Reducing uncertainties in predictions of future air pollution, heat extremes, and associated health risks in Asia with dynamical downscaling and machine learning

Asian countries, particularly China and India, are facing the dual challenge of air pollution and heat extremes, due to intense ongoing economic development. Both hazards pose great threats to human health and ecosystem. Therefore, prediction of future air pollution and heat extremes has been a major concern of both science and regulatory communities. Predicting future air pollution and heat extremes for a certain region is particularly challenging, due to coarse resolution of climate models, uncertain emission estimates, missing or poorly parameterized physical and chemical processes, etc. Here we aim to combine the advantages of dynamical downscaling and machine learning to reduce uncertainties in predictions of future air pollution, heat extremes and associated health risks.

Compared to North America and Europe, predictions of future air pollution and heat extremes in Asia are commonly more uncertain due to rapidly changing emissions in Asia and assumptions in climate-chemistry models that developed mainly in the US and Europe. In this project, we propose to combine our expertise in numerical modeling of regional climate and chemistry and machine learning, and to use historical observations
of atmospheric composition and meteorological variables to improve predictions of future air pollution and heat extremes in Asia. We will apply a sophisticated coupled meteorology-chemistry model to obtain a refined prediction of PM2.5 concentrations, O3 concentrations and occurrence of heat extremes for both present (2011-2020) and future (2091-2100) decades at higher resolution. With the ability to learn without being explicitly programed, a bias-correction model based on machine learning algorithms will then be developed to learn the systematic bias of the predictions using dynamical downscaling, and the developed bias-correction model will be further used to obtain bias-corrected future predictions. We will also assess present and future health risks of exposure to air pollution and heat extremes using improved predictions. This project will be a good demonstration of the use of complementary advantages of numerical modeling and machine learning. The results will be potentially useful for air quality or weather/climate forecasting in Asia, and are likely to benefit public health or ecology communities as well.