resolution data but required extended processing times, stationary positioning, and manual repositioning within the plot area. Conversely, SLAM offered greater mobility and efficiency, albeit with slightly lower resolution. TLS achieved an accuracy of approximately 75% in tree detection, while SLAM ranged between 60% and 65%, demonstrating its operational viability despite a slight reduction in precision. Overall, this study underscores the potential of advanced 3D modelling techniques and efficient parameter extraction methods for biodiversity relevant information extraction over forest protected areas.
How to cite: Alexandridis, V., Karolos, I.-A., Bellos, K., Tsioukas, V., Diamantopoulou, M., Chrysafis, I., and Mallinis, G.: Comparative analysis of SLAM and TLS LiDAR technologies for biodiversity relevant information extraction over two Natura 2000 Sites in Greece, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5247, https://doi.org/10.5194/egusphere-egu25-5247, 2025.
Boreal drained-peatland forests provide a wide range of interlinked and overlapping biodiversity and ecosystem services (BES). Mapping BES is essential for informed decision-making to manage and conserve forest resources effectively. Centimeter-level resolution is often deemed necessary for mapping BES in complex landscapes like northern boreal drained peatland forests. However, systematic tests on the optimal spatial resolution, data combinations, and the impact of including or omitting specific datasets on mapping accuracy remain limited. In this study, we mapped several indices of BES, including biodiversity conservation, habitat suitability, non-timber forest products (e.g., bilberry and cowberry yield), scenic beauty, timber production and carbon storage, using multiple remote sensing (RS) data. These data sources included airborne laser scanning, unmanned aerial vehicle (UAV) data, and optical satellite data from PlanetScope, Sentinel-2, and Landsat 8-9. The specific objectives were: (1) to identify the best RS predictors for individual BES; (2) to develop random forest regression models for predicting BES; (3) to compare the performance of difference RS data and (4) to upscale the pixel-level distribution of BES across different canopy covers closed, partial, and open. Our preliminary findings indicate that BES can be accurately predicted using a set of height, density, and multispectral predictors, with explained variance ranging from 13% to 90% for individual BES. Model performance varied among individual BES and across different RS data sources. Furthermore, we successfully upscaled the BES predictions to map the spatial distribution of BES across the entire study area. Areas with closed canopies exhibited higher BES potential compared to partial and open canopies. These results demonstrate that RS data can be effectively used to predict BES on a spatial scale, providing a valuable tool for sustainable forest management.
How to cite: Rana, P., Chowdhury, A. I., Keränen, K., and Balazs, A.: Data and resolution requirements in mapping biodiversity and ecosystem services in boreal drained peatland forests, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9599, https://doi.org/10.5194/egusphere-egu25-9599, 2025.
The locally invasive insect pest Gonipterus sp. n. 2 (Coleoptera: Curculionidae) threatens Eucalyptus plantations, causing defoliation and yield loss through adult and larval feeding. Early detection is important for early intervention to prevent pest outbreaks. As conventional insect pest monitoring methods are time-consuming and spatially restrictive, this study assessed the potential of UAV monitoring. Multispectral imagery was obtained with Unmanned Aerial Vehicles (UAVs) in South Africa’s Midland region across seven different sites in 14 datasets of young stands of Eucalyptus dunnii with varying levels of Gonipterus sp. n. 2 infestation. Reference damage levels were obtained through visual assessments of (n= 80-100) trees at each site. Across sites, a decrease in canopy reflectance in both the visual and the near-infrared domains with increasing damage levels was consistently observed. Several vegetation indices showed consistent patterns, but none showed site independence. XGBoost was used to predict damage levels. The best-performing models included reflectance, vegetation indices and grey-level co-occurrence matrix data. When data from a 10-band multispectral camera were used, the highest classification accuracy was 90% across all sites in classifying defoliation levels. With a classical 5-band multispectral camera, accuracy was 82%, but distinguishing medium damage from absence remained challenging. Regardless the sensor, the method was less reliable when the training and validation sets were completely separated. This study highlights the potential of UAV-based multispectral imagery to assess Gonipterus sp. n. 2 damage, demonstrating reliable upscaling from individual tree assessments to stand scale. However, larger training datasets across multiple damage levels and additional image corrections are required for broader applicability.
How to cite: Nzuza, P., Schröder, M., Heim, R., Daniels, L., Slippers, B., Hurley, B., Germishuizen, I., Sivparsad, B., Roux, J., and Maes, W.: Assessing Gonipterus sp. n. 2 defoliation levels using multispectral Unmanned Aerial Vehicle (UAV) data in Eucalyptus plantations, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6619, https://doi.org/10.5194/egusphere-egu25-6619, 2025.
Posters virtual: Wed, 30 Apr, 14:00–15:45 | vPoster spot A
EGU25-7236 | ECS | Posters virtual | VPS4
Assessment of In-situ Canopy Cover Measurement Techniques and GEDI Vertical Canopy Cover in the Indian Western Himalayan RegionWed, 30 Apr, 14:00–15:45 (CEST) | vPA.30
Global Ecosystem Dynamics Investigation (GEDI) mission, operating from International Space Station, is a full-waveform LiDAR measuring vertical 3-dimensional structure of terrestrial ecosystems. The vertical canopy cover (CC) available from the GEDI L2B product has applications in forest ecosystem, forest health and climate change studies, and management practices. Some studies have assessed the accuracy and uncertainty of the GEDI vertical canopy cover profile product using aerial LiDAR scans and in-situ measurements. However, in-situ measurements taken using angle-of-view effectively produces canopy closure whereas GEDI measures vertical CC. Cajanus tube, regarded as ideal canopy cover measurement technique, is time consuming and impractical for larger areas. In this study, suitable in-situ canopy cover measurement methodologies were assessed alongside GEDI vertical CC. Canopy cover measurements were taken under GEDI footprints in the Indian Western Himalayan region using spherical densiometer, hemispherical photographs and digital canopy photographs with narrow angle-of-view. The plot dimensions were adjusted to accommodate horizontal geolocation uncertainty of GEDI version 2 data products. Following data collection, measurement techniques were assessed based on R-squared, RMSE and MAE.
How to cite: Paygude, A., Pande, H., and Tiwari, P. S.: Assessment of In-situ Canopy Cover Measurement Techniques and GEDI Vertical Canopy Cover in the Indian Western Himalayan Region, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-7236, https://doi.org/10.5194/egusphere-egu25-7236, 2025.
EGU25-208 | ECS | Posters virtual | VPS4
Mapping Alpine Treeline Ecotones in the Tungnath Himalaya Using Terrestrial Laser Scanning and GEDI LiDARWed, 30 Apr, 14:00–15:45 (CEST) | vPA.32
Alpine treeline ecotones are extremely vulnerable to climate change, making them important early warning systems in climate research. Advanced remote sensing tools, such as Light Detection and Ranging (LiDAR), enable detailed mapping and monitoring of these high-altitude zones, offering critical baseline data for future change detection. This study combines ground-based Terrestrial Laser Scanning (TLS) and space borne Global Ecosystem Dynamics Investigation (GEDI)- LiDAR data to analyze the structural attributes and delineate the position of alpine treelines in the Tungnath Himalaya, India, located at elevations between 3252 and 3,590 meters above mean sea level (a.m.s.l). TLS provided high-resolution three-dimensional data on alpine vegetation, including tree height, diameter at breast height (DBH), and canopy structure. Using an automated algorithm, 84.84% of individual trees were segmented from TLS data. TLS-derived tree height and DBH estimates achieved root mean square errors of 44.74 cm and 78.45 cm, respectively, compared to field-measured values. A semi-automated method using GEDI LiDAR identified trees taller than 3 meters to delineate the treeline, achieving a positional accuracy of ~ ±40 m a.m.s.l when validated against TLS-derived data. The results show that combining TLS with GEDI provides a non-destructive and effective method for assessing treeline structure and location in the Indian Himalaya. Future research might use multi-temporal LiDAR datasets to track treeline movements and obtain a better understanding of the long-term effects of climate change on alpine ecosystems.
How to cite: Mathew, J., Singh, C. P., Solanki, H., and Sedha, D.: Mapping Alpine Treeline Ecotones in the Tungnath Himalaya Using Terrestrial Laser Scanning and GEDI LiDAR, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-208, https://doi.org/10.5194/egusphere-egu25-208, 2025.
