Spatiotemporal Prediction of PM_2.5 Pollution in Dhaka District Using Ensemble Machine Learning and Satellite-Derived Indices (2019–2025)

Dhaka districts suffer from severe air pollution, with PM₂ concentrations averaging 98 µg/m³. ₅ annually, nearly 20 times the WHO recommended guideline of 5 µg/m³. However, only three PM₂. ₅ sensors covered 1,463 km² of the district. There was no data on PM₂. ₅ in a remaining neighbourhood and nowhere to restrain emissions within Dhaka's megacity of 21.8 million people. The study from 2019 to 2025 uses Google Earth Engine, GIS, and machine learning to understand where and when PM₂.₅ is present. It also maps air quality risks across the Dhaka District. The study examines NO₂, CO, SO₂, and CH₄ data from Sentinel-5P/TROPOMI. It also uses the MODIS Aerosol Index and Land Use and Land Cover data. All this data was put together at 119 points in ArcGIS. The Google Earth Engine, GIS and machine learning are used to understand PM₂.₅ distribution. Random Forest outperformed multiple linear regression (MLR) and Extreme Gradient Boosting, with R²=0.8341 vs 0.6449 (p<0.001), RMSE=16.94 µg/m³, and MAE=11.38 µg/m³, confirming that the collective model captures the nonlinear relationship between the predictor variables and PM₂. ₅ at different spatial and temporal resolutions that the MLR model fails to. The analysis showed that NO₂ accounted for 66.2% of the PM₂. ₅ variability, highly suggesting that emissions from transportation are the primary pollution contributor. Secondly, predicted winter PM₂. ₅ concentration reaches 156.70 µg/m³ (31.3 times the WHO guidelines), a peak likely exacerbated by a meteorological thermal reversal. Brick kiln industries, untreated pollutant emissions from industries, and deforestation documented a 12.91% reduction. The study shows an increase in CH ₄ plumes in the Matuail area near Jatrabari, Dhaka. The findings of our study aim to show real-time air pollution in Dhaka and the health hazards caused by uncontrolled air pollution. Ultimately, this research provides the AQI risk estimates for Dhaka, offering spatially explicit evidence to support precision emission control and urban afforestation initiatives.