基于哑变量的湿地松林分断面积生长模型  

Basal area growth model for Pinus elliottii forest based on dummy variables

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

机构地区：[1]福建农林大学林学院,福建福州350002

出　　处：《中南林业科技大学学报》2021年第1期117-123,150,共8页Journal of Central South University of Forestry & Technology

基　　金：国家林业和草原局项目(KFA17283A,KLB18H18A)。

摘　　要：【目的】为更好地估测福州市湿地松人工林林分断面积生长情况,同时为湿地松人工林的经营提供参考依据。【方法】基于样地调查数据选用理查德方程、逻辑斯蒂、Mitscherlich和Schumacher等基础模型,引入优势木平均高和年龄因子为自变量并将林分密度指数作为密度指标加入到基础断面积模型中。在最优基础模型中引入哑变量,建立可兼用于采脂与未采脂林分的湿地松人工林断面积模型。利用R软件的遗传算法求解模型参数,采用均方差(RMSE)、残差平方和(SSE)、决定系数(R2)、模型精度(v)和模型准确度(P)等模型评价指标。并运用含熵权值的TOPSIS法对哑变量模型选优,选用的各项评价指标熵权值由其本身数值决定,该方法具有较好的客观性,避免主观因素影响综合评价结果。【结果】4个基础模型的拟合效果均较好,模型决定系数均超过了0.9,其中Schumacher模型为最优基础模型,决定系数达0.983 31,模型精度达98.20%。在该模型中b3参数中引入哑变量所得模型拟合效果最优,较最优基础模型决定系数提升到0.998 96,模型精度提升到99.37%,距离最优解距离总和最小为0.000 016 7。并对最优哑变量模型做适用性检验,分别检验模型对采脂林分、未采脂林分和总体林分的拟合效果,发现其预估精度均超过99%。根据检验数据断面积实测值和最优哑变量模型预测值建立的线性回归方程,其R2值达到0.999 2,常数项接近0且残差基本均匀分布于横轴两侧,表明模型预估精度较高。【结论】引入哑变量后模型拟合效果得到了提升,可用于估测采脂与未采脂不同经营措施湿地松林分断面积,为测算湿地松林分材积与规划林分经营模式提供参考。【Objective】Estimation of the basal area and the basis for the management of fat-harvesting and non-harvested forests in the Pinus elliottii plantation in Fuzhou, China.【Method】Based on the sample survey data, selected the basic models such as Richard, Logistic, Mitscherlich and Schumacher, Introduce the average height of dominant trees and the age factor as independent variables and add the stand density index as the density index to the basic cross-sectional area model,the dummy variables were introduced into the optimal foundation model to establish a model of the basal area growthing of Pinus elliottii plantation suitable for both fat-harvesting and non-harvested forest stands.The genetic algorithm of the R software was used to solve the model parameters, and model evaluation indicators such as mean square error(RMSE), residual square sum(SSE), determination coefficient(R2), model accuracy(v), and model accuracy(P) were used. The TOPSIS method with entropy weight is used to select the dummy variable model. The entropy weight of each evaluation index selected is determined by its own value. This method has good objectivity and avoids subjective factors affecting the comprehensive evaluation results.【Result】The fitting effects of the four basic models were better, models correlation coefficients were all over 0.9 and the Schumacher model was the optimal foundation model with a correlation coefficient of 0.983 31 and a model accuracy of 98.20%. The model with the dummy variable introduced into the b3 parameter in this model had the best fitting effect. The correlation coefficient of the optimal basic model was raised to 0.998 96, the model accuracy was improved to 99.37% and the minimum sum of the optimal solution distances is 0.000 016 7. Applicability test for the dummy variable model, the prediction accuracy of the fatharvesting forest and the non-harvested forest stand and the overall stand are more than 99%. The linear regression equation established based on the measured area of the test data an

关 键 词：湿地松 林分断面积 哑变量 含熵权值的TOPSIS 相容性 

分 类 号：S791.246[农业科学—林木遗传育种]