欢迎访问林业科学,今天是

林业科学 ›› 2018, Vol. 54 ›› Issue (8): 1-12.doi: 10.11707/j.1001-7488.20180801

• 论文与研究报告 • 上一篇    下一篇

基于随机森林模型的毛竹林CO2通量模拟及其影响因子

陈亮, 周国模, 杜华强, 刘玉莉, 毛方杰, 徐小军, 李雪建, 崔璐, 李阳光, 朱迪恩   

  1. 浙江省森林生态系统碳循环与固碳减排重点实验室 浙江农林大学环境与资源学院 临安 311300
  • 收稿日期:2016-09-25 修回日期:2018-06-14 出版日期:2018-08-25 发布日期:2018-08-18
  • 基金资助:
    浙江省与中国林业科学研究院省院合作林业科技项目(2017SY04);国家自然科学基金项目(31670644,31500520);浙江省自然科学基金项目(LR14C160001);浙江省竹资源与高效利用协同创新中心开放项目(S2017011)。

Simulation of CO2 Flux and Controlling Factors in Moso Bamboo Forest Using Random Forest Algorithm

Chen Liang, Zhou Guomo, Du Huaqiang, Liu Yuli, Mao Fangjie, Xu Xiaojun, Li Xuejian, Cui Lu, Li Yangguang, Zhu Di   

  1. Zhejiang Province Key Laboratory of Carbon Cycling in Forest Ecosystems and Carbon Sequestration School of Environmental and Resources Science, Zhejiang A & F University Lin'an 311300
  • Received:2016-09-25 Revised:2018-06-14 Online:2018-08-25 Published:2018-08-18

摘要: [目的]探究环境因子对毛竹林CO2通量的影响,为亚热带典型森林碳循环模拟提供技术和理论支撑。[方法]基于浙江省安吉县山川乡2011—2014年毛竹林通量塔观测数据,采用随机森林模型对毛竹林CO2通量进行模拟,以均方根误差(RMSE)、决定系数(R2)和林氏调和系数(LCCC)3个指标评价模型精度,并通过计算环境因子重要性得分来分析环境因子对毛竹林CO2通量的影响。[结果]随机森林模型能以较高精度模拟毛竹林CO2通量,但由于2013年7和8月发生极端高温干旱,模型难以刻画这种短时、剧烈的干扰,最终导致测试阶段模型精度(R2=0.845 5,RMSE=0.437 7 mg·m-2s-1,LCCC=0.914 1)低于训练阶段模型精度(R2=0.961 5,RMSE=0.005 4 mg·m-2s-1,LCCC=0.980 1);十折交叉验证表明,随机森林模型拟合效果稳定,且模型内部参数设置合理,模型误差主要来自于输入数据;在月尺度上,环境因子对毛竹林CO2通量影响的重要性得分表现为光合有效辐射PAR (63.332)> 土壤5 cm深处温度TS(29.932)> 空气相对湿度RH(25.839)> 大气温度TA(25.581)> 空气CO2浓度CCO2(25.095)> 饱和水汽压差VPD (24.123)> 风速WS(23.504)> 生态系统有效能量AE(19.323)> 土壤热通量QS(18.502),PAR对毛竹林CO2通量变化影响最大,PAR、TS和VPD对毛竹林CO2通量的影响较显著(P<0.05),这3个因子是影响月尺度上毛竹林CO2通量变化的主导因子。[结论]随机森林模型能以较高精度拟合毛竹林CO2通量;在相关环境因子中,光合有效辐射、土壤5 cm深处温度和饱合水汽压差对毛竹林CO2通量影响的贡献最显著,这3个因子对毛竹林CO2通量月尺度上的变化具有控制作用。

关键词: 毛竹林, 涡度相关技术法, 通量塔, CO2通量, 随机森林模型, 环境因子, 安吉县

Abstract: [Objective]This paper aims to investigate the influence of environmental factors on the CO2 flux of moso bamboo forest and to provide technical and theoretical support for carbon cycle simulation of typical subtropical forests.[Method]The CO2 flux of moso bamboo forest was simulated using random forest model based on eddy covariance flux data collected from 2011 to 2014, and the accuracy of model was evaluated using the mean squared root error (RMSE), the coefficient of determination (R2) and the lin's concordance correlation coefficient (LCCC). The contribution of each environmental factor to CO2 flux was analyzed based on importance score calculated using the random forest algorithm.[Result]The random forest model accuracy of testing data (R2=0.845 5, RMSE=0.437 7 mg·m-2s-1; LCCC=0.914 1) was lower than that of training data (R2=0.961 5; RMSE=0.005 4 mg·m-2s-1; LCCC=0.980 1), because the model was hard to accurately depict this kind of short, intense interference to CO2 flux during the extreme drought occurred in July and August in 2013. The accuracy was stable for different training data based on 10-fold cross-validation method and the parameters of the model were appropriately set. The error in the model was mainly caused by the input data. The importance score of each environmental factor was decreased in the following order:PAR (63.332) > TS (29.932) > RH (25.839) > TA (25.581) > CCO2 (25.095) > VPD (24.123) > WS(23.504) > AE (19.323) > QS(18.502). PAR was the dominant factor for explaining the change of CO2 flux in moso bamboo forest. Based on the significance test, monthly CO2 flux was significantly influenced by PAR, TS, and VPD (P<0.05).[Conclusion]The random forest model can simulate the CO2 flux of moso bamboo with a high accuracy; PAR, TS, and VPD remarkably affect the CO2 flux at the 0.05 significance level, according to their importance score, indicating that PAR, TS, and VPD play an important role in controlling the CO2 flux of moso bamboo forest at a monthly time scale.

Key words: moso bamboo forest, eddy covariance, flux tower, CO2 flux, random forest model, environmental factors, Anji County

中图分类号: