中亚热带米槠阔叶林单木质量分类评价

Classification and Quality Assessment of Individual Tree in Castanopsis carlesii Broad-leaved Forest in Mid-subtropical Zone

于2022年在福建省中部尤溪县国有林场,以中亚热带米槠(Castanopsis carlesii)群落作为研究对象,通过构建胸径、树高、冠幅、活冠层比重、树冠枯梢比重、树冠密度等11项指标体系,运用模糊综合评价进行单木质量分类,分类结果选择线性判别分析和K近邻算法建立预测模型,10折交叉验证并做精度评估,建立一种快速评价单木质量的研究方法。结果表明:单木质量分类结果中,4种评价等级差、中、良、优的林木占比分别为22.20%、30.19%、30.45%、17.16%,总体质量等级以良和中为主。交叉验证分析线性判别预测模型误判率、准确率分别为14.18%、85.82%,K近邻算法误判率、准确率分别为11.84%、88.16%。混淆矩阵分析线性判别误差主要出现在中和优之间,K近邻算法误差主要出现在中和良之间。综合结果表明,K近邻算法的准确率、平均精确率和召回率均优于线性判别分析,所建立的单木质量预测模型,可以在保证一定预测精度的前提下,快速评判研究区内米槠阔叶林单木质量。

In 2022,the Castanopsis carlesii community in mid-subtropical zone was selected as the research object at the state-owned forest farm in Youxi County,central Fujian Province.By constructing a system of 11 indicators including DBH,tree height,crown width,live canopy ratio,crown dieback and crown density,fuzzy comprehensive evaluation was used to classify the quality of individual trees.Linear discriminant analysis and K-nearest neighbor algorithm were used to establish a prediction model for the classification results,and 10 fold cross validation was conducted to evaluate the accuracy.The results showed that in the single tree quality classification results,the proportion of trees with four evaluation levels of poor,medium,good,and excellent was 22.20%,30.19%,30.45%and 17.16%,respectively.The overall quality level was mainly good and medium.The cross validation analysis showed that the error rate and accuracy of the linear discriminant prediction model were 14.18%and 85.82%,respectively.The error rate and accuracy of the K-nearest neighbor algorithm were 11.84%and 88.16%,respectively.The confusion matrix analysis linear discrimination error mainly occurs between medium and excellent,while the K-nearest neighbor algorithm mainly occurs between medium and good.The K-nearest neighbor algorithm outperformed linear discriminant analysis in terms of accuracy,average precision,and recall.Therefore,the individual tree quality prediction model established by the K-nearest neighbor algorithm can quickly evaluate the single tree quality of the Castanopsis carlesii broad-leaved forest in the study area while ensuring a certain prediction accuracy.