Machine Learning Approach Improves Bioaerosol Detection in the Atmosphere

The Science                                  

A primary biological aerosol is a type of bioaerosol that is directly emitted into the air from biological sources such as plants, animals, fungi, and humans. These aerosol particles are known to significantly affect the Earth’s radiation budget and human health. Biological particles are ubiquitous in the atmosphere. However, accurate detection of bioaerosols using standard analytical techniques is inherently difficult due to their remarkable similarity to other atmospheric carbonaceous (carbon-rich) particles. This study reveals the abundance of bioaerosols at different altitudes, as well as different geographical locations, by combining single particle measurements with advanced supervised machine learning techniques.

The Impact

Bioaerosols play a critical role in atmospheric processes, acting as seeds for cloud droplets and ice crystals. While only a small fraction of atmospheric particles can initiate ice formation, usually at relatively cold temperatures, bioaerosols are particularly efficient at forming ice at relatively warm temperatures. Improving the quantification of these particles is essential for understanding their impact on cloud formation and the environment. This study presents a novel tool to enhance the measurement and characterization of bioaerosols in the atmosphere.

Summary

Accurate detection of bioaerosols in the atmosphere remains challenging. Traditional methods are known to underestimate the fraction of bioaerosols in the atmosphere, requiring arduous manual verification of high-resolution images of individual particles. A new multi-institutional study led by a scientist at EMSL, the Environmental Molecular Sciences Laboratory, a Department of Energy Office of Science user facility at Pacific Northwest National Laboratory, developed a supervised machine learning (ML) model that accounts for chemical composition and morphological features from individual particles. The supervised ML (SML) model, called XGBoost, has been validated by both standard and manually annotated field-collected particles. The team applied the SML method to understand the vertical distribution of biological particles over the Atmospheric Radiation Measurement (ARM) User Facility’s Southern Great Plains atmospheric observatory in Oklahoma and found that traditional methods significantly underestimate the fraction of biological particles compared to the use of SML. The team also used SML to investigate the abundance of biological particles in the Amazon rainforest, where a high fraction of biological particles is expected. This second case study showed that SML models outperform and provide improved detection of bioaerosols in the atmosphere. This study has significant implications for assessing the impact of bioaerosols on the environment, the atmosphere, and human health.

Contacts

Ashfiqur Rahman | EMSL | ashfiqur.rahman@pnnl.gov

Aivett  Bilbao | EMSL | aivett.bilbao@pnnl.gov

Swarup China | EMSL | swarup.china@pnnl.gov

Funding

The study was supported by EMSL’s intramural science and technology funding. EMSL is a Department of Energy (DOE) Office of Science user facility sponsored by the Biological and Environmental Research program. This research also used resources from DOE’s Atmospheric Radiation Measurement User Facility.

Publications

A. Rahman, et al. “Enhanced Detection of Primary Biological Aerosol Particles Using Machine Learning and Single-Particle Measurement.” ACS ES&T Engineering. 4, 10, 2393–2402 (2024). [DOI: 10.1021/acsestengg.4c00262]

Related links

This research was featured on the cover of the American Chemical Society ES&T Engineering journal.