Given the sparse ground-level measurements of fine particle (PM2.5) mass and composition over global south locations and other data sparse regions of the world, we explore the potential to glean insights into air pollution sources by combining satellite retrievals of tropospheric trace gases and aerosols.
We train a machine learning model (XGBoost) that takes input MODIS and TROPOMI satellite data against available reference and low cost air sensor data to create a gridded surface PM2.5 dataset. Geographic locations include sub-Saharan Africa (work funded by Clean Air Fund), Puerto Rico, and data sparse regions of the USA.
These data sources will be applied to policy and epidemiology datasets, for example with our partners at Kintampo Health Research Centre in Ghana and Columbia's Mailman School of Public Health