基于随机森林的长白落叶松树干液流影响因素研究

Recognition of Dominant Factors behind Sap Flow of Larix olgensis Based on Random Forest Method

【目的】研究植物树干液流密度的变化及其对环境因子的响应对理解植物蒸腾作用、植物生理生化特性和植物耗水规律具有重要意义。目前基于皮尔逊相关系数的树干液流主导因素分析方法无法准确描述液流密度与环境因子间非线性关系,本研究为树干液流密度主导因子分析以及树干液流密度预测提供新的技术手段。【方法】本文提出基于随机森林的树干液流主导因子分析方法,以黑龙江省佳木斯市孟家岗林场长白落叶松为实验对象,利用Grainer探针和气象因子观测站监测长白落叶松树干液流密度及环境因子;构建基于随机森林的树干液流主导因子分析模型,通过滞后曲线分析环境因子与树干液流的时滞效应,采用反向传播神经网络搭建树干液流预测模型进行实验验证,对比分析不同主导因子分析方法及时滞效应对液流预测的影响。【结果】(1)采用随机森林主导因子作为输入变量,长白落叶松树干液流预测精度为0.904,高于皮尔逊相关系数主导因子作为输入的预测精度。(2)在输入变量中引入时滞效应,能够有效提高预测精度。【结论】随机森林模型可以更好的分析树干液流密度的主导因子,利用随机森林分析结果并引入时滞效应能够有效提高树干液流密度的预测精度。

【Objective】The study of the variation of plant trunk sap density and its response to environmental factors is important for understanding plant transpiration,plant physiological and biochemical characteristics and plant water consumption patterns.The Pearson correlation coefficient-based method for the analysis of the dominant factors of trunk sap flow cannot accurately describe the non-linear relationship between sap flow density and environmental factors.This paper provides a new technical tool for the analysis of the dominant factor of stem sap density and the prediction of stem sap density.【Method】In this paper,a method for analyzing the dominant factors of trunk sap flow based on random forest is proposed.Taking Larix olgensis in Mengjiagang Forest,Jiamusi City,Heilongjiang Province as the research object.The Grainer probe and the meteorological factor observation station were used to monitor the sap flow density and environmental factors of Larix olgensis.The random forest-based dominant factor analysis model for stem sap flow was constructed,the time lag effect between environmental factors and tree trunk sap flow was analysed by lag curves.The back propagation neural network was used to build a sap flow prediction model for experimental validation,and the effects of different dominant factor analysis methods and time lag effects on sap flow prediction were compared.【Result】(1)Using the random forest dominant factor as the input variable,the coefficient of determination of trunk sap flow of Larix olgensis was 0.904,which was higher than the prediction accuracy of Pearson's correlation coefficient dominant factor as the input.(2)The prediction accuracy is improved by introducing time lag effects in the input variables.【Conclusion】The random forest model can better analyze the dominant factors of stem sap flow density,and using the results of random forest analysis and introducing the time lag effect can effectively improve the prediction accuracy of stem sap flow density.