Using machine learning to derive daily PM2.5 concentration estimates at fine spatial scales for the western US 2008-2018

Keywords: spatiotemporal modeling, air pollution, health, machine learning
Session Type: Virtual Paper Abstract
Day: Friday
Session Start / End Time: 2/25/2022 05:20 PM (Eastern Time (US & Canada)) - 2/25/2022 06:40 PM (Eastern Time (US & Canada))
Room: Virtual 30

Abstract

Fine particulate matter (PM2.5) levels are declining in many areas of the US due to enforcement of the Clean Air Act, yet, in much of the western US, PM2.5 concentrations have been increasing, likely due to increased wildfire activity. There is growing evidence of various health impacts of PM2.5 exposures, even at levels below the federal standard, yet these studies are limited by spatial sparseness of monitoring data. To improve population exposure assessment of PM2.5, researchers are increasingly using statistical methods to “blend” information from multiple data sources to better estimate PM2.5 in space and time. We created daily PM2.5 concentration estimates at the centroids of each county, ZIP code, and census tract across the western US, from 2008–2018 from ensemble machine learning models trained on 24-hour PM2.5 measurements from monitoring station data across 11 states in the western US. Predictor variables were derived from satellite, land cover, chemical transport model (just for the 2008–2016 model), and meteorological data. Ten-fold spatial and random CV R2 were 0.66 and 0.73, respectively, for the 2008–2016 model and 0.58 and 0.72, respectively for the 2008–2018 model. Comparing areal predictions to nearby monitored observations demonstrated overall R2 of 0.70 for the 2008–2016 model and 0.58 for the 2008–2018 model, but we observed higher R2 (>0.80) in many urban areas. Our data is publicly available for use in future studies of the health impacts of fine particulate air pollution in the western US.