Authors: Colleen Reid*, University of Colorado, Boulder, Ellen Considine, University of Colorado Boulder, Melissa Maestas, University of Colorado Boulder, Gina Li, University of Colorado Boulder
Topics: Spatial Analysis & Modeling, Global Change, Medical and Health Geography
Keywords: air pollution, machine learning, wildfires, health, spatiotemporal modeling
Session Type: Paper
Start / End Time: 8:00 AM / 9:15 AM
Room: Governors Square 16, Sheraton, Concourse Level
Presentation File: No File Uploaded
Fine particulate matter (PM2.5) levels are declining in many areas of the US due to policies and enforcement of the Clean Air Act. In much of the western US, however, PM2.5 concentrations have been increasing, likely due to the increased presence of wildfires in this region. There is growing evidence of various health impacts of PM2.5 exposures, even at levels below the federal standard. Health studies of PM2.5 in the western US 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. Previously made daily fine-resolution estimates of PM2.5 for the whole US perform poorly in the western US. We have tailored a machine learning model to the western US, combining satellite, meteorological, monitoring, land use and other spatiotemporal data to estimate daily PM2.5 estimates at the census tract, ZIP code, and county levels during 2008-2018. Our methods improve upon previous models by: use of a more extensive monitoring station network to train the data, thus capturing more spatial locations and proximity to wildfires; use of ensembles of machine learning algorithms, which have been shown to improve model performance; and coverage of a longer period of time.