Daily 1 km satellite-based PM2.5 exposure estimates in Bangladesh, 2020-2025: spatiotemporal patterns and implications for environmental health.

Bangladesh experiences some of the highest ambient PM2.5 concentrations globally and remains one of the most monitoring-sparse high-burden regions. To address this gap, we developed daily 1 km PM2.5 exposure estimates for Bangladesh from January 1, 2020, to January 31, 2025 by integrating satellite observations, reanalysis products, and machine learning. We first reconstructed missing MAIAC aerosol optical depth retrievals, which exceed 60% of potential retrievals across the study period, using LightGBM to generate a spatially complete AOD dataset. We then combined the gap-filled AOD with meteorological, atmospheric composition, and geospatial predictors within an attention-based TabNet model to estimate ground-level PM2.5 concentrations. Model performance is high on independent test data (R2 = 0.95, RMSE = 17.41 μg/m3, MAE = 9.38 μg/m3), and it tracks observed pollution extremes more closely than existing global products. The predicted Bangladesh national annual mean PM2.5 concentration ranged from 68 to 75 μg/m3 from January 1, 2020, to January 31, with seasonal peaks in December-February exceeding 130 μg/m3 in the most polluted corridor. To demonstrate the relevance of the exposure dataset for population-level research, we linked 2022 annual mean PM2.5 estimates to the 2022 Bangladesh Demographic and Health Survey. In an exploratory division-level ecological analysis, each 1 μg/m3 increase in annual mean PM2.5 was associated with a 0.08 percentage-point higher prevalence of acute respiratory infections in children under five (95% CI: 0.003-0.157; p = 0.043) and a 0.527 percentage-point higher prevalence of hypertension in adult men (95% CI: 0.26-0.79; p = 0.003).