This website shows the concentration of particulate matter air pollution (PM2.5) in cities around the world. Very few historical observations of PM2.5 exist before the year 2000 so instead we use data produced from a mix of computer model simulations and satellite observations.
For the most recent years (2000-2021) we use a dataset that combines ground-level and satellite observations of PM2.5 concentrations from Van Donkelaar et at (2021, V5 0.1 degree resolution), this dataset can be found here.
Satellite observations of PM2.5 aren’t available for the years before 1998, so instead we take the historical trend in air pollution concentrations from computer models (Turnock 2020); publicly available model data was taken from the Coupled Model Intercomparison Project (CMIP6) which is made freely available via the Earth System Grid Federation (ESGF), these are the climate models used for the . We used data from the UKESM submission to CMIP (data is here). The historical concentrations for the UKESM model are calculated using the emissions inventory developed through the Community Emissions Data System (CEDS) by Hoesly et al, 2018 -
Modelling global concentrations of pollutants is very challenging, and models are continuously evaluated and improved. Previous research has shown that the CMIP6 multi-model simulations tend to underestimate PM2.5 concentrations when compared to global observations (Turnock et al., 2020). To address this issue and to ensure a smooth time series between the model and satellite data, we take the following steps: for each city, we first calculate a three-year (2000-2002) mean of the satellite data for that city. Next, we calculate the three-year (2000-2002) mean of model concentrations for the same city. The ratio between these values represents the model's bias compared to observations. We then adjust (or "weight") the model values using this ratio. This is a similar approach to that taken by Turnock et al. (2023) and Reddington et al. (2023).
+Modelling global concentrations of pollutants is very challenging, and models are continuously being evaluated against observations to improve their representation of physical and chemical processes. Previous research has shown that the CMIP6 multi-model simulations tend to underestimate PM2.5 concentrations when compared to global observations (Turnock et al., 2020). To address this issue and to ensure a smooth time series between the model and satellite data, we take the following steps: for each city, we first calculate a three-year (2000-2002) mean of the satellite data for that city. Next, we calculate the three-year (2000-2002) mean of model concentrations for the same city. The ratio between these values represents the model's bias compared to observations. We then adjust (or "weight") the model values using this ratio. This is a similar approach to that taken by Turnock et al. (2023) and Reddington et al. (2023).
Because so few historical observations of PM2.5 exist, so it is challenging to evaluate how good this approximation is, but in our approach the historical trend is taken from the computer model and the values are informed by the satellite.
-This is the first versions of these images, they will be updated in the future as improved model simulations and observations become available.
+This is the first versions of the Air Quality Stripes, they will be updated in the future as improved model simulations and observations become available. We welcome comments and suggestions for improvements!
The data used to create these images is here.