期刊名称：Journal of Integrative Environmental Sciences
印刷版ISSN：1943-815X
电子版ISSN：1943-8168
出版年度：2020
页码：1-28
DOI：10.1080/1943815X.2020.1768550
出版社：Taylor & Francis
摘要：Statistical relationships between surface ozone (O 3 ) concentration, precursor species and meteorological conditions in continental South Africa were examined from data obtained from measurement stations in north-eastern South Africa. Three multivariate statistical methods were applied in the investigation, i.e. multiple linear regression (MLR), principal component analysis (PCA) and –regression (PCR), and generalised additive model (GAM) analysis. The daily maximum 8-h moving average O 3 concentrations were considered in these statistical models (dependent variable). MLR models indicated that meteorology and precursor species concentrations are able to explain ~50% of the variability in daily maximum O 3 levels. MLR analysis revealed that atmospheric carbon monoxide (CO), temperature and relative humidity were the strongest factors affecting the daily O 3 variability. In summer, daily O 3 variances were mostly associated with relative humidity, while winter O 3 levels were mostly linked to temperature and CO. PCA indicated that CO, temperature and relative humidity were not strongly collinear. GAM also identified CO, temperature and relative humidity as the strongest factors affecting the daily variation of O 3 . Partial residual plots found that temperature, radiation and nitrogen oxides most likely have a non-linear relationship with O 3 ,while the relationship with relative humidity and CO is probably linear. An inter-comparison between O 3 levels modelled with the three statistical models compared to measured O 3 concentrations showed that the GAM model offered a slight improvement over the MLR model. These findings emphasise the critical role of regional-scale O 3 precursors coupled with meteorological conditions in daily variances of O 3 levels in continental South Africa.
关键词：Tropospheric ozone (O3); multiple linear regression; principal component analysis; generalized additive models; Welgegund
