This paper describes a Bayesian hierarchical approach to predict short-term concentrations of particle pollution in an urban environment, with application to inhalable particulate matter (PM10) in Greater London. We developed and compared several spatiotemporal models that differently accounted for factors affecting the spatiotemporal properties of particle concentrations. We considered two main source contributions to ambient measurements: (i) the long-range transport of the secondary fraction of particles, which temporal variability was described by a latent variable derived from rural concentrations; and (ii) the local primary component of particles (traffic- and non-traffic-related) captured by the output of the dispersion model ADMS-Urban, which site-specific effect was described by a Bayesian kriging. We also assessed the effect of spatiotemporal covariates, including type of site, daily temperature to describe the seasonal changes in chemical processes affecting local PM10 concentrations that are not considered in local-scale dispersion models and day of the week to account for time-varying emission rates not available in emissions inventories. The evaluation of the predictive ability of the models, obtained via a cross-validation approach, revealed that concentration estimates in urban areas benefit from combining the city-scale particle component and the long-range transport component with covariates that account for the residual spatiotemporal variation in the pollution process.