中国农业气象 ›› 2021, Vol. 42 ›› Issue (05): 390-401.doi: 10.3969/j.issn.1000-6362.2021.05.004

• 农业生态环境栏目 • 上一篇    下一篇

基于随机森林方法分析环境因子对空气负离子的影响

施光耀,周宇,桑玉强,张劲松,孟平,蔡露露,裴松义,王尧   

  1. 1.中国林业科学研究院林业研究所/国家林业局林木培育重点实验室, 北京 100091;2.南京林业大学南方现代林业协同创新中心,南京 210037;3.河南农业大学,郑州 450002;4.河南省地球物理空间信息研究院,郑州 450016;5.国有建平县黑水机械化林场, 朝阳 122000
  • 收稿日期:2020-10-25 出版日期:2021-05-20 发布日期:2021-05-16
  • 通讯作者: 张劲松,研究员,研究方向为林业气象,E-mail:zhangjs@caf.ac.cn E-mail:zhangjs@caf.ac.cn
  • 作者简介:施光耀,E-mail:shiguangyao01@163.com
  • 基金资助:
    中央级公益性科研院所基本科研业务费专项资金项目(CAFYBB2018ZA002)

Influence of Environmental Factors on Negative Air Ion Using Random Forest Algorithm

SHI Guang-yao, ZHOU Yu, SANG Yu-qiang, ZHANG Jin-song, MENG Ping, CAI Lu-lu, PEI Song-yi, WANG Yao   

  1. 1. Research Institute of Forestry, Chinese Academy of Forestry/Key Laboratory of Tree Breeding and Cultivation, State Forestry Administration, Beijing 100091, China; 2. Co-Innovation Center for Sustainable Forestry in Southern China, Nanjing Forestry University, Nanjing 210037; 3. Henan Agricultural University, Zhengzhou 450002; 4. Henan Geophysical Space Information Research Institute, Zhengzhou 450016; 5. State Owned Jianping County Heishui Mechanized Forest Farm, Chaoyang 122000
  • Received:2020-10-25 Online:2021-05-20 Published:2021-05-16

摘要: 空气负离子是衡量一个地区空气清洁度的重要指标,对人体的心理和生理机能的调节发挥着重要作用。随着森林生态旅游的兴起,空气负离子的发生过程及影响机制已成为研究热点。本研究基于华北低丘山地森林植被主要生长季的气象数据和栓皮栎人工林空气负离子浓度观测资料,利用机器学习中随机森林模型从非线性角度全面分析确定影响空气负离子浓度变化的重要环境因子,通过独立样本对构建的随机森林模型进行模拟和检验,确定模型的预测精度,同时筛选出对空气负离子影响程度最大的环境因子。结果表明:随机森林模型在分析环境因子对空气负离子影响方面具有较高的精度以及较好的拟合效果,通过对模型的拟合值与实测值进行验证,均方根误差(RMSE)为59.349,决定系数R2达到了0.887。同时利用独立样本数据对随机森林模型进行十折交叉验证,决定系数R2均达到了0.904以上,且均方误差(RMSE)较小,为24.851。此外,模型筛选出影响空气负离子的主要因素,按重要性排序依次为颗粒物PM2.5(48.037)、饱和水汽压差(46.169)、土壤湿度(43.984)、风速(43.779)、紫外辐射(41.130)、土壤温度(40.107)、总辐射(36.838)、大气压力(34.532),其中对模型重要性贡献相对较高的3个变量分别为颗粒物PM2.5、饱和水汽压差和风速,它们对空气负离子的影响起决定性作用。因此,随机森林模型适合分析环境因子对空气负离子影响,且拟合效果精度高,稳定性强。

关键词: 空气负离子, 环境因子, 机器学习, 随机森林, 栓皮栎

Abstract: Negative air ion(NAI) is an essential indicator for measuring the air cleanliness of a given area, which plays an important role in promoting the psychological and physiological functions of the human body. With the development of forest eco-tourism, NAI has attracted substantial attention, while research on NAI has become increasingly active, especially for the topics of the occurrence process and impact mechanism of NAI in related fields. Based on meteorological data and observation data of NAI during the main growing season of Quercus variabilis BI. plantation in the hilly area of North China, the random forest model in machine learning was used to analyze the environmental factors that affected NAI concentration changes from a non-linear view, and independent samples were used to simulate the random forest model to determine the prediction accuracy of the model. The estimation model of NAI was established for revealing the response mechanism and predicting the response pattern of NAI to environmental factors for further research. Results showed that the random forest model had higher accuracy and better fitting effect in analyzing the impact of environmental factors on NAI, and by verifying the fitted and measured values of the model, the root mean square error(RMSE) was 59.349, and the coefficient of determination R2 reached 0.887. While using independent test data to 10-fold cross-validation of the random forest model, the average R2 was above 0.904 and the root mean square error(RMSE) was small at 24.851. In addition, the model screened out that the main factors affecting NAI were particulate matter PM2.5(48.037), vapor pressure deficit(46.169), soil moisture(43.984), wind speed(43.779), ultraviolet radiation(41.130), soil temperature(40.107), direct radiation(36.838) and atmospheric pressure(34.532) sorted by importance scores. Among them, the three variables contributed relatively high importance to the model were particulate matter PM2.5, vapor pressure deficit and wind speed, which prove that they play a decisive role in the variations of NAI. Therefore, the random forest model is better to simulate the NAI with high accuracy and strong stability.

Key words: Negative air ion, Environment factor, Machine learning Random forest model, Quercus variabilis BI.