Air pollution represents one of the most important environmental risk factors to human health with a number of pollutants associated with adverse health outcomes. Epidemiological studies designed to estimate the risks associated with air pollution require accurate measures of concentrations with comprehensive coverage of a study area. Traditionally, these have been based on measurements from ground monitors but this may not provide information of sufficient quality and coverage to allow accurate spatial (and temporal) prediction at any location at which estimated concentrations are required. Ground monitoring data may therefore need to be supplemented with information from other sources, such as estimates from satellite remote sensing, chemical transport models, land use and topography. Set within a Bayesian hierarchical modelling framework, downscaling models are used to align data generated at different geographical resolutions, including point locations and a series of (potentially non-aligned) grids of varying resolutions. The proposed modelling approach is used to predict concentrations of nitrogen dioxide, together with measures of uncertainty, at a high-resolution throughout Western Europe by integrating data from ground monitoring, chemical transport models together with land-use information. Performing complex Bayesian inference combining potentially large datasets may be computationally challenging and we perform approximate Bayesian inference using INLA.