基于机器学习的油松人工混交林单木地上生物量混合效应模型构建

Construction of an Aboveground Biomass Mixed Effect Model for Individual Trees of Mixed Forests of Pine Based on Machine Learning

利用现地调查数据,构建油松人工混交林单木地上生物量估算模型。通过野外调查,共获取了1313株油松人工混交林单木生物量实测数据,基于最小二乘法、线性混合效应模型和随机森林混合效应模型,分别构建油松人工混交林单木地上生物量模型。结果表明:1)线性混合效应模型(LME)和随机森林混合效应模型(MXRF)结果与普通最小二乘模型(OLS)相比,立地因子作为随机效应能够提升模型的解释能力;2)随机森林混合效应模型(MXRF)的决定系数(0.9907)大于线性混合效应模型(0.9347)和普通最小二乘模型(0.9332),且随机森林混合效应模型的均方根误差(RMSE)和平均绝对误差绝对值(MAE)均小于线性混合效应模型和普通最小二乘模型,说明MXRF模型拟合效果优于OLS模型;3)MXRF模型拟合克服了2种线性模型拟合中存在的异方差问题。本研究构建的模型较好地反映了油松人工混交林地上生物量与生长指标间的关系,形式简单、使用方便,可以为林分生长预测和可持续经营提供依据。

The objective of this study was to establish model to estimate the individual aboveground biomass of mixed Pinus tabuliformis plantation by using the data of site investigation.Through field surveys,the measured data of the individual biomass of 1313 P.tabuliformis plants were obtained in mixed plantations.Based on the ordinary least square model(OLS),linear mixed effect model(LME),and mixed effect random forest model(MXRF),the biomass estimation models were constructed and compared.The results showed that 1)compared to the results of LME,MXRF and OLS,the application of site factors as random effects enhanced the explanation capability of the models.2)The coefficient of determination of MXRF(0.9907)was larger than those of LME(0.9347)and OLS(0.9332).The root mean square error(RMSE)and mean absolute error(MAE)of MXRF were less than those of LME and OLS,indicating that the fitting effect of MXRF is better than the OLS model.3)MXRF model solved the problem of heteroscedasticity other than LME and OLS models.It is concluded that MXRF well reflects in the relationship between the biomass and growth indicators of P.tabuliformis plantations.The form of MXRF model is simple and convenient to use,which can provide a basis for the growth forecast and sustainable management of the plantations.