基于哑变量的木荷人工林树皮厚度模型研究  

Bark Thickness Model of Schima superba Plantation Based on Dummy Variable

作　　者：张伟志 吴宏炜 黄光灿 田意 庄崇洋[1] 江希钿[1] ZHANG Weizhi;WU Hongwei;HUANG Guangcan;TIAN Yi;ZHUANG Chongyang;JIANG Xidian(College of Forestry,Fujian Agriculture and Forestry University,Fuzhou,Fujian 350002,China)

机构地区：[1]福建农林大学林学院,福建福州350002

出　　处：《热带作物学报》2020年第12期2562-2570,共9页Chinese Journal of Tropical Crops

基　　金：福建省林业局森林生长动态预测模型及数表编制项目(No.KLB18H18A);森林资源资产精准计测及评估技术项目(No.KFA17283A)。

摘　　要：在木荷人工林中,基于哑变量模型法构建树皮厚度模型,以提高预测模型参数的稳定性,为木荷人工林出材率计算和经济价值评估方面提供参考依据。基于福建省南平市213块木荷人工林固定样地的调查数据,选取19个包含胸高处、任意高度处、相对树皮厚度和去皮直径的理论方程作为构建木荷人工林树皮厚度的基础模型,使用R软件进行模型拟合分析,运用决定系数(R2)、均方根误差(Erms)、和方差(Ess)、赤池信息准则(AIC)、贝叶斯准则(BIC)等模型评价指标,结合含熵值的TOPSIS法筛选出拟合度较高的4个基础模型(MI、M10、M12、M16),进一步构建含龄组和立地质量哑变量的木荷人工林树皮厚度模型。用于建模的胸高处样本数为130个,任意高度处、相对树皮厚度和去皮直径样本数为2386个,用于检验模型拟合效果相对应的样本数分别为55个和1013个。结果表明:含龄组哑变量胸高处、任意高度处、相对树皮厚度模型(M20、M22、M25)的R2分别为0.9769、0.9214、0.9111,比基础模型(M1、M10、M12)的R2 (分别为0.6981、0.5540、0.5056)提高了39.9%、66.3%、80.2%,有了明显提升;含龄组哑变量去皮直径模型M27的AIC为21.62,BIC为76.44,比其基础模型M16分别降低了95.8%、85.46%。通过含熵值的TOPSIS法对模型指标综合评价,并经模型配对t检验,这4个含龄组哑变量模型(M20、M22、M25、M27)拟合效果更佳,适合于福建木荷人工林树皮厚度预测。In the artificial forest of Schima superba,a bark thickness model was constructed based on the dummy variable model method,in order to improve the stability of the prediction model parameters,and provide reference for the calculation of the timber yield and the evaluation of the economic value of the artificial forest of S.superba.Based on the survey data of 213 sample plots of S.superba plantation in Nanping City,Fujian Province,19 theoretical equations including breast height,any height,relative bark thickness and peeling diameter were selected as the basic models for building the bark thickness of S.superba plantation.Based on R software model fitting analysis,determination coefficient(R2),root mean square error(Erms),the sum of squares due to error(Ess),akaike information criterion(AIC)and bayesian information criterion(BIC)models as the evaluation index,and the entropy TOPSIS method to screen high fitting degree of four basic models(M1,M10,M12,M16)model,an age group and the site quality of S.superba plantation bark thickness of dummy variable model was further built.The number of samples at the breast height used for modeling was 130,the number of samples at any height,the relative bark thickness,and the peeling diameter was 2386,and the corresponding numbers for testing the model fitting effect was 55 and 1013 respectively.The results showed that R2 of the model with age dummy variable at breast height, any height and relative bark thickness (M20, M22 and M25)was 0.9769, 0.9214 and 0.9111, respectively, which was 39.9%, 66.3% and 80.2% higher than that of the basic model(M1, M10 and M12), and AIC and BIC of M27 with age dummy variable was 21.62 and 76.44, respectively, it was95.8% and 85.46% lower than that of the basic model M16. Through the comprehensive evaluation of the model indexesby the TOPSIS method with entropy value and t-test of model pairing, the fitting effect of these four age group dummyvariable models (M20, M22, M25, M27) was better, which is suitable for the prediction of bark thickness of

关 键 词：树皮厚度 哑变量 龄组 立地质量 木荷 

分 类 号：S758[农业科学—森林经理学] S79[农业科学—林学]