Application of Machine Learning Techniques in Predicting Primary and Secondary Organic Aerosols

Description

A key knowledge gap in estimating ambient PM concentration is the prediction of secondary organic aerosols (SOA), which are formed in the atmosphere.  Conventional chemical transport modeling typically underestimates SOA concentrations, primarily due to uncertainties in emission of the chemical precursors and primary organic aerosols (POA) and the mechanisms and kinetics of the formation of SOA from precursors and POA themselves.  Machine learning (ML) techniques can circumvent the above complexity to predict POA and SOA concentrations. This project proposes to use machine learning (ML) techniques to predict the concentrations of POA (traffic and cooking) and SOA components using aerosol mass spectrometry and other air quality and meteorological datasets we have  obtained in Mong Kok (1 year, urban) and HKUST (4 months, rural).   We will train and test the prediction model based on “Random Forest”.  Finally, we will establish the dependence of ambient SOA concentrations and traffic and cooking surrogates to estimate the reduction of ambient SOA from emission controls that reduce traffic and cooking emissions.  The proposed project can potentially improve the predictions of ambient organic aerosol concentrations and provide scientific basis on the importance of reducing vehicle and cooking emissions in mitigating overall PM pollution in HK.