A008-0002
GOES16-Based Estimation of Hourly PM2.5 Levels during the Camp Fire Episode in California

Monday, 7 December 2020
Poster
Bryan N. Vu1, Jianzhao Bi1, Amy K Huff2, Shobha Kondragunta3 and Yang Liu4, (1)Emory University, Atlanta, GA, United States, (2)Pennsylvania State Univ, University Park, PA, United States, (3)NOAA College Park, College Park, MD, United States, (4)Emory University, Gangarosa Department of Environmental Health, Atlanta, GA, United States
Abstract:
Wildland fires release vast amounts of PM2.5 into the atmosphere, which may be transported via smoke plumes to a long distance downwind, resulting in excess mortality and morbidity. Although many studies have shown positive associations between exposure to ambient PM2.5 and adverse health outcomes, few studies investigate the effects of acute and intense exposures during wildfire events due to lack of hourly exposure measurements. We investigate the feasibility of using aerosol optical depth (AOD), smoke mask, and aerosol detection parameters retrieved by GOES16, a geostationary satellite, to estimate hourly PM2.5 levels in California between October 1 and November 30 2018 at 3 km spatial resolution to capture the Camp Fire wildfire event. We trained a Random Forest Model to fit GOES16 AOD, smoke mask, and aerosol detection parameters, meteorological variables from the High-Resolution Rapid Refresh (HRRR) atmospheric model, and land use data including percent of various vegetation, road distance, and elevation to existing ground measurements from Purple Air sensors and EPA Air Quality System (AQS) monitors. A convolutional layer derived from existing ground measurements for each hour is added to enhance spatial and temporal correlation in the model. There were 255,936 hourly measurements from a total of 921 combined Purple Air sensors and AQS monitors during the study period. The out of bag R2 (RMSE) was 0.88 (8.75 μg/m3). Importance ranking indicated that aside from the convolutional layer, AOD and land use variables including percent herbaceous and forest were the top predictors. Hourly predictions from the model also aligns well with true color imagery from MODIS. Our study showcases the feasibility and utilization of GOES16 products for modeling PM2.5 at finer spatial and temporal resolution. It also allows for construction of historical hourly PM2.5 levels during a significant wildfire event to support air pollution epidemiological studies that investigate the effects of acute and intense exposure to smoke PM2.5.
<< Previous Abstract | Next Abstract >>