A microscale hybrid modelling system to assess the air quality over a large portion of a large European city

The role of atmospheric dispersion models is becoming increasingly relevant to assess air pollution urban population exposure for epidemiological studies. Estimating urban air quality is challenging, because of the intrinsic characteristics of cities atmospheric structure, such as high density of primary emissions and presence of local dispersion processes, that produce strong concentration gradients. Therefore, very high spatial resolution simulations may often be required to improve the accuracy of estimations. The objective of this study is developing a microscale hybrid modelling system (HMS) to carry out, in a reasonable computational time, long-term simulations providing hourly concentration fields at building-resolving scale in extended urban areas in order to calculate annual indicators to evaluate exposure. The proposed system couples two atmospheric dispersion models suited for different scales: a Eulerian chemical transport model, FARM (Flexible Air quality Regional Model), accounting for dispersion phenomena due to regional and local emission sources, and a Lagrangian particle micro-scale dispersion model, PMSS (Parallel Micro Swift Spray), used to compute concentrations induced by vehicular traffic inside the city. The HMS has been applied on 12 × 12 km2 domain in Rome with a horizontal resolution of 4 m for calculating NO2 and PM10 concentrations for all year 2015. This study has been carried out in the frame of project BEEP (Big data in Environmental and occupational Epidemiology), that is an Italian research project in epidemiological field. Results show that the combined use of the two models reproduces the spatial and temporal variability of the observed atmospheric pollutants with a good agreement. The statistical analysis performed on daily average concentrations proves that the HMS suits the standard acceptance criteria for urban dispersion model evaluation, with a FAC2 of 0.92 and 0.80 and a Fractional Bias of −0.03 and −0.2 for NO2 and PM10 respectively. Furthermore, the implementation of an innovative kernel method to calculate concentrations in PMSS has made possible to reduce the computational time by 80%, leading to an average computational time of 3 h per simulated day on an HPC (High Performance Computing) system with 180 cores.