Sub-grid variability and its impact on European wide air quality exposure assessment Bruce Denby a, * , Massimo Cassiani a , Peter de Smet b , Frank de Leeuw b , Jan Horálek c a The Norwegian Institute for Air Research (NILU), PO Box 100, 2027 Kjeller, Norway b The Netherlands Environmental Assessment Agency (PBL), The Netherlands c The Czech Hydro-Meteorological Institute (CHMI), Czech Republic article info Article history: Received 4 January 2011 Received in revised form 12 April 2011 Accepted 3 May 2011 Keywords: Air quality Exposure Sub-grid variability Covariance abstract Estimates of population exposure to air pollution on the European scale are required for policy devel- opment and health impact assessment. Long term exposure estimates of this type can be made using spatially distributed air quality and population density data. Gridded chemical transport models (CTMs) are often used for this purpose, however the grid resolution of CTMs that cover entire continents is usually limited to 25e100 km and there may be a significant level of unresolved variability within the grids that will impact on the exposure estimates. In this paper sub-grid variability and its impact on long term exposure estimates is assessed by investigating the covariance of concentration and population, which is shown to be the defining term in estimating the average sub-grid population exposure. A parameterisation of the sub-grid covariance is described, based on the sub-grid covariance of other proxy data, and this is applied using EMEP model results for all of Europe. The study shows that the error made in the exposure calculations for all of Europe is significant for the typical CTM resolution of 50 km. The error is largest for NO 2 , where the average European urban background exposure using CTMs is underestimated by 44  4%. Particulate matter with a diameter less than 10 mm (PM 10 ) is also under- estimated, but only by 15  4%. Calculation of the sub-grid covariance for the ozone health indicator SOMO35 (Sum of Ozone 8 h running Means Over 35 ppb) was not carried out for all of Europe as the sub- grid parameterisation was considered too uncertain for useful application. However, estimates based on observations alone show that population weighted SOMO35 concentrations are overestimated by around 13% when using model grid resolutions of 50 km. Ó 2011 Elsevier Ltd. All rights reserved. 1. Introduction The health impacts of ambient air pollution have been investi- gated in a number of studies. Many of these studies make use of gridded air quality data either directly from models, e.g. CAFÉ (IIASA, 2005), or using a combination of models and monitoring, e.g. Fiala et al. (2009), to estimate the static long term population exposure for large regions, e.g. Europe. Typical grid sizes in regional or global scale chemical transport models (CTMs) range from 10 to 100 km, dependent on the model and the application. For regional scale models, covering the European continent, a 10 km grid resolution represents the maximum feasible resolution in regard to computing time for long term (annual) calculations, but most operational regional scale CTMs will run at resolutions of 25 km or larger. The concentrations calculated by CTMs are intended to repre- sent the mean gridded concentrations in each grid. However, it is clear that there can be a substantial amount of variability within these grids, known as sub-grid variability (SGV), and this can impact upon the exposure estimates made using these models. The following question arises when using finite grid resolutions: “What is the error in the exposure calculation due to the use of mean gridded concentration data and can a correction be imple- mented to account for this?” There are a small number of studies investigating the sub-grid variability due to both grid resolution and turbulence treatment, e.g. Ching et al. (2006a) and Cassiani et al. (2010). These studies focus on the variance of the concentrations and whilst this infor- mation is useful for describing the frequency distribution of pollutant concentrations it does not tell us directly how this vari- ance will impact on the sub-grid exposure assessment. To study this further it is possible to increase model resolution and assess the differences in exposure resulting from this. In studies carried out by Ching et al. (2006b) and Isakov et al. (2007), varying CTM grid resolutions (36e4 km), as well as local Gaussian dispersion models, * Corresponding author. Tel.: þ47 63898164; fax: þ47 63898050. E-mail address: bde@nilu.no (B. Denby). Contents lists available at ScienceDirect Atmospheric Environment journal homepage: www.elsevier.com/locate/atmosenv 1352-2310/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.atmosenv.2011.05.007 Atmospheric Environment 45 (2011) 4220e4229