Estimating High-Resolution Global Daily Ambient and Wildfire-related PM2.5 using Machine Learning Models – 2021 Seed Funding from the Centre for Air pollution, energy and health Research
In this study, we aim to estimate the global daily concentrations of ambient PM2.5 and wildfire PM2.5 based on data of multiple predictors, such as meteorological factors, satellite observations (e.g., land use information, fire size and behaviour), and inputs from GEOS-Chem model.
First, we will link data of predictors with ground-based observations of PM2.5 from 11,404 monitor stations in 85 countries and regions during 2014−2019 (data already collected, Figure 1).
Based on this dataset, we will train a machine learning model that can accurately predict daily ambient PM2.5 with data of predictors.
With the trained model, we will then predict global high-resolution (0.1° × 0.1° or finer) daily ambient PM2.5 from 2014 to 2019 based on data of predictors.
Finally, the global wildfire PM2.5 at the same spatial and temporal resolution will be estimated as the differences between the simulated ambient PM2.5 with and without fire events.