Intercomparison of Air Quality Models in a Megacity: Toward an Operational Ensemble Forecasting System for S?o Paulo

An intercomparison of four regional air quality models is performed in the tropical megacity of Sao Paulo with the perspective of developing a forecasting system based on a model ensemble. Modeled concentrations of the main regulated pollutants are compared with combined observations in the megacity center, after analyzing the spatial scale of representativeness of air monitoring stations. During three contrasting periods characterized by different types of pollution events, the hourly concentrations of carbon monoxide (CO), nitrogen oxides (NOx), sulfur dioxide (SO2), and particulate matter (PM2.5 and PM10) modeled by the ensemble are in moderate agreement with observations. The median of the ensemble provides the best performance (R approximate to 0.7 for CO, 0.7 for NOx, 0.5 for SO2, 0.5 for PM2.5, and 0.4 for PM10) because each model has periods and pollutants for which it has the best agreement. NOx concentration is modeled with a large inter-model variability, highlighting potential for improvement of anthropogenic emissions. Pollutants transported by biomass burning events strongly affect the air quality in Sao Paulo and are associated with significant inter-model variability. Modeled hourly concentration of ozone (O3) is overestimated during the day (approximate to 20 ppb) and underestimated at night (approximate to 10 ppb), while nitrogen dioxide (NO2) is overestimated at night (approximate to 20 ppb). The observed O3 concentration is best reproduced by the median of the ensemble (R approximate to 0.8), taking advantage of the variable performance of the models. Therefore, an operational air quality forecast system based on a regional model ensemble is promising for Sao Paulo. Forecasting air quality in megacities is particularly difficult due to the diversity and temporal variability of emission sources. Sao Paulo is the largest metropolitan area in South America and has no operational air quality forecast. We perform an intercomparison of four regional air quality models with the perspective of developing an air quality forecasting system. During three contrasting periods characterized by different types of pollution events, we analyze the modeled concentrations of the main regulated pollutants (trace gases and aerosols) compared to observations from the Sao Paulo air quality monitoring network. The modeled concentrations of the main regulated pollutants agree well with the observations range of variation, although we show the potential in improving the treatment of anthropogenic emissions. In addition, the long-range transport of pollutants due to forest fires strongly affects the air quality in Sao Paulo and also reduces the performance of the models. The observed hourly ozone concentration is well reproduced by the models, and its median has the best performance, taking advantage of the capabilities of each model. Therefore, an operational air quality forecasting system for the megacity of Sao Paulo is promising. An ensemble of regional air quality models performs well in the Sao Paulo megacity for the main regulated pollutants (CO, NOx, O3, SO2, PM2.5, and PM10)Transport of pollutants due to biomass burning events, affecting strongly the air quality of the megacity, is represented with high variability by the ensembleIn the center of the megacity, the median of the regional model ensemble leads to the best performance for these pollutants compared to each model that composes it