Abstract

This study was motivated by the use in air pollution epidemiology and health burden assessment of data simulated at 5 km  ×  5 km horizontal resolution by the EMEP4UK-WRF v4.3 atmospheric chemistry transport model. Thus the focus of the model–measurement comparison statistics presented here was on the health-relevant metrics of annual and daily means of NO2, O3, PM2. 5, and PM10 (daily maximum 8 h running mean for O3). The comparison was temporally and spatially comprehensive, covering a 10-year period (2 years for PM2. 5) and all non-roadside measurement data from the UK national reference monitor network, which applies consistent operational and QA/QC procedures for each pollutant (44, 47, 24, and 30 sites for NO2, O3, PM2. 5, and PM10, respectively). Two important statistics highlighted in the literature for evaluation of air quality model output against policy (and hence health)-relevant standards – correlation and bias – together with root mean square error, were evaluated by site type, year, month, and day-of-week. Model–measurement statistics were generally better than, or comparable to, values that allow for realistic magnitudes of measurement uncertainties. Temporal correlations of daily concentrations were good for O3, NO2, and PM2. 5 at both rural and urban background sites (median values of r across sites in the range 0.70–0.76 for O3 and NO2, and 0.65–0.69 for PM2. 5), but poorer for PM10 (0.47–0.50). Bias differed between environments, with generally less bias at rural background sites (median normalized mean bias (NMB) values for daily O3 and NO2 of 8 and 11 %, respectively). At urban background sites there was a negative model bias for NO2 (median NMB  =  −29 %) and PM2. 5 (−26 %) and a positive model bias for O3 (26 %). The directions of these biases are consistent with expectations of the effects of averaging primary emissions across the 5 km  ×  5 km model grid in urban areas, compared with monitor locations that are more influenced by these emissions (e.g. closer to traffic sources) than the grid average. The biases are also indicative of potential underestimations of primary NOx and PM emissions in the model, and, for PM, with known omissions in the model of some PM components, e.g. some components of wind-blown dust. There were instances of monthly and weekday/weekend variations in the extent of model–measurement bias. Overall, the greater uniformity in temporal correlation than in bias is strongly indicative that the main driver of model–measurement differences (aside from grid versus monitor spatial representivity) was inaccuracy of model emissions – both in annual totals and in the monthly and day-of-week temporal factors applied in the model to the totals – rather than simulation of atmospheric chemistry and transport processes. Since, in general for epidemiology, capturing correlation is more important than bias, the detailed analyses presented here support the use of data from this model framework in air pollution epidemiology.