摘要
本研究以贵州省国有扎佐林场98块样地内的马尾松数据为研究对象,从7种常见的树高-胸径模型中挑选出最佳基础模型,然后引入对模型预测精度有显著影响且效果最大的林分因子,构建了广义模型。最后基于基础和广义模型构建不同分位点的分位数回归模型,选取决定系数(R^(2))、平均绝对误差(MAE)和均方根误差(RMSE)对不同模型进行检验。在7个基础模型中以Richards模型预测效果最佳,确定为最优基础模型;引入优势木平均高构建广义模型比基础模型的预测效果明显提高;在9个分位数中,当τ=0.4时基础和广义模型回归拟合能力最好。基于分位数回归构建的马尾松树高-胸径模型有很好的预测效果,但本研究数据量单一以及考虑影响模型因子较少,所以在兼顾数据样本和模型影响因子方面有待进一步研究。
Taking the data of Pinus Massoniana from 98 plots in Zhazuo State-owned Forest Farm in Guizhou Province as research object,the best base model from 7 common tree height-diameter at breast height(DBH)models were selected,and the stand factor,which significantly affected the model prediction accuracy,was introduced to construct a generalized model.The quantile regression models at different quantiles built based on base and generalized models with coefficient of determination(R^(2)),mean absolute error(MAE),and root mean square error(RMSE)were evaluated.Among the 7 base models,the Richards model had the best prediction effect and was determined to be the optimal base model.The generalized model incorporating the mean height of dominant trees significantly improved the prediction accuracy over the base model.Among the 9 quartiles,the regression fitting capacity of the base and generalized models performed best whenτ=0.4.The Height-DBH model of Pinus Massoniana based on quantile regression showed good prediction performance.Due to the limitations in terms of data volume and few factors considered in the model,it is suggested to take more data samples and influencing factors in further research.
作者
余昆隆
曹霸
王利治
张雷
何成正
YU Kun-long;CAO Ba;WANG Li-zhi;ZHANG Lei;HE Cheng-zheng(Zhazuo State-owned Forest Farm of Guizhou Province,Guiyang,Guizhou 550299;Guizhou Institute of Forestry Inventory and Planning,Guiyang,Guizhou 550081)
出处
《陕西林业科技》
2025年第3期45-50,共6页
Shaanxi Forest Science and Technology
基金
基于激光雷达技术的马尾松森林结构参数估测研究(黔林科合[2022][37]号)。
关键词
马尾松
树高-胸径模型
分位数回归
Pinus massoniana
height-diameter model
quantile regression
作者简介
余昆隆,硕士研究生,主要从事森林可持续经营工作,E-mail:971547515@qq.com;通信作者:曹霸。
