High Resolution Air Quality Estimation and Forecasting using Advanced Deep Learning Algorithms

The impact of near-surface aerosol is a pressing current issue for human health. The Institute for Health Metrics and Evaluation's (HME) Global Burden of Disease (GBD) report of 2019 ranks outdoor fine particulate matter (PM_2.5) as the 5th-leading risk factor for early death. Therefore, early detection and better air-quality estimation and forecast might help in improving mitigation strategies and dissemination of human health advisories. The main challenge in developing mitigation strategies is the sparseness of the monitoring stations, gaps in satellite observations due to clouds, low-quality retrievals, and the accuracy and computational requirements of fine-resolution numerical models. Additionally, fine spatial and temporal datasets are required for assessing and forecasting air quality due to specific events such as wildfires, dust storms, volcanic eruptions, or lockdowns due to a pandemic. In this study, we will present results of using advance machine learning and deep learning algorithms along with surface, satellite and global model outputs to i) develop a global reanalysis of PM_2.5 record for 25 years; ii) improving forecasted air quality at surface; and iii) a downscaling model to downscale the estimated PM_2.5 at higher spatial resolution. We will discuss the estimation accuracies, application to regional haze transport and examples of implementation at stakeholder agencies.