Integration of Terrestrial Laser Scanning and field measurements data for tree stem volume estimation: Exploring parametric and non-parametric modeling approaches

• Published in: Annals of forest research Vol. 67; no. 1; pp. 77 - 94
• Main Authors: Capalb, Florin, Apostol, Bogdan, Lorent, Adrian, Petrila, Marius, Marcu, Cristiana, Badea, Nicolae Ovidiu
• Format: Journal Article
• Language: English
• Published: Voluntari Forest Research and Management Institute (ICAS) - editor of Annals of Forest Research, owner of Silvica Publishing House 01.01.2024
• Subjects: Accuracy; Data acquisition; Data collection; Effectiveness; Estimates; Fagus sylvatica; Forest ecosystems; Forest management; Height; Laser applications; Modelling; Parameter identification; Parameter robustness; Plant species; Regression analysis; Regression models; Stems; Terrestrial ecosystems; Variables
• ISSN: 1844-8135; 2065-2445
• DOI: 10.15287/afr.2024.3664

Terrestrial laser scanning (TLS) has emerged as a powerful tool for acquiring detailed three-dimensional information about tree species. This study focuses on the development of models for tree volume estimation using TLS data for even aged Fagus sylvatica L. stands located in the western part of the Southern Carpathians, Romania. Both parametric and non-parametric modeling approaches were explored, leveraging variables extracted from TLS point clouds such as diameter at breast height (DBH), height, crown radius, and other relevant crown and height parameters. Reference data were collected through high-precision field measurements across 76 circular Permanent Sample Areas (PSA) spanning 500 m2 each. A multi-scan approach was implemented for TLS data collection, involving four scanning stations within each PSA. Concurrently, parametric (regression equations) and non-parametric (Random Forest - RF) models were applied, leveraging all TLS-derived variables to explore potential enhancements in volume estimation accuracy. Among the parametric models, the most effective performer was the one featuring solely DBH as an input variable. The RF non-parametric model yielded more accurate stem volume estimates (RMSE = 1.52 m3*0.1ha-1; RRMSE = 3.62%; MAE = 1.22m3*0.1ha-1) compared to the best-performing regression model (RMSE = 5.24 m3*0.1ha-1; RRMSE = 12.48%; MAE = 4.28 m3*0.1ha-1). Both types of models identified DBH as the most important predictive variable, while the RF model also included height and crown related parameters among the variables of importance. Results demonstrate the effectiveness of the non-parametric RF model in providing accurate and robust estimates of tree stem volume within even aged European beech stands. This highlights the significance of TLS data, increasingly employed in diverse forest inventory and management applications. Nevertheless, additional research and refinement of the proposed models are needed. This includes thorough validation across various forest ecosystems and continued efforts to enhance the accuracy of tree height determination from point cloud data.