Digital Twin of Atmospheric Environment: Sensory Data Fusion for High-Resolution PM2.5 Estimation and Action Policies Recommendation

Abstract

This study addresses the health risks posed by particulate matter smaller than 2.5 microns (PM2.5), which can penetrate deep into the lungs and bloodstream, leading to respiratory and cardiovascular issues. To overcome the inefficiencies and limited coverage of traditional monitoring methods, a digital twin (DT) of the atmospheric environment is developed, integrating remote sensing and observational data with feedback mechanisms for dynamic representation. PM2.5 levels are estimated using an ensemble of Random Forest and Gradient Boosting models, yielding robust predictions with high accuracy (RMSE of 38.94, R² of 0.728 within a 95% confidence interval). The predicted values inform actionable recommendations such as cloud seeding to disperse pollutants, scheduled air filtering to reduce indoor pollution, and public SMS alerts to encourage preventive measures during high pollution days. An optimization problem is formulated to identify the most impactful action policies. Additionally, the study quantitatively assesses variable contributions to PM2.5 levels, demonstrating the potential of digital twins in enhancing environmental monitoring and management while offering strategies to mitigate the adverse health effects of air pollution.