Understanding the spatiotemporal dynamics of PM 2.5 exposure among vulnerable populations is essential for promoting environmental equity and guiding effective public health interventions. In this study, we developed a comprehensive framework for estimating daily PM 2.5 concentrations and assessing age-specific exposure across China from 2009 to 2020. We proposed a novel deep learning architecture, PMMamba, which integrated convolutional layers with the Mamba block to effectively capture both local spatial features and long-range temporal dependencies. Trained on multi-source satellite, meteorological, and land surface datasets, PMMamba achieved high estimation accuracy, with an overall R 2 of 0.93, RMSE of 8.12 μg/m 3 , and MAPE of 23.79%. Based on the 1-km gridded PM 2.5 estimates, we constructed a Spatialized Population Exposure Index (SPEI) that incorporated pollutant concentration and vulnerable population density. We further performed hotspot analysis, pixel-level trend estimation, and developed a dual-index system—the Regional Inequality Index (RII) and Regional Burden Contribution (RBC)—to evaluate spatial disparities. Finally, we classified provinces into four distinct exposure typologies based on their joint RII and RBC values, providing a multidimensional view of environmental health equity. The results revealed a persistent east-west gradient in exposure burden, with particularly high levels in the North China Plain and Yangtze River Delta. Despite national improvements, substantial spatial inequalities remained. This study provides an integrated and scalable approach to PM 2.5 exposure estimation and offers critical insights for targeted air quality management and equity-oriented environmental policy.
Liu et al. (Wed,) studied this question.