A Space/Time Data Fusion Method for Accurately Estimating Wildfire Smoke Concentrations During the October 2017 California Fires to Inform Population-Level Exposure

Abstract

Background. Exposure to wildfire smoke causes a range of adverse health outcomes, suggesting the importance of accurately estimating wildfire smoke concentrations. While chemical transport models (CTMs) and spatial interpolation of observations are often used to assess smoke exposure, geostatistical methods can combine surface observations with modeled and satellite-derived concentrations to produce more accurate exposure estimates.Methods. Here we estimate ground-level PM2.5 concentrations during the October 2017 California wildfires, using the Constant Air Quality Model Performance (CAMP) Method and the Bayesian Maximum Entropy (BME) Framework to bias-correct and fuse together three PM2.5 datasets: concentrations from permanent and temporary monitoring stations, a CTM, and satellite observations. Four different BME space/time (s/t) kriging and data fusion methods using these three datasets were evaluated for accuracy.Results. All four BME methods produced more accurate estimates of ground-level PM2.5 than the standalone CTM and satellite products, with the addition of temporary monitoring station data further improving accuracy. While BME s/t kriging on observations performed best near monitoring stations, the BME data fusion of observations with the CAMP-corrected CTM provided the best overall estimate, especially in smoke-impacted regions. Using these smoke concentrations, we estimate more than 60,000 people were exposed to very unhealthy air, PM2.5 concentrations greater than 150.5 µg/m3, during the 2017 wildfires.Conclusions. We show that the BME framework, used in combination with the CAMP correction method, can be used to accurately estimate ground-level PM2.5 concentrations during a wildfire event. Our results emphasize the importance of combining multiple data sources to characterize population-level exposure during extreme air pollution events.