Terrestrial LiDAR scans (TLS) offer a rich data source for high-fidelity vegetation characterization, addressing the limitations of traditional fuel sampling methods by capturing spatially explicit distributions that have a significant impact on fire behavior. However, large volumes of complex, high resolution data are difficult to use directly in wildland fire models. In this study, we introduce a novel method that employs a voxelization technique to convert high-resolution TLS data into fine-grained reference voxels, which are subsequently aggregated into lower-fidelity fuel cells for integration into physics-based fire models. This methodology aims to transform the complexity of TLS data into a format amenable for integration into wildland fire models, while retaining essential information about the spatial distribution of vegetation. We evaluate our approach by comparing a range of aggregate geometries in simulated burns to laboratory measurements. The results show insensitivity to fuel cell geometry at fine resolutions (2–8 cm), but we observe deviations in model behavior at the coarsest resolutions considered (16 cm). Our findings highlight the importance of capturing the fine scale spatial continuity present in heterogeneous tree canopies in order to accurately simulate fire behavior in coupled fire-atmosphere models. To the best of our knowledge, this is the first study to examine the use of TLS data to inform fuel inputs to a physics based model at a laboratory scale.
Above-ground vertical structure is a critical variable for ecosystem monitoring, carbon accounting, and land management. However, the high cost and limited coverage of airborne lidar hinder its widespread application. To address this, we developed NAIP-CHM, a 0.6-meter resolution canopy height and structure model (CHM) covering the contiguous United States, derived from National Agriculture Imagery Program (NAIP) aerial imagery. Unlike forestry-specific models that exclude human-made features, NAIP-CHM characterizes the full vertical structure of the landscape including vegetation, buildings, and infrastructure. We utilized a U-Net convolutional neural network with attention mechanisms and environmental conditioning, training and validating the model with a peer-reviewed, publicly available dataset of 22.8 million co-registered NAIP imagery and lidar-derived CHM pairs, with stratified sampling to ensure robustness in open-canopy ecosystems. The model achieved a pixel-wise root mean square error (RMSE) of 2.28 meters and an r 2 of 0.87. Forested sites alone produced an r 2 of 0.82 and RMSE of 3.82 meters. We provide the dataset, source code, and cloud-based tools to enable broad application without requiring specialized computational resources.
Accurate three-dimensional representations of tree structure are essential for fire modeling, radiative transfer simulation, synthetic data generation, and digital twins of forests, yet detailed 3D structure is rarely available at required scales. Current approaches approximate crowns with smooth geometric primitives, discarding the clumping, gaps, and irregular branching present in real trees. We present TreeFlow, a conditional flow matching model that generates realistic 3D tree point clouds from species, acquisition platform, and height. The model uses a transformer trained on real laser scanning data from the FOR-species20K benchmark to learn a velocity field transporting samples from a Gaussian distribution to the source data distribution. We evaluate generation quality by comparing conditioning and distributional fidelity metrics to scans of real trees. Generated trees match or approach the intra-class baseline on five of six metrics, with a Chamfer distance of 0.581 m versus 0.559 m for real trees of the same genus and height class. Performance is strongest below 25 m and degrades with increasing height. TreeFlow is the first flow matching model to produce 3D tree point clouds from scalar inventory attributes using real laser scanning data.
Accurate three-dimensional representations of tree structure are essential for fire modeling, radiative transfer simulation, synthetic data generation, and digital twins of forests, yet detailed 3D structure is rarely available at required scales. Current approaches approximate crowns with smooth geometric primitives, discarding the clumping, gaps, and irregular branching present in real trees. We present TreeFlow, a conditional flow matching model that generates realistic 3D tree point clouds from species, acquisition platform, and height. The model uses a transformer trained on real laser scanning data from the FOR-species20K benchmark to learn a velocity field transporting samples from a Gaussian distribution to the source data distribution. We evaluate generation quality by comparing conditioning and distributional fidelity metrics to scans of real trees. Generated trees match or approach the intra-class baseline on five of six metrics, with a Chamfer distance of 0.581 m versus 0.559 m for real trees of the same genus and height class. Performance is strongest below 25 m and degrades with increasing height. TreeFlow generates individual-tree point clouds conditioned on scalar inventory attributes using a model trained entirely on real laser scanning data.