Discussion of “Data Fusion for Air Pollution Exposure Monitoring”
View Journal ArticleIntroduction
The A&WMA 2026 Critical Review (CR) (Liu et al. 2026) analyzes the methods, applications, and future directions of data fusion in air pollution exposure assessment. It has long been noted that surface air monitoring, because of the complexity and expense required to establish regulatory sites, cannot provide complete coverage of all areas needed for health effects exposure analyses. Supplemental data from lower-cost air sensors and numerical modeling can enhance data coverage if fused with regulatory-grade monitoring data. The CR explores the benefits and challenges around data fusion as a potential tool for enhancing air pollution exposure modeling and the outputs that are derived from that modeling. It examines the key input data sets available for data fusion, details 5 specific technical approaches for data fusion, and summarizes the strengths and weaknesses of each approach.
Lead author Yang Liu presented the CR at the A&WMA Annual Conference in Austin, Texas, U.S.A., June 24, 2026. Dr. Liu began his superb presentation by noting that previous critical reviews by Pope and Dockery (2006), Hoff and Christopher (2009), and Holloway et al. (2025) had influenced his early career trajectory and made positive contributions to his current work. He quickly summarized key points of the CR: (1) Modern air pollution exposure assessment increasingly requires spatially complete, temporally continuous air pollution concentration fields, but no single data source can provide these properties simultaneously. (2) A key advantage of data fusion is that it integrates complementary information from monitoring networks, satellite retrievals, chemical transport models (CTM), low-cost sensors, and geospatial predictors to construct internally consistent exposure surfaces. (3) The scientific value of a fused product depends not only on predictive accuracy, but also on validation design,uncertainty characterization, transparency, and appropriateness for downstream use. He noted that while ground observations are the empirical anchor of data fusion applications, validation based only on this data can overstate performance in unmonitored areas. Dr. Liu concluded that the best model is the one whose assumptions are most appropriate for the intended use.
Also, at the AWMA conference, three experts invited by the Critical Review Committee commented on the CR: Dr. Bryan Hubbell, Nathan R. Pavlovic, and Dr. Jia Xing. This article summarizes comments from these invited Discussants, all of whom concurred that the CR does an outstanding job presenting and explaining benefits and challenges of data fusion as a viable tool for use in air pollution exposure analyses. Each Discussant brings different perspectives on the topic, thus providing additional thoughts for consideration as well clarifying potential uses and challenges. The Discussant’s comments are their own.
Dr. Hubbell focuses on the uncertainties of fusion due to limitations in available monitoring sites and pollutants, which affect prediction accuracy and bias, particularly for hyper-local areas and rural regions with sparse data. He also emphasizes the importance of dynamic responsiveness to changes in emissions and meteorology. Mr. Pavlovic discusses uses of data fusion for applications not covered in the paper, such as real-time and forecasting applications, offers guidance for selecting fusion approaches, describes challenges associated with data quality and extreme events, and highlights additional frontiers in data fusion applications. Dr. Xing comments on the challenges faced with fusion models based on artificial intelligence (AI), the representativeness of ground monitoring networks, and the resulting uncertainty in model generalizability.
This article concludes with responses from the CR’s coauthors.
Authors
Richard A. Wayland
US Environmental Protection Agency (retired)
Nathan R. Pavlovic
Spheros Environmental
Jia Xing
University of Tennessee, Knoxville
Yang Liu
Emory University
Howard H. Chang
Emory University
Joshua S. Fu
University of Tennessee
Susan S.G. Wierman
Johns Hopkins University