Modeling diameter using UAV LiDAR data for Chinese fir (Cunninghamia lanceolata) in Southern China

Ziyang Liu¹Dongbo Xie¹Zheyuan Wu¹Linyan Feng¹Xingyong Liao²Yongjun Wang²Wendong Zhu²Ram P. Sharma³Liyong Fu^1*

• ¹Institute of Forest Resource Information Techniques, Chinese Academy of Forestry, Haidian, China
• ²Chengdu Agriculture And Forestry Academy Of Sciences, Chengdu, Sichuan Province, China
• ³Institute of Forestry, Tribhuvan University, Kathmandu, Nepal

Current research on predicting individual tree diameter at breast height (DBH) using airborne LiDAR data at large scales remains limited by the lack of comprehensive models that effectively account for variations in tree growth stages and regional differences in tree growth responses. Existing models often overlook the non-linear effects of growth stages and fail to integrate stand-level and regional-level variations, leading to suboptimal predictive accuracy. To address these gaps, we developed a DBH model for Cunninghamia lanceolata forests suitable for large-scale applications, utilizing ground-based measurements and corresponding airborne LiDAR data from 26,768 trees across 130 forest sites in Guangdong Province, China. Due to differences in tree growth rates and responsiveness to environmental factors at different growth stages, we introduced a dummy variable model to group the full range of growth stages, thereby reducing correlations between stage-specific observations. We incorporated two-level random effects to capture the influence of regional and intra-regional differences on DBH predictions at different growth stages. The contribution of stand density and tree characteristics to improving the DBH model was also evaluated. The best combination of dummy variables and random effects was selected using the Akaike information criterion (AIC) and error metrics. Model performance was evaluated using an independent dataset, showing that incorporating stand competition and crown width indicators improved model fit statistics. Moreover, the model that accounted for growth stage variations and introduced random effects to capture regional and sample plotlevel differences outperformed the basic model, enhancing predictive accuracy. This study addresses the limitations of previous models by integrating growth stage-specific effects and multi-level variability, providing a robust modeling approach that can be adapted for large-scale applications of airborne LiDAR data in predicting individual tree DBH of Cunninghamia lanceolata across different regions.