Retrieval of Annual Air Quality Statistics from a Limited Number of LES Model Simulations

Abstrakt

The annual mean and other annual statistics, such as the n-th highest hourly or daily concentration, are an essential part of legislative air quality limits. The resulting statistics are usually calculated from data collected at a sparsely distributed network of monitoring stations and evaluated at spatial resolutions of hundreds of meters. More detailed spatial information in urban environments can be obtained from complex CFD models, which can provide air quality simulations at the street level. However, simulating the entire year is not feasible because it is computationally too expensive. In the case of RANS models, the standard approach is the simulation of steady-state fields for different wind directions and then the calculation of the required annual statistics from them. However, this approach is not suitable for LES models with time-evolving resolved turbulence. A different approach suitable for LES is to retrieve the annual statistics using a limited number of “typical” days. The set of “typical” days should be carefully identified, because it should cover the different scenarios occurring during the year. We proposed a statistical solution to this problem — identification based on k-medoids clustering. Our method was validated using observations from monitoring stations and used to retrieve the annual means of PM10 and NOx from the PALM model simulations covering most of the Prague territory at a 10-m resolution.