3D4est

Research Project

Forest ecosystems play a critical role in multiple domains. Acting as carbon sinks, they provide an essential service to mitigate global warming. Forests also host a significant portion of the world’s biodiversity. However, their current state and dynamics are shaped by a complex interplay of short-term events, human activities, and long-term climate change. To support effective forest management and policymaking, it is crucial to understand and monitor these dynamics accurately. Achieving this requires precise stand-scale monitoring of forest ecosystems. This research aims to enhance the estimation of forest tree characteristics using Unmanned Aerial Vehicle (UAV) data.

UAVs enable the rapid collection of diverse data types over large areas. In this study, particular emphasis is placed on LiDAR technology, which provides detailed insights into the vertical structure of forests. UAV-mounted LiDAR sensors can help detect understory vegetation and assess forest regeneration. To process this data, deep learning models will be explored for tree detection and the extraction of relevant attributes from 3D point clouds.

Tree extraction from aerial point clouds is the first step of this research. Aerial LiDAR data will be used to train deep learning object detection models capable of identifying individual trees. Once extracted, various geometric parameters for each tree (e.g., diameter at breast height (DBH), stem height, tree height, crown size, and biomass) can be computed. While some of these parameters can be derived using rule-based algorithms, others, like DBH, remain challenging to estimate from aerial LiDAR data. To address these challenges, artificial intelligence (AI), including machine learning and deep learning techniques, will be investigated.

Tree species classification is another key objective of this research. To achieve this, the merging of 3D LiDAR data with multispectral and hyperspectral data will be explored. While spectral information is the primary source for species identification, the structural and geometric details provided by LiDAR data can enhance classification accuracy.

In conclusion, this research will contribute to a better understanding of forest ecosystems by advancing forest monitoring techniques. Ultimately, it will support the transition to more sustainable forest management practices.

The Team

Cadwal Borremans (PhD Student)

Dr. Romain Neuville (Co-Supervisor)

Prof. François Jonard (Supervisor)

Publications