Using a combination of satellite data, meteorological analyses, ground based PM2.5 observations, and machine learning we have created a global estimate of the PM2.5 abundance (micro grams per cubic meter) at a resolution of 1 degree x 1 degree (100 km x 100 km). Movies showing examples for January 2001 are shown above.
We use aerosol optical depth (AOD) information from multiple platforms, this helps to fill in the gaps caused by cloud cover. The relationship between aerosol optical depth and PM2.5 abundance is a function of many factors including surface type, wind speed, atmospheric pressure, temperature, humidity and the height of the planetary boundary layer. If these other factors are not considered the correlation between the PM2.5 abundance and the AOD is poor, as can be seen in the figure below left. When we include the other factors mentioned the situation is improved considerably, as can be seen in the figure below right. We use a non-linear multi-variate non-parametric fit provided by machine learning to estimate the PM2.5 abundance from the observed AOD and the surface type, wind speed, atmospheric pressure, temperature, humidity and the height of the planetary boundary layer.