Growth model development and lidar-driven ai prediction of single tree-level attributes for çamkoru scots pine stands

Sustainable forest management requires species- and site-specific growth models, whose key components include stem taper, height–diameter (H–D) relationships, diameter distributions, and biomass functions. In this study, taper, H–D, aboveground fresh biomass, and diameter-distribution models were developed for natural Scots pine (Pinus sylvestris L.) stands in the Çamkoru Research Forest (Ankara, Türkiye). Among taper models, the formulation by Jiang et al. (2005) exhibited the highest goodness-of-fit (R² = 0.9708; RMSE = 1.6735 cm), whereas the Biging (1984) model yielded the weakest performance. For H–D modeling, several functional forms were tested, with the mixed-effects Gompertz model providing the best results, particularly when calibrated with four sample trees per plot. Stand structure was effectively characterized using the Weibull distribution, and among nine alternatives, the Seemingly Unrelated Regression (SUR) approach with percentile-based parameter estimation produced the most accurate results. To ensure additivity in biomass estimation for whole trees and components, nonlinear SUR (NSUR) and the Generalized Method of Moments (GMM) were jointly applied. In the remote-sensing integration, a knowledge-distillation–based artificial intelligence model predicted DBH from airborne LiDAR-derived crown diameter and height metrics with high accuracy (R² = 0.891; RMSE = 2.7456 cm). Species classification using UAV-borne multispectral imagery showed improved accuracy of the LightGBM algorithm from 85.5% to 93.5% following hyperparameter optimization. Furthermore, mobile LiDAR provided highly precise estimates of diameter, height, and volume. Overall, integrating field measurements with LiDAR and multispectral data offers a robust and transferable framework for modeling diameter distributions, growth dynamics, and biomass in Scots pine stands.