Atmospheric aerosol particles produced by burning of biomass, such as wildfires, can be transported over long distances and reach even the most remote places including the Arctic.
One important tracer commonly used to identify such emissions is levoglucosan, which can be measured with advanced mass spectrometry techniques (FIGAERO-CIMS). When aerosol particles are collected on a filter and are afterwards evaporated, a signal curve, also known as ‘thermogram’ is recorded, where the peak of the curve (Tmax) is commonly used as proxy for aerosol volatility – a property that controls whether a compound remains in the atmosphere as particle or gas. This is relevant as it has consequences on how these compounds interact with radiation, or their ability to form new particles or even cloud droplets.
In a recent study based on year-long observations from the Zeppelin Observatory on Svalbard, Yvette Gramlich and her collaborators investigated what factors, including atmospheric and instrumental conditions, could impact variations in Tmax, in the example of levoglucosan. Machine learning methods revealed that the dominant driver of variability in Tmax is not atmospheric conditions or particle composition, but rather the amount of particles collected on the instrument filter, known as ‘mass loading’. As the authors of the study explain ‘the mass loading on the filter has the largest influence on levoglucosan Tmax, with a tendency of overall lower Tmax values at higher mass loadings.’ The findings highlight the importance of accounting for instrumental conditions when interpreting atmospheric observations from thermograms and demonstrates how machine learning can help improve the interpretation of complex atmospheric measurements.
Journal: Environmental Science & Technology
