基于不同周期PM_(2.5)组成高时间分辨观测的PMF源解析研究

PMF source apportionment based on high time-resolved measurements of PM_(2.5) components during different observation periods

于2017年1月1日—12月31日对南京市城区大气细粒子(PM_(2.5))化学组分(元素、水溶性离子和碳质组分)的小时质量浓度进行连续观测,采用正矩阵因子分析(Positive Matrix Factorization,PMF)模型分别基于全年观测数据(PMF全年)和逐月观测数据(PMF月份)进行源解析,比较不同观测周期源解析结果的差异以及对PM_(2.5)各组分浓度估算的准确性.结果表明:不同观测周期下,PMF源解析结果中因子类型未发生改变,但因子组成和贡献分布存在较大差异.由于PMF模型假设同一观测周期内源成分谱不发生变化,只有基于逐月观测数据的PMF源解析才能体现全年范围内因子组成和贡献分布的变化.尽管PMF全年和PMF月份的分析结果均能准确估算PM_(2.5)组分的月均浓度,但PMF月份结果对各组分小时浓度的估算值和观测值在时间变化上更一致.这是因为PMF模型要求对各组分浓度的平均值进行拟合,易低估(或高估)PM_(2.5)组分在观测周期内的极大(或极小)值.因此,基于短期(例如,月份)高分辨观测数据的PMF分析能改进对PM_(2.5)组分浓度时间变化的模拟.

In this study,hourly-resolved fine particle(PM_(2.5))components(elements,water-soluble inorganic ions,and carbon contents)were continuously monitored in urban Nanjing from January 1 to December 31,2017.Positive Matrix Factorization(PMF)was deployed for source apportionment based on full-year(PMFfull-year)and month-by-month(PMFmonth)measurement data,and the source apportionment results for different observation periods and PMF estimations of PM_(2.5)component concentrations were compared.The results indicated that the identified types of factors were the same,but the factor profiles and distributions differed substantially between PMFfull-year and PMFmonth solutions.As the PMF model presumes constant factor profiles during a certain observation period,only the analysis based on month-by-month data can reflect the changes in factor profiles and distributions for the whole year.Although both PMFfull-year and PMFmonth estimated monthly averages of PM_(2.5)components accurately,the hourly concentration time series of individual species derived from PMFmonth results were in better agreement with measurements.This is because the PMF model requires that the measured average concentrations to be fit well,and tends to underestimate(or overestimate)the extremely large(or small)values during a specific observation period.Therefore,the PMF analysis based on short-term(e.g.,monthly)measurement data with high time-resolution can improve the simulation of temporal variations of PM_(2.5)components.