In this study Computational Fluid Dynamics (CFD) are used to predict the air flow structure and pollutant retention inside urban canyons taking into account thermal effects due to solar radiation in order to imitate diurnal heating over a city. The diurnal heating induces buoyancy therefore the flow structure and pollutant concentration will be completely different depending on the daytime. The presented method departs from a 2D domain comprised by eight urban canyons of aspect ratio 1 and different 2D transient RANS (Reynolds-Averaged Navier-Stokes) simulations that have been conducted using OpenFoam. Isothermal, single-surface heating and multi-surface heating cases are considered. For all of them, the results obtained have been post-processed using ParaView and Matlab to be able to analyse all the information related with velocity, pressure, temperature, pollutant concentration, streamlines, etc. These results have been also contrasted and validated with data obtained from wind-tunnel experiments and other CFD simulations which have been conducted using a 2D model and conditions similar to the ones used in this thesis. Finally, the conclusions obtained for an airflow with a Reynolds number of 12000 are that for most cases, a single main vortex is formed which is strengthen for the cases where leeward or ground surfaces are heated up compared to the vortex formed in isothermal case. In all cases where the windward side is heated up, a second counter-rotating vortex appears due to buoyancy and displaces the main one from the center of the canyon. Also, when the windward side is heated, the concentration of pollutants increases at the pedestrian level which may be considered as the worst case. Further improvements may be considered to continue with this work such as improving the mesh quality, considering an additional heat source, using another turbulence modelling (LES) or even considering a 3D model since the buoyant flows have an unsteady and 3D characteristc that require complex numerical models to obtain more accurate results