Spatial characteristics and temporal evolution of the relationship between PM2.5 and aerosol optical depth over the eastern USA during 2003–2017

Previous research has proposed use of satellite-retrieved aerosol optical depth (AOD) to generate geospatial assessments of near-surface PM₂.₅ mass concentrations and potentially to provide air quality forecasts. Spatiotemporal variability in PM₂.₅–AOD relationships over the eastern United States of America are analyzed using surface observations, satellite data, reanalysis data, and WRF-Chem simulations. Three primary metrics are analyzed: eta (η, the ratio of PM₂.₅ to AOD), the correlation coefficient (ρ) between daily values of PM₂.₅ to AOD, and hit rate (θ, defined as co-occurrence of high PM₂.₅ and AOD). It is shown that mean daily η exhibits substantial geospatial variability and a pronounced seasonal cycle. η computed for 301 EPA stations ranges from 21 to 155 μg m⁻³ and has a domain-wide median value of 70 μg m⁻³. Larger values of η occur in winter and fall. There is also evidence of diurnal variability in η. Lower values are derived in analyses using AOD from Terra (i.e. the morning overpass) than when AOD from the MODIS instrument onboard Aqua (i.e. the afternoon overpass). The spatial median η from Terra is 70 vs. 93 μg m⁻³ from Aqua. A majority of sites exhibit statistically significant lower values of η, ρ, and θ during 2013–2017 than 2003–2007, indicating a declining association between AOD and PM₂.₅. This has implications for the potential to use remotely-sensed AOD to generate geospatial estimates of near-surface PM₂.₅. The spatial distribution of η across the 301 locations exhibits a negative dependence on planetary boundary layer height and 10 m wind speed and a positive dependence on integrated humidity in lower troposphere, urban fraction, 2 m temperature, and vegetation coverage. Simulations with WRF-Chem indicate model-derived estimates of η, ρ, and θ are highly dependent on the aerosol scheme employed but the model captures some of the spatial variability and the correct dependence of η on meteorological and land use causes of that variability.