Individual tree above-ground biomass estimation by integrating LiDAR and machine learning

Abstract

Global warming represents a critical challenge globally, while tree carbon sequestration is essential for achieving carbon neutrality. The existing global allometric models face challenges in accurately modelling local trees’ biomass. To develop a localized allometric model using a small dataset, this study proposes an innovative framework for estimating tree above-ground biomass (AGB) that involves local tree felling data collection, Light Detection and Ranging (LiDAR) implementation, and the development of a machine learning-based allometric model. During the data collection period, 100 trees were felled in Hong Kong from March 2023 to April 2024, encompassing 31 tree species and 17 tree families. Point-cloud models of the felled trees were collected using a LiDAR backpack. Each felled tree’s AGB was measured by integrating point-cloud technology and oven drying of samples. A data augmentation method was developed with a proposed tree point-cloud ‘degrowth’ algorithm to address the challenge of data limitation in allometric model development. The allometric models in this study were trained using advanced tree parameters measured by TreeQSM and tree family parameters. The best-performing allometric model developed by XGBoost, scored an accuracy of R² = 0.82, mean absolute percentage error (MAPE) = 40.70 %, and mean absolute error (MAE) = 214.37 kg. To summarize, this study enhanced AGB estimation in the local region by incorporating LiDAR, tree data augmentation, and machine learning for allometric model development.