Abstract
Information on the distribution of trees with different diameters at breast height (DBH) is needed to inform management programs aimed at achieving conservation objectives in high stem density forest stands. This article explored the feasibility of mapping the density of trees with different DBHs using airborne LiDAR data. Experiments were conducted in the largest river red gum forest in the world, located in the southeast of Australia. Field measured data on trees with different DBHs were used for the supervised learning of airborne LiDAR scans with a pulse density of 5.92 pulses/m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> . Specifically, the hyperparameters of gradient boosting and random forest regressors were tuned to produce a viable solution for mapping the density of different sized trees at the plot level. Our results indicate that the total tree density (DBH > 0 cm; height > 1.37 m) can be mapped using airborne LiDAR data with the coefficient of determination R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of up to 0.67, with gradient boosting outperforming random forest. However, the accuracy of mapping the density of saplings (DBH ≤ 10 cm), small trees (10 cm <; DBH ≤ 50 cm), and large trees (DBH > 50 cm) differed with R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.65, 0.60, and 0.42, respectively. These results show that the airborne LiDAR data can provide a viable solution for mapping the density of small trees (DBH ≤ 50 cm) over large areas and has the potential for mapping the density of large trees (DBH > 50 cm).
Highlights
F OREST thickening caused by land management, altered disturbance regimes, and climatic factors is increasingly common globally [1]
Our results indicate that the total tree density (DBH > 0 cm; height > 1.37 m) can be mapped using airborne light detection and ranging (LiDAR) data with the coefficient of determination R2 of up to 0.67, with gradient boosting outperforming random forest
Gradient boosting outperformed random forest in predicting the densities of different sized trees when using predictor variables averaged across five plot areas
Summary
F OREST thickening caused by land management, altered disturbance regimes, and climatic factors is increasingly common globally [1]. After gazettal as Manuscript received July 27, 2020; revised October 8, 2020; accepted December 4, 2020. Date of publication December 21, 2020; date of current version January 8, 2021. To inform management programs that aim to achieve conservation objectives, a spatial representation of the density of different sized trees overtime is required for BMF. In this respect, the remote sensing technology may provide essential dynamic information on a scale that field-based studies cannot match
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
