摘要
大气颗粒物是最重要的空气污染物之一,会对人类健康产生负面影响。激光雷达探测是实现颗粒物分布高精度测量的可行手段。气溶胶消光系数在一定程度上能反映气溶胶质量浓度的相对大小,气象要素对消光系数和质量浓度的影响不容忽视。本团队利用反演得到的消光系数,结合地面温度、相对湿度、风速、地面气压等地面气象要素,与PM_(2.5)、PM_(10)质量浓度建立数据集;通过主成分分析法计算数据特征,基于广义回归神经网络(GRNN)对PM_(2.5)、PM_(10)质量浓度建立评估模型。GRNN模型得到的PM_(2.5)和PM_(10)质量浓度的评估值与真实值的相关系数分别为0.86和0.85,均方根误差(RMSE)分别为2.58μg/m^(3)和10.84μg/m^(3),平均绝对误差(MAE)分别为0.81μg/m^(3)和1.53μg/m^(3)。将GRNN模型应用于激光雷达扫描模式下,对南京市浦口区颗粒物质量浓度的水平分布进行了评估,评估值和实际站点测量值的一致性较好,进一步验证了GRNN模型用于颗粒物质量浓度评估的有效性。
Objective Atmospheric particulate matter is regarded as one of the most serious air pollutants that endanger human health.Lidar detection is a viable method for achieving high-precision particle distribution measurements.To some extent,the aerosol extinction coefficient reflects the relative size of aerosol mass concentration.Meteorological elements,such as temperature,relative humidity,wind speed,and surface pressure,have a significant impact on the extinction coefficient and mass concentration.In this study,a dataset of PM_(2.5) and PM_(10) was created by combining the extinction coefficient obtained from lidar and surface meteorological elements such as temperature,relative humidity,wind speed,and surface pressure.The data characteristics were calculated using principal component analysis.The mass concentration of PM_(2.5) and PM_(10) were estimated using the generalized regression neural network(GRNN)model.The results show that the correlation coefficients between the estimated mass concentrations obtained by the GRNN model and the true values collected from environmental monitoring stations for PM_(2.5) and PM_(10) were 0.86 and 0.85,respectively.Moreover,the root mean square errors(RMSEs)were 2.58μm/cm^(3)and 10.84μm/cm3,and the mean absolute deviations(MAEs)were 0.81μg/m^(3)and 1.53μg/m^(3)for PM_(2.5) and PM_(10),respectively.The GRNN model was applied to lidar scanning mode to evaluate the horizontal distribution characteristics of particle over Pukou District of Nanjing.The estimated mass concentrations of PM_(2.5) and PM_(10) were consistent with the measured values,demonstrating the GRNN model’s effectiveness in particle mass concentration evaluation.Methods The aerosol extinction coefficient,wind speed,temperature,relative humidity,and surface pressure were used as input variables in the GRNN,and the PM_(2.5) and PM_(10) mass concentrations were used as output variables for model training and verification.First,principal component analysis was used to calculate the characteristics of sample dat
作者
莫祖斯
卜令兵
王勤
林雪飞
Samuel A.Berhane
杨彬
邓晨
李智
Mo Zusi;Bu Lingbing;Wang Qin;Lin Xuefei;Samuel A.Berhane;Yang Bin;Deng Chen;Li Zhi(Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters.School of Atmospheric Physics,Nanjing University of Information Science and Technology,Nanjing 210044,Jiangsu,China;Nanjing Mulei Laser Technology Co.,Ltd.,Nanjing 210038,Jiangsu.China)
出处
《中国激光》
EI
CAS
CSCD
北大核心
2022年第17期125-135,共11页
Chinese Journal of Lasers
基金
国家自然科学基金(42175145)。
关键词
遥感
气溶胶消光系数
颗粒物质量浓度
神经网络
气象要素
remote sensing
aerosol extinction coefficient
particulate matter mass concentration
neural networks
meteorological elements
