Abstract The development of smoke fine particulate matter (PM 2.5 ) exposure surfaces for estimating air pollution trends and associated health effects has advanced considerably. Currently available smoke exposure products rely on various data sources and modeling techniques, as there is no gold standard method for modeling wildfire smoke PM 2.5 . This study compares multiple daily smoke PM 2.5 data sets developed using diverse methodologies spanning 2008–2018. Incorporating metrics for short‐ and long‐term exposure, we compare four data sets at the census tract level in California: one using the U.S. Environmental Protection Agency’s chemical transport model (CTM), the Community Multiscale Air Quality Modeling System (CMAQ); two using statistical methods, also referred to as machine learning (ML) techniques; and one combining these approaches to develop an ML‐calibrated CTM‐based exposure surface. Our analysis highlights differences between the data sets in terms of long‐term exposure metrics, with the CTM data set estimating the highest concentrations overall, and considerable differences between estimates produced by the two ML models. An analysis of six case studies of large fires across the state finds that even data sets with similar inputs and methods produced estimates that varied several‐fold, with additional differences by region and over time. Our findings have important implications for quantifying smoke PM 2.5 exposures for use in population health impact studies, which rely on exposure estimates to accurately estimate health burden from pollution exposure.
Connolly et al. (Mon,) studied this question.