13/11/2026 –, Multifuncional (Remoto)
Abstract
A tree-stage methodological approach is described for addressing the problem of identifying patters in 3D LiDAR point clouds and distinguished vegetation diversity captured in a Digital Elevation Model (DEM), considering the spatial distribution of the 3D point clouds, tree morphology, and canopy size. Strengthening the processes of segmenting, characterizing, and classifying 3D point clouds using MatLab and Python tools will allow for their implementation and application in larger forest areas, improve the efficiency of inventory estimation, and increase their usefulness in forest management.
Keywords: LiDAR Remote Sensing, Machine Learning, Point Cloud Segmentation.
1 Introduction
Concerning to the United Nations Sustainable Development Goals (SDGs), the scientific community is joining forces to address environmental challenges that belong the good health and well-being (SDG 3), climate action (SDG 13), and life on land (SDG 15) [4, 5]. This includes a focus on global concerns such as uncontrolled deforestation and the protection of natural areas within communities growing due to urban expansion. Furthermore, Mexico’s National Strategic Programs (PRONACES) are aligned with the SDGs through their application and availability to tackle on national strategic problems in current initiatives that encompass concerns ranging from local to global [2]. Given that remote sensing and artificial intelligence technologies have been used for a variety of applications in the context of social and environmental management, and that related scientific work has demonstrated the utility and capabilities of tree canopy classification, enabling biomass estimation and monitoring of natural
areas within developing urban areas [1, 3, 6]. Therefore, the next section describes a procedure that highlights the segmentation and characterization of tree canopies as hard preprocessing steps, previously to machine learning algorithms validation using MatLab, as implemented in reference [1]. However, for this approach, classification is performed by validating and testing different machine learning algorithms and identifying seven structural forms of tree canopies within the study region using the Python scikit-learn library [7]. The scope and development of this part of the project focuses on strengthening the classification process through the strategic selection of feature extraction and the expansion of the study region, as well as the tree canopy sample bank provided by 3D LiDAR point clouds.
2. Methodology
The proposed workflow for our approach is based on the methodology of reference [1], beginning with LiDAR data acquisition, which enables the generation of Digital Elevation Models (DEMs) of the study region. Next, the preprocessing stages are performed, first applying a strategic segmentation algorithm based on the Euclidean distance relationship between 3D points, primarily to delineate point clouds related to the morphological trend of tree cover. This is followed by the characterization stage, which involves extracting features related to the geometry, spatial relationships of the solids, and statistical data of each segmented individual tree. This extracted data is then stored in feature vectors, which aim to formulate the training/test database that must be supplied to the cover classification stage in order to prove and validate its performance.
3 Results
The robustness of the classification relies on selecting the highest-performing machine learning methods included in the Python 3.10 sklearn toolkit for training and validation data, using the five mentioned classifiers. The validation accuracies obtained, such as the area under the curve (AUC) for each training model, remain at 0.98 for the Multilayer Perceptron using the neural network approach, 0.97 for the Weighted Support Vector Machine, and 0.95 for the Random Forest method. The lower classification performance results for the remaining methods regards with an AUC of 0.91 for Naive-Bayes and an AUC of 0.80 for the Gradient Boosting classifier.
4 Conclusions
This manuscript briefly details the proposed methodology for addressing the problem of tree cover classification from the LiDAR point cloud capture. The data undergoes preprocessing steps, such as segmentation and characterization, to clean and organize the LiDAR point clouds, mainly highlighting five tree cover morphologies present in the study region. The classification process is performed using five machine learning methods available in the Python sklearn toolkit, comparing them to select and optimize their performance. Therefore, the validation and testing results of the chosen classifiers, implemented in Python, support the proposed methodology for application in complex study regions with large dimensions and diverse tree cover, with the aim of estimating biomass and monitoring vegetation areas for environmental management.
References
Monroy-Anieva, J. A., Téllez-Quiñones, A., Lopez-Farías, R., & Aguilar-Sierra, H. (2026).
Tree Canopy Segmentation and Characterization Using LiDAR for Machine Learning Models.
IEEE/SICE International Symposium on System Integration (SII).Pavón, N. P., Larios Lozano, O., & Bravo Cadena, J. (2025). Los sistemas socio-ecológicos: un
enfoque transdisciplinario para la atención de los problemas socioambientales. Herreriana, 7(1),
37-42. doi:10.29057/h.v7i1.13469Laino, D. et al. (2024). 3DFin: a software for automated 3D forest inventories from terrestrial point clouds. Forestry: An International Journal of Forest Research, 97(4), 479-496. doi:10.1093/forestry/cpae020
Bennich, T., Persson, A ., Beaussart, R., Allen, C., & Malekpour, S. (2023). Recurring patterns of SDG interlinkages and how they can advance the 2030 Agenda. One Earth, 6(11), 1465-1476. doi:10.1016/j.oneear.2023.10.008
Scown, M.W., Craig, R. K., Allen, C. R., Gunderson, L., Angeler, D. G., Garcia, J. H., & Garmestani, A. (2023). Towards a global sustainable development agenda built on social–ecological resilience. Global Sustainability, 6, e8. doi:10.1017/sus.2023.8
Ma, Q., Su, Y., & Guo, Q. (2017). Comparison of Canopy Cover Estimations From Airborne LiDAR, Aerial Imagery, and Satellite Imagery. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 10(9), 4225-4236. doi:10.1109/JSTARS.2017.2711482
Pedregosa, F. et al. (2011). Scikit-Iearn: Machine learning in python. Journal of Machine Learning Research, 12.