应用机器学习算法预测樟子松树皮厚度  

Prediction of bark thickness of Pinus sylvestris var.mongolica by machine learning algorithm

作　　者：谌俊燃 张兹鹏 姜立春[1,2] CHEN Junran;ZHANG Zipeng;JIANG Lichun(College of Forestry,Northeast Forestry University,Harbin 150040,Heilongjiang,China;Key Laboratory of Sustainable Forest Ecosystem Management Ministry of Education,Northeast Forestry University,Harbin 150040,Heilongjiang,China)

机构地区：[1]东北林业大学林学院,黑龙江哈尔滨150040 [2]东北林业大学森林生态系统可持续经营教育部重点实验室,黑龙江哈尔滨150040

出　　处：《中南林业科技大学学报》2025年第2期82-90,共9页Journal of Central South University of Forestry & Technology

基　　金：国家自然科学基金项目(32271866)。

摘　　要：【目的】研究传统模型和机器学习算法预测樟子松Pinus sylvestris var.mongolica的树皮厚度,为树皮厚度的精准预测提供理论依据和实践指导。【方法】以大兴安岭图强林业局245株樟子松伐倒木数据为研究对象,构建6个传统模型(线性、非线性)和2种机器学习模型(人工神经网络ANN、支持向量回归SVR),同时比较不同自变量组合时机器学习模型的表现。【结果】(1)树皮厚度的拟合和检验结果均表明2种机器学习模型均优于传统模型,具体结果排序为SVR6>ANN6>M5;(2)SVR6的最优输入变量组合为胸径、树高、距离地面高度和相对高。与传统模型M5相比,SVR6的预测精度有明显提高,其中R2提高了12.66%,RMSE和MAE分别降低了17.71%和20.27%;(3)将数据划分为不同径阶组合分析各模型的预测精度时,发现2种机器学习模型(ANN6、SVR6)的预测效果均优于传统模型M5。其中,当树木为小径阶(5 cm≤DBH<15 cm)和大径阶(DBH≥25 cm)时,建议采用SVR6进行预测;当树木多为中等径阶(15 cm≤DBH<25 cm)时,建议采用ANN6进行预测;(4)比较各模型在树干不同高度的预测能力时,发现在相对高0~70%处,ANN6和SVR6的预测能力较优;在相对高70%~100%处,M5的预测能力较优。总体来看,ANN6和SVR6在大部分高度处的预测能力都优于M5。【结论】机器学习建模方法可以有效地提高树皮厚度的预测精度。相较传统模型,机器学习模型预测效果更优。其中SVR的拟合和检验效果最好,适合该区域树皮厚度的精准预测。【Objective】This paper studies the traditional model and machine learning algorithm to predict the bark thickness of Pinus sylvestris var.mongolica,and provides theoretical basis and practical guidance for the accurate prediction of bark thickness.【Method】Taking 245 Pinus sylvestris var.mongolica felled data from Tuqiang Forestry Bureau in Daxing’anling as the research object,six traditional models(linear,nonlinear)and two machine learning models(artificial neural networks,support vector regression)were constructed,and the performance of machine learning models with different combinations of independent variables was compared.【Result】(1)The fitting and validation results of bark thickness show that the two machine learning models are better than the traditional model,and the specific results are SVR6>ANN6>M5;(2)The optimal combination of input variables for SVR6 is 2 DBH,H,h and RH.Compared with the traditional model M5,the prediction accuracy of SVR6 is significantly improved,in which R is increased by 12.66%,RMSE and MAE are reduced by 17.71%and 20.27%,respectively;(3)When the data is divided into different path order combinations to analyze the prediction accuracy of each model,it is found that the prediction effect of the two machine learning models(ANN6,SVR6)is better than that of the traditional model M5.Among them,SVR6 is recommended to be used for prediction when the trees are small diameter(5 cm≤DBH<15 cm)and large diameter(DBH≥25 cm);When most trees are of medium diameter(15 cm≤DBH<25 cm),ANN6 is recommended for prediction.4)When comparing the prediction ability of each model at different trunk heights,it was found that the prediction ability of ANN6 and SVR6 was better at the relative height of 0-70%;At the relatively high 70%-100%,M5 has better prediction ability.Overall,the prediction ability of ANN6 and SVR6 at most heights is better than that of the corresponding M5.【Conclusion】The results show that the machine learning modeling method can effectively improve the prediction a

关 键 词：樟子松 树皮厚度 预测精度 人工神经网络 支持向量回归 

分 类 号：S757[农业科学—森林经理学]