PNNL

Abstract

Accurately identifying primary biological aerosol particles (PBAPs) using analytical techniques poses inherent challenges due to their resemblance to other atmospheric carbonaceous particles. We present a study on an enhanced method for detecting PBAP by combining single particle measurement with advanced supervised machine learning (SML) techniques. We analyzed ambient particles from a variety of environments and lab-generated standards, focusing on chemical composition for traditional classifications and incorporating morphological features into the SML approaches Neural Networks (NN) and XGBoost for improved accuracy. SML methods achieved over 92% precision, in contrast to about 86% precision obtained by traditional approaches. The adaptability of proposed SML model is showcased in comparison to conventional approaches in categorizing PBAPs for blind datasets from different geographical locations. Two field case studies were investigated, over agricultural land, and Amazonia rain forest representing relatively low and high concentrations of PBAPs respectively. Our findings show that traditional methods underestimate biological particles by at least 19% of the total labeled particles compared to SML methods. Precise detection of PBAPs in the atmosphere would significantly improve the prediction of climatic impacts by PBAPs.