Abstract View
Characterization and Quantification of Novel Ambient Organic Aerosol Compounds using Machine Learning and the UCB-GLOBES Mass Spectral Database
EMILY FRANKLIN, Lindsay Yee, Robert Weber, Paul Grigas, Allen Goldstein, University of California, Berkeley
Abstract Number: 692
Working Group: Aerosol Standards
Abstract
The chemical composition of ambient organic aerosols plays a critical role in driving their climate relevant properties and holds important clues to their sources and the formation mechanisms of secondary aerosol material. In most environments, this composition remains incompletely characterized. Mass spectral analysis of organic aerosol material collected during several recent ambient sampling campaigns consistently shows that >80% of individually catalogued compounds lack definitive mass spectral matches in the literature or NIST/NIH/EPA mass spectral databases. This creates significant challenges in utilizing the full analytical capabilities of techniques which separate and generate mass spectra for complex environmental samples. In particular, TD-GCxGC-MS achieves advanced separation of complex organic samples with two GC columns in sequence, with each individual species characterized by its 1st (volatility) and 2nd (polarity) dimension retention times and 70 ev EI mass spectrum. Typically, high hundreds to low thousands of individual organics are isolated from any given sampling medium, yet the majority cannot be identified or quantified by traditional means. In this work, we develop the use of machine learning techniques and further update the Goldstein Library of Biogenic and Environmental Spectra (UCB-GLOBES) to quantify and characterize novel organic material. A random forest model trained and tested on a known 135 component custom representative external standard predicts the quantification factors of novel environmental organics based on position in volatility-polarity space and mass spectrum, enabling reproducible, efficient, and optimized quantification of novel environmental species. Intercomparison between mass spectra, both known and novel, identified across a wide range of ambient conditions as well as from single precursor oxidation experiments through UCB-GLOBES database enables the identification of new tracers of specific atmospherically relevant oxidation mechanisms. Finally, clustering of novel environmental spectra based on chemical similarity provides insight into the true chemical complexity of organic aerosol.