基于非线性混合效应的华东地区主要优势树种林分水平树高-胸径模型

Tree height-diameter models of main tree species in East China based on nonlinear mixed-effects model

【目的】使用非线性混合效应(nonlinear mixed-effects,NLME)方法建立华东地区林分水平树高-胸径模型,提供一个较大地域尺度的林分树高预测工具,并与传统非线性回归方法比较,探索非线性混合效应方法在构建树高-胸径模型上的表现。【方法】基于华东地区7个省级单元国家森林资源连续清查数据,构建主要优势树种哑变量,采用非线性混合效应方法建立林分水平树高-胸径模型。【结果】非线性混合效应方法在训练集与测试集上均表现优秀,训练集R2、均方根误差(RMSE)和平均绝对偏差(MAE)分别为0.678 5、2.396 7 m和1.832 6 m,测试集R2、RMSE与MAE分别为0.673 8、2.352 0 m和1.798 5 m。与不加入随机效应的传统回归方法相比,混合效应模型在训练集中,R2提升了8.13%,MAE与RMSE分别下降了7.09%与6.87%;在测试集表现中,R2上升了7.75%,RMSE与MAE分别下降了6.61%和5.97%。【结论】使用非线性混合效应方法建立了华东地区主要树种的林分水平树高-胸径模型,最终模型的拟合精度和预测精度均优于传统回归模型,模型适用的覆盖区域较广,能够为森林经营实践提供依据。

Objectives The tree height-diameter model is a highly efficacious instrument in forest management.The purpose of this article was to establish the tree height-diameter model at the stand level in East China using the nonlinear mixed-effects method and provide an available prediction tool for forest stand tree height at a large regional scale.The final model was compared with the traditional nonlinear regression model to explore the performance of the nonlinear mixed-effects method in developing the tree height-diameter model.Methods The nonlinear mixed-effects method was employed to construct a stand-level tree height-diameter model for the Each China region,utilizing data from the national forest resources inventory of seven provinces in the region.This involved the creation of dummy variables for the predominant tree species.Results The nonlinear mixed-effects method performed well on both the training and testing sets.The performance of the training set of the final model in terms of R2,root mean square error(RMSE),and mean absolute error(MAE)is 0.6785,2.3967 m,and 1.8326 m,respectively.The performance of the validation set of the final model in terms of R2,RMSE and MAE is 0.6738,2.3520 m and 1.7985 m,respectively.Compared with traditional regression methods,the mixed-effects model demonstrated an 8.13%increase in R2 within the training set,accompanied by a 7.09%reduction in MAE and a 6.87%reduction in RMSE.In the validation set,R2 exhibited a 7.75%increase,while RMSE and MAE exhibited a 6.61%and 5.97%reduction,respectively.Conclusions The stand-level tree height-diameter model for the main tree species in East China was established using the nonlinear mixed effects method.The final model demonstrated superior fitting and prediction accuracy compared to the traditional regression model.The model has a relatively wide applicable coverage area,which can provide a basis for forest management practices.