Computational Fluid Dynamics (CFD) simulations can be used to assess indoor natural ventilation by solving the interaction between the urban wind flow and the indoor airflow. The air exchange rate (ACH) can be obtained from the simulated volume flow rates through the ventilation openings or by the concentration decay method that is often used in experimental studies. This paper presents 3D unsteady Reynolds-averaged Navier-Stokes (RANS) CFD simulations to reproduce the decay of CO2 concentration in a large semi-enclosed stadium. The study focuses on the hours after a concert, when the indoor CO2 concentration generated by the attendants has reached a maximum. The wind flow, indoor airflow and dispersion of heat, water vapour and CO2 are modelled on a high-resolution grid based on grid-sensitivity analysis. The simulations are validated with on-site measurements of wind velocity and CO2 concentration decay. The validated CFD model is used to analyse the significant horizontal and vertical CO2 concentration gradients in the stadium, showing local differences at t = 300 s up to 700 ppm (i.e. 37% of the maximum of 1900 ppm). A specific piecewise linear technique is applied for the concentration decay method to determine the ACH values for smaller time intervals. This is needed because the plotted semi-logarithmic decay curve itself is not linear because the ventilation rate changes over time, due to the changing buoyancy forces. It shows that the ACH values decrease from about 2 h-1 at the beginning of the concentration decay simulations to about 0.3 h-1 at the end.