May 13, 2022
Article

An Automated, Faster Way to Analyze Aerosols

Machine learning makes analyzing aerosol data faster

Aerosols from oil refineries

Researchers developed a faster way to analyze aerosols, like those from oil refineries shown here.

(Photo by Chris LeBoutillier | Pexels.com)

Within the atmosphere, tiny, airborne, carbon-containing microscopic particles called organic aerosols (OA) remain suspended. They are directly emitted from a variety of sources ranging from cooking oils and meats to burning wood and fossil fuels. They can also be formed by the condensation of oxidized organic gases, making secondary organic aerosol particles. Each OA particle consists of thousands of organic compounds that dynamically change due to their chemical processing in the atmosphere. These aerosols can affect human health, global climate, and air quality. Therefore, it is important to understand their key sources.

Researchers from the Pacific Northwest National Laboratory (PNNL) and collaborators from the University of California, Davis, Peking University, the Georgia Institute of Technology, Texas A&M University, and the City University of Hong Kong, established a new way to make analyzing the chemical makeup and sources of aerosols even easier.

The chemical makeup of aerosols and their source signatures can be measured in a variety of ways. One method involves the use of a mass spectrometer—an instrument that measures their chemical composition online and in real time. As part of the Atmospheric Radiation Measurement (ARM) user facility for the Department of Energy (DOE) Office of Science, a research aircraft is specially equipped with instruments to collect such data. However, the analyses of aerosol mass spectrometry data to determine the sources of OA is often time-consuming, challenging, and relies a lot on user judgement. Research aircraft flying over different regions encounter different OA types, which pose challenges for subsequent analyses.

“We wanted to develop an automated approach to rapidly analyze the aerosol mass spectrometer data in real time bypassing the need for subjective user judgement, as samples are being collected from the atmosphere,” said Earth Scientist Manish Shrivastava, corresponding author of the publication in ACS Earth and Space Chemistry. Shrivastava and his colleagues applied machine learning techniques to mass spectrometry datasets in ways that have never been done before.

They developed a two-step machine learning approach that can predict key OA sources and their corresponding fractional mass abundances. They applied this approach to analyze aircraft data captured during the Holistic Interactions of Shallow Clouds, Aerosols, and Land-Ecosystems (HI-SCALE) campaign in 2016.

“This approach can be used in a variety of past and upcoming field measurements since it can yield results in seconds and could analyze even a single aerosol mass spectrum, which cannot be done in traditional analysis methods,” said Shrivastava.

Additional PNNL authors are Paritosh Pande, John E. Shilling, Alla Zelenyuk, Quazi Z. Rasool, Yuwei Zhang, and Ying Liu.

This research was supported by the Seed Laboratory Directed Research and Development Program of the Earth and Biological Sciences Directorate at PNNL, and by Shrivastava's DOE Early Career award and DOE's Atmospheric System Research, both of the Office of Science Biological and Environmental Research program. Additional funds were provided by the National Institutes of Health and the National Science Foundation. Computational resources were provided by the Environmental Molecular Sciences Laboratory, a DOE Office of Science user facility located at PNNL. Support for data collection onboard the aircraft was provided by ARM.