落叶松人工林树皮厚度预测模型

Bark thickness prediction models for larch plantation

【目的】建立落叶松人工林树皮因子及任意高度处树皮厚度的预测模型,以期更加准确地对树皮厚度进行预测,为实际木材生产和森林经营提供更加准确的指导。【方法】基于2015年黑龙江省佳木斯市孟家岗林场49株人工落叶松的1 186个圆盘数据,利用SAS 9.4软件中的MIXED模块构建落叶松人工林树皮厚度(树皮因子、任意高度处树皮厚度)的线性混合效应预测模型。模型评价指标选用赤池信息准则(AIC)、贝叶斯信息准则(BIC)、-2倍的对数似然值(-2LL)及似然比检验(LRT)。【结果】对于树皮因子模型,基于树木效应时含b1、b2、b4随机参数组合的树皮因子模型为最优混合模型;基于样地效应时含b1、b2随机参数组合的模型是最优混合效应模型。对于任意高度处树皮厚度模型,基于树木效应时含b1、b2的组合为最优混合模型;基于样地效应时含b0、b2、b3组合的为最优模型。所有最优模型在具有无结构(UN)方差-协方差矩阵时拟合效果最好。【结论】不论是树皮因子还是树皮厚度模型,树木效应对模型的影响最大;混合效应模型的预估精度与传统回归模型相比有明显提高。

【Objective】The prediction model of bark factor and bark thickness at any height of larch plantation was established in order to predict bark thickness more accurately and provide more accurate prediction model and guidance for actual wood production and forest management.【Method】Based on 1 186 disk data of 49 artificial larch trees in Mengjiagang Forest Farm of Jiamusi City, Heilongjiang Province in 2015, the bark thickness(bark factor, bark thickness at any height) of larch plantation linear mixed effects prediction model was constructed using the MIXED module in SAS 9.4 software. The model evaluation indicators were Akaike Information Criterion(AIC), Bayesian Information Criterion(BIC),-2 log likelihood(-2 LL) and likelihood ratio test(LRT). 【Result】For the bark factor model, the bark factor model with b1,b2, b4 and random parameter combination is the optimal mixed model based on the tree effect, and the model with b1, b2 random parameter combination is the optimal model based on the plot effect. For the bark thickness model at any height, the combination of b1, b2 is the optimal mixed model based on the tree effect, and the combination of b0,b2,b3 is the optimal model based on the plot effect. All the optimal models have the best fitting effect when they have unstructured(UN) variance-covariance matrix. 【Conclusion】The tree effect has the greatest influence on the model whether it is the bark factor or the bark thickness model. The prediction accuracy of the mixed-effect model is significantly improved compared with the traditional regression model.