Direct Prediction of PM 2.5 from Raw Satellite Data – A Prospectus

Accurate estimation of ground-level particulate matter (PM₂.₅) concentrations is crucial for effective air quality management and public health protection. While ground-based monitoring networks, such as those operated by the EPA, provide precise PM₂.₅ measurements, their limited geographic coverage poses significant challenges. As a result, satellite-based estimations have become critical. Traditionally, satellite-based PM₂.₅ estimation relies on Aerosol Optical Depth (AOD), which is derived from top-of-the-atmosphere reflectance (TOAR) using algorithms such as Dark Target and Deep Blue. This process also produces surface reflectance (SR), a variable that ideally should not correlate with aerosol concentrations. Our first objective is to investigate whether AOD is properly estimated from TOAR by examining the relationship between SR and ground-based PM₂.₅ — a novel approach in this field. If a strong relationship exists, it would suggest inaccuracies in the current AOD estimation process. Using daily MODIS SR data (Bands 1–7) and EPA ground-based PM₂.₅ data across the United States for 2023, we applied Dynamic Panel Regression models incorporating lagged PM₂.₅ variables. The model achieved strong predictive performance (overall R-squared 0.78), challenging the existing paradigm and motivating our first recommendation: to explore the direct prediction of PM₂.₅ from TOAR without relying on AOD. Our second objective was to examine the seasonality of PM₂.₅ patterns across the United States. Spatio-temporal clustering analysis using 2020–2024 data revealed that 3 out of 5 identified clusters exhibited seasonal behavior. Based on this, we recommend specific cluster-based de-seasonalization to improve prediction accuracy. Our third objective was to assess heterogeneity among ground-based measuring instruments. Data from multiple instruments at the same stations revealed random variations in PM₂.₅ measurements, attributable to calibration and measurement errors. Thus, we recommend standardizing measurements across different instrument types. Finally, our fourth objective was to review the influence of meteorological variables and extreme events on PM₂.₅ levels. Empirical analysis confirmed that commonly used meteorological variables remain significant predictors, supporting their continued use. Overall, this study presents a prospectus for the direct prediction of PM₂.₅ from raw satellite data (TOAR) combined with meteorological factors, incorporating cluster-based de-seasonalization and standardizing ground measurements for improved accuracy.