We report on the application and comparative assessment of two rational turbulence modelling approaches for the computer simulation of air flow and pollutant dispersion in real urban environment: a stand-alone unsteady Reynolds-averaged Navier-Stokes eddy-viscosity model (URANS), and its blend with Large-eddy simulations (LES) in a hybrid mode (HRL). The elliptic-relaxation (k−ε)ζ−feddy-viscosity model was applied in both methods. The models, verified earlier in a range of engineering flows including heat transfer, were here validated in two environmental benchmark cases, a single building and an idealized urban settlement, both subjected to a steady wind. The velocity field and pollutant concentration are better predicted by the HRL method in both benchmarks. The HRL is then applied to predictions of air flow and spreading of pollutant from road traffic in downtown of a real city (Sarajevo) containing 100 realistic buildings. The simulations were performed with the in-house open-source CFD code T-Flows, specifically optimized for a fast and efficient high-quality unstructured meshing of the terrain orography and building configurations. The benchmark validations demonstrated that, compared with the standard k−ε, in typical urban applications with complex urban shapes and arrangements the accuracy and credibility of the predictions can substantially be improved by applying a physically sounder and yet still relatively simple ζ−f eddy-viscosity model that specifically accounts for the elliptic inviscid wall-blocking effects and near-wall stress anisotropy. The benchmarking also showed that still further improvements are achieved by using the HRL scheme which ensures a rational balance in capturing the important physics and the computational economy.@en

We describe the low Mach variable-density Navier-Stokes numerical iterative solution procedure implemented in the finite-volume unstructured T-FlowS code. As the test cases we use a number of analytic manufactured solutions and Rayleigh-Taylor instability problem from the literature for algorithm verification purposes. The tests show that the code is second-order accurate in agreement with the spatial discretization scheme. We outline the recent combustion ADEF model implemented in the program.@en