Random forest models for PM2.5 speciation concentrations using MISR fractional AODs

Abstract

It is increasingly recognized that various chemical components of PM2.5 might have differential toxicities to human health, although such studies are hindered by the sparse or non-existent coverage of ground PM2.5 speciation monitors. The Multi-angle Imaging SpectroRadiometer (MISR) onboard the Terra satellite has an innovative design to provide information about aerosol shape, size and extinction that are more related to PM2.5 speciation concentrations. In this study, we developed random forest models that incorporated ground measurements of PM2.5 species, MISR fractional AODs, simulated PM2.5 speciation concentrations from a chemical transport model (CTM), land use variables and meteorological fields, to predict ground-level daily PM2.5 sulfate, nitrate, organic carbon (OC) and elemental carbon (EC) concentrations in California between 2005 and 2014. Our models had out-of-bag R2 of 0.72, 0.70, 0.68 and 0.70 for sulfate, nitrate, OC and EC, respectively. We also conducted sensitivity tests to explore the influence of variable selection on model performance. Results show that if there are sufficient ground measurements and predictor data to support the most sophisticated model structure, fractional AODs and total AOD have similar predicting power in estimating PM2.5 species. Otherwise, models using fractional AODs outperform those with total AOD. PM2.5 speciation concentrations are more sensitive to land use variables than other supporting data (e.g., CTM simulations and meteorological information).