Symmetry Patterns and Machine Learning Prediction of Urban Pollutant Dispersion

Urban pollutant dispersion is governed by highly nonlinear interactions between flow structures, turbulence characteristics, and scalar transport processes, making accurate prediction computationally demanding. The present study investigates pollutant dispersion behaviour within an idealised urban configuration using a combined computational fluid dynamics (CFD) and surrogate machine learning. Reynolds-averaged Navier–Stokes simulations were performed to analyse the effects of Reynolds number, Schmidt number, turbulence kinetic energy, concentration transport, velocity distribution, and injection ratio on flow symmetry and pollutant redistribution. The numerical results showed that variations in Reynolds number and injection ratio modified the spatial organisation and relative symmetry of the computed vorticity and concentration fields, together with the corresponding centreline concentration, velocity, and turbulent kinetic energy responses. To complement the CFD analysis, an Extreme Gradient Boosting (XGBoost) model was trained to predict concentration, velocity, and turbulence kinetic energy using CFD-generated datasets. Feature importance analysis further revealed physically meaningful relationships among the governing transport variables, with Reynolds number dominating velocity and turbulence kinetic energy behaviour, while concentration transport remained strongly influenced by injection ratio.