关键词: 硬阔叶林, 地基激光雷达, 树高, 枝下高, 混合效应模型

Abstract: Tree height and height to crown base are key parameters for the monitoring of forest growth and sustainable forest management. Accurate prediction of these variables is essential for the conservation and efficient utilization of valuable hardwood broad-leaved forests. In this study, we developed prediction models for tree height and height to crown base of Fraxinus mandshurica, Juglans mandshurica, and Phellodendron amurense in the Zhuan-shan Experimental Forest Farm of Heilongjiang Province, to by combining terrestrial laser scanning (TLS) point cloud data with field-measured data. First, optimal basic models were screened, and generalized models were then constructed by introducing individual-tree factors (height to crown base), competition factors (relative diameter and basal area of trees larger than the subject tree), stand structure variables (dominant diameter and Gini coefficient of diameter), and species diversity indices (mingling index and Pielou evenness index). Finally, mixed-effects models were established by incorporating plot-level random effects. The results showed that the overall fitness for individual trees reached 96%, and the coefficients of determination (R²) between TLS-derived tree height, height to crown base, and diameter at breast height and field measurements were 0.80, 0.75, and 0.98, respectively, with corresponding root mean square error (RMSE) values of 2.00 m, 1.96 m and 0.93 cm. The optimal basic models for tree height and height to crown base were the exponential reciprocal model (R²=0.637, RMSE=2.739 m) and the bivariate interaction model (R²=0.373, RMSE=2.981 m), respectively. The introduction of multiple variables in the generalized models significantly improved prediction accuracy, with R² increasing by 24.2% and 20.9% and RMSE decreasing by 39.4% and 6.3% for tree height and height to crown base models, respectively. The mixed-effects models considering plot-level random effects performed the best, with R² further increasing by 11.8% and 7.1% and RMSE decreasing by 29.1% and 3.8% for tree height and height to crown base models, respectively, compared with the corresponding generalized models. Overall, mixed-effects modeling combining TLS data with multi-level ecological factors effectively improved the prediction accuracy of vertical structural parameters in valuable hardwood broad-leaved forests, which could provide reliable technical support for precision forest management and dynamic monitoring in Northeast China.

Key words: hardwood broad-leaved forest, terrestrial laser scanning (TLS), tree height, height to crown base, mixed-effects model