Department Seminar: Chi Li

Title: Advances in characterizing air pollution from modeling, observations, and machine learning

Abstract: Air pollution is the greatest environmental risk to human health. Observations have witnessed rapid responses of air pollution to emissions worldwide, meanwhile large spatial gaps exist. Specific pollutant exhibits unique variability and sensitivity to emissions, as characterized by their driving sources and sinks in models. My research interfaces observations and modeling to advance the quantification and understanding of air pollution, with the goal of informing policy. For the short-lived nitrogen oxides (NO x ), I examine multi-resolution air quality modeling to reveal the resolution-dependent NO x simulation biases due to non-linear chemistry. I elucidate how these biases depend on specific chemical environments and diurnally varying driving mechanisms. For fine particulate matter (PM 2.5 ) with variable relationship with aerosol optical depth (AOD), I show how leveraging information from satellite AOD, air quality modeling, and ground- based data can effectively fill observational gaps toward high-resolution and seamless PM 2.5 estimates. From such estimates I report timely the recent reversal of trend in global PM 2.5 exposure. For ozone with strong dependence on meteorology and non- linear responses to precursors, I use machine learning to isolate emission-only ozone trends from short-term ground-based observations in China. I use the relationship between odd oxygen and NO x to interpret the emission drivers of the local de-weathered ozone trends, with implications on previous and future policies. Going forward, I will carry on with multi-scale modeling, space-based observations, and machine learning to characterize and interpret the variability of prevailing air pollutants as well as emerging climate-sensitive composition.