Particulate matter smaller than 2.5 microns (PM2.5) is one of the main pollutants that has considerable detrimental effects on human health. Estimating its concentration levels with ground monitors is inefficient for several reasons. In this study, we build a digital twin (DT) of an atmospheric environment by fusing remote sensing and observational data. Integral part of DT pipeline is a presence of feedback that can influence future input data. Estimated values of PM2.5 obtained from an ensemble of Random Forest and Gradient Boosting are used to provide recommendations for decreasing the agglomeration levels. A simple optimization problem is formulated for computing the recommendations. Possible action policies, such as cloud seeding, scheduling of air filtering, and SMS notifications have been identified. The PM2.5 estimation part of the proposed DT pipeline has achieved RMSE and R2 of 38.94 and 0.728 (95%, CI 0.717-0.740). In addition, we investigate different approaches for quantitatively examining the contribution of each independent variable. Author
air pollution, Atmospheric modeling, Computational modeling, Data models, Deep learning, digital twins, Digital twins, health risk, Pipelines, PM2.5, Predictive models, satellite data
K. Abutalip, A. Al-Lahham and A. El Saddik, "Digital Twin of Atmospheric Environment: Sensory Data Fusion for High-Resolution PM2.5 Estimation and Action Policies Recommendation," in IEEE Access, vol. 11, pp. 14448-14457, 2023, doi: 10.1109/ACCESS.2023.3236414.