Application of Clustering Algorithms for Improved Source Apportionment with Photothermal Infrared Spectra from Individual Particles Collected in New York City
ABBYGAIL AYALA, Yao Xiao, Eduardo Ochoa Rivera, Kayleigh Reilly, Emily Costa, Xu He, Corin Tyler, Drew Gentner, Rachel O'Brien, Ambuj Tewari, Andrew P. Ault, University of Michigan
Abstract Number: 368
Working Group: Advancing Aerosol Science through Data Analysis Tools
Abstract
Characterizing the physiochemical properties of individual aerosol particles is critical to establish sources, improve climate models, and understand health impacts from air quality. Microspectroscopic techniques provide valuable single-particle characterization, but analyzing thousands of spectra collected from a campaign requires extensive time and human effort, which would benefit from improvements in automated processing. This work uses an unsupervised clustering method to group infrared (IR) spectra collected from field samples into classes that can be assigned to specific sources and forms of atmospheric aging. Particles collected during the New York City metropolitan Measurements of Emissions and TransformationS (NYC-METS) campaign were analyzed with optical photothermal infrared (O-PTIR) spectroscopy. This instrument uses a pump-probe system, with a continuous wave visible laser (532 nm) and a tunable IR laser (948-1860 and 2698-3002 cm-1) to obtain photothermal IR (PTIR) spectra of individual aerosol particles, with a theoretical spatial resolution of ~0.5 µm. Unlike Raman spectroscopy, O-PTIR does not have issues with fluorescence overwhelming the signal, which can commonly occur with environmental samples. To analyze the thousands of spectra collected from this campaign, k-means clustering was used to group similar spectra together, reducing both analysis time and the potential for human bias. We were then able to distinguish particles containing primarily nitrate, ammonium sulfate, mineral or road dust, and organic species. Collection of PTIR spectra on field samples at this scale has not been done to date prior to this study. By using this clustering process, we were able to obtain an insight into aerosol composition from the NYC-METS field campaign, distinguishing trends as a function of sample location, time of day, and size of particles using vibrational spectroscopy. These findings will enable an improved understanding of source apportionment and atmospheric evolution of particulate matter in the New York metropolitan area.