Accuracy assessment on eight public PM2.5 concentration datasets across China

Economic development has historically led to environmental challenges, notably in China where fine particulate matter with an aerodynamic diameter no greater than 2.5 μm (PM_2.5), has significantly influenced human health and social issues. However, the scarcity and uneven distribution of ground-based PM_2.5 observation sites hinder studies about air pollution impacts at regional and national scales. Although PM_2.5 datasets based on remote sensing retrieval algorithms have provided long-term and high-resolution gridded near surface PM_2.5 concentration data recently, comparisons on accuracy between datasets were not conducted by previous studies. This study evaluated eight publicly accessible PM_2.5 datasets (i.e., CHAP, GHAP, GWRPM25, HQQPM25, LGHAP v1, LGHAP v2, MuAP, and TAP) across China using independent records at 1020 monitoring sites from 2017 to 2022 at monthly and annual granularities. Mean Absolute Errors (MAEs) showed a seasonal trend, with higher errors in winter and lower in summer. Datasets exhibited a bias towards overestimation or underestimation based on concentration levels. CHAP, GWRPM25, and HQQPM25 had better estimation control. Additionally, the incorporation of spatiotemporal features into original machine learning based algorithms was likely credited to the outperformance compared to conventional PM_2.5 simulation methods. Overall, this study contributed to comprehensive references for PM_2.5 concentration dataset users and potential explanations to the variations within and among datasets.