• 关键词：浙江省  PM_2.5  时空分布  相关因子  随机森林模型  多元线性回归

• 基金项目：国家自然科学基金（No.31770697）；2017年浙江省访问工程师项目（No.FG2017240）；浙江省教育厅一般项目（No.Y201840513）

• 陈兵红
• 丽水职业技术学院, 丽水 323000

• 靳全锋
• 1. 丽水职业技术学院, 丽水 323000;2. 福建农林大学林学院, 福州 350002

• 柴红玲
• 丽水职业技术学院, 丽水 323000

• 郭福涛
• 福建农林大学林学院, 福州 350002

• 摘要：该研究以浙江省2014-2019年PM_2.5浓度数据为研究对象，应用多元线性回归和随机森林方法结合气象、植被、地形、经济、人口和基础设施等因子进行分析.研究结果表明PM_2.5浓度时空分布不均匀，时间上季节变化差异显著，总体呈冬季>春季>秋季>夏季分布规律，每年呈下降趋势；空间上呈西北多东南少的分布特征.多元线性回归和随机森林模型显示日最低地表气温（MI-GST）、日最低气压（MI-PRS）、日蒸发量（EVP）、日最小相对湿度（MI-RHU）、月植被覆盖度（FVC）、日降水量（PRE）、日极大风速（MM-WIN）、日平均相对湿度（AV-RHU）、铁路密度（Railway）、日最大风速（MA-WIN）、日照时长（SSD）、海拔（DEM）、日平均风速（AV-WIN）和河流密度（River）等15个因子对PM_2.5浓度影响显著；随机森林模型均方根误差（RMSE）、均方绝对百分比误差（MAPE）和变异解释量（R²）分别为0.133、17.83%和0.834，明显优于多元线性回归（0.278、40.48%和0.575），表明随机森林更适合浙江省PM_2.5浓度估测，该研究揭示PM_2.5时空分布及相关因子分析，为限制空气污染提供有效策略.

• Abstract：The multiple linear regression and random forest methods, which combine factors including meteorology, vegetation, terrain, economy, population and infrastructure, were applied to analyze the data of PM_2.5 concentrations in the Zhejiang region from 2014 to 2019. Results show that the spatial and temporal distribution of PM_2.5 concentrations varied, with a significant seasonal order of winter > spring > autumn > summer, and the trend of annual PM_2.5 concentrations gradually decreased over the study period. The spatial concentration distribution of PM_2.5 in northwest part of the province is much higher in the northeast part of the province. Results of the multiple linear regression and random forest models show that PM_2.5 concentrations were significantly impacted by a variety of 15 factors, such as the daily minimum surface temperature (MI-GST), daily minimum pressure (MI-PRS), daily evaporation (EVP), daily minimum relative humidity (MI-RHU), monthly vegetation coverage (FVC), daily precipitation (PRE), daily maximum wind speed (MM-WIN), daily average relative humidity (AV-RHU), railway density (Railway), daily maximum wind speed (MA-WIN), sunshine duration (SSD), digital elevation model (DEM), wind speed (AV-WIN) and river density (River). The root mean square error (RMSE), absolute mean square error (MAPE) and variance interpretation (R²) of the random forest model were 0.133, 17.83% and 0.834, respectively, which were substantially better than the multiple linear regression analysis of 0.278, 40.48% and 0.575. This indicates that the random forest model is better for the estimation of Zhejiang data of PM_2.5 concentrations. Overall this study characterized the spatiotemporal distribution and correlation factors for PM_2.5 concentrations, which can provide important information on which to base an effective strategy for controlling air pollution.

• Key words：Zhejiang Province  PM_2.5  spatiotemporal distribution  correlation factors  random forest model  multiple linear regression

• 
  摘要点击次数： 568 全文下载次数： 660