top of page

Air quality model intercomparison


Global forecasts of air quality are now available. The resolution of these forecasts is coarse compared to the size of a megacity. Also, the anthropogenic emissions are strong and their spatial and temporal variability is high in densely populated areas.  Diurnal cycles change between days of a week, pollutant concentrations along major roads with heavy traffic can be extremely elevated, by nearby green areas or residential neighborhoods can show a completely different scenario. Therefore, there is a need for high resolution modelling of air quality in order to accurately forecast the pollutant concentrations in megacities, such as São Paulo.

Moreover, a measurement network is necessary to evaluate and build an accurate model-based forecasting system. In São Paulo, the CETESB (Companhia Ambiental do Estado de São Paulo, provides an excellent dataset to analyze the modelling outputs from different air quality simulations with 63 available measurement sites in the São Paulo region.
The model intercomparison that we carry out within the framework of Klimapolis aims to analyze the performance of the Chemistry-Transport models compared to air quality observations. The ultimate goal is to provide an operational system of air quality predictions for South America with a focus on the largest urban areas in Brazil, based on an ensemble forecast of the air quality using several models. This approach has already been used in different regions, showing the added value of the high-resolution modelling compared to the global forecast.

We evaluate strengths and weaknesses of the forecasted concentrations of the regulated pollutants (CO, O3, NO2, PM2.5, PM10), during three contrasting periods. For the first step, four Chemistry-Transport models are involved in the intercomparison using high-resolution, lower than 5 km.


Three periods of 15-days have been selected which correspond to specific events:

  1. From 27 January to 12 February 2019, five days above the ozone air quality standard in São Paulo were monitored despite the intense precipitation occurring during this period.

  2. From 8 to 21 August 2019, a period during which Biomass Burning aerosols from the Amazon basin transported to São Paulo, have created “black rain”.

  3. From 6 to 20 September 2019, there were episodes of high concentration of ozone and aerosols (PM2.5 and PM10).

These three periods are presented for CO, NO2, O3 and PM2.5 daily concentrations in Figure 1.  Using the CETESB network, we estimate a “city center” mean and standard deviation for these concentrations. To do so, we interpolated the observed and modelled data at a specific point, for São Paulo center (reference : Catedral da Sé). The city center mean is compared to the average of all the stations (orange line compared to green line in Figure 2).  Four additional areas will be studied in São Paulo state.


Figure 1: Map of the CETESB station locations. The circles represent the studied areas.


Figure 2: Daily concentrations of CO, NO2, O3 and PM2.5 during the year 2019 in São Paulo using all stations of the CETESB network that are within a radius of 30 km from the center (defined as Catedral da Sé). The three studied periods are highlighted with the red frames.



Five institutes are involved in the activity, three institutes in Brazil and two in Germany. Three different air quality models will be used with four different model set-ups.

  1. The Max Planck Institute for Meteorology (MPI-M) provides simulations made with the WRFchem model. The simulations will be led by Dr. Adrien Deroubaix.

  2. The Universidade Federal de Minas Gerais (UFMG) provides simulations made with the WRF-SMOKE-CMAQ model. The simulations will be led by Prof. Taciana Toledo de Almeida Albuquerque.

  3. The Universidade Federal do Rio Grande do Norte (UFRN) together with the Rhenish Institute for Environmental Research (RIU) at the University of Cologne provide simulations made with EURAD-IM (EURopean Air Pollution Dispersion - Inverse Model). The simulations will be led by Prof. Judith Hoelzemann and Dr. Ediclê de Souza Fernandes Duarte.

  4. The Instituto de Astronomia, Geofísica e Ciências Atmosféricas (Universidade de São Paulo – USP) provides simulations made with the WRFchem model, led by Prof. Rita Yuri Ynoue and Prof. Maria de Fatima Andrade.

The model set-ups used by each institution is different depending on their input datasets for emission and meteorology, and also the chemical scheme, the spatial and temporal resolution, etc.  The aim of the intercomparison is to take advantage of this variability in the set-up of the models to identify the possible spread in the modelled air pollutant concentrations. The aim is to cover the range of possible forecasts, which represents the uncertainties in air quality forecasting using state-of-the-art meteorological-chemistry models. Previous studies have shown that the average of the models is often better than each individual model (e.g. Brasseur et al., 2019, ACP). We will analyze this question in the São Paulo region (see Figure 3 showing an example of analysis corresponding to 26-31 January 2019. 


Figure 3: Example of analysis of modelled (WRF-CHEM by MPI-M) and observed ozone (O3) concentration using (a) hourly data during the period 26-31 January 2019, and (b) average hourly concentrations and standard deviation. The modeled planetary boundary layer height (PBLH) are the green lines. The set-ups of the two models differ only in the number of vertical levels used, which is fixed at 37 for model 1 and 47 for model 2.

bottom of page