Understanding Boundary Layer Transport of Carbon-Based Pollutants During the TRACER Campaign Using Unsupervised Machine Learning
Elizabeth is a second-year master’s student at the University of Oklahoma studying carbon-based pollutants. After earning her BS in chemical engineering at Virginia Tech in 2021, meteorological research was the natural next step to further her understanding of urban and industrial emissions and their impacts on human health and the climate. Elizabeth’s current research focuses on greenhouse gas and related data collected with EM27/SUN spectrometers which she deployed during the Department of Energy Atmospheric Radiation Measurement User Facility’s TRACER (Tracking Aerosol Convection Interactions Experiment) campaign in Houston, Texas.
Abstract
Analysis of urban air quality, atmospheric chemistry, and climate change is incomplete without an understanding of carbon-based species in the Earth’s atmosphere and their dynamics. Greenhouse gases methane and carbon dioxide, along with carbon monoxide, all contribute to atmospheric heating and are commonly produced by anthropogenic sources as well as natural ones. The GeoCarb-TRACER Campaign was designed to observe these trace gases and their dynamics in the fourth most populated city in the United States as a part of the greater TRacking Aerosol Convection interactions ExpeRiment (TRACER) Campaign, organized by the U.S. Department of Energy’s (DOE) Atmospheric Radiation Measurement (ARM) user facility.
During this campaign, Bruker EM27/SUN spectrometers were deployed at various urban and background sites around the city of Houston, Texas from late May through mid-September of 2022 during seven one-to-two-week intense operational periods (IOPs). The portable Fourier transform spectrometers employed during the campaign capture high-resolution (0.5 cm-1) spectra in the near- and shortwave-infrared wavelength range. These spectra are commonly analyzed to retrieve column-averaged abundances of species present in the atmosphere, primarily carbon dioxide, carbon monoxide, methane, and water vapor. A maximum of three EM27/SUNs were deployed simultaneously alongside instruments gathering boundary layer, aerosol, and near-surface meteorological information.
Using dimension reduction and unsupervised machine learning techniques, the data collected over the course of the campaign can be simultaneously analyzed, avoiding the need to rely on case studies to decipher the near-surface transport of carbon-based pollutants in the dynamically complex atmosphere over Houston. In this study, Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) is used to represent the dataset in a low-dimensional space for qualitative analysis of the relationship between the state vector space and concentrations of interest. Additionally, Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is used in conjunction with UMAP to quantify the qualitative relationships, laying the groundwork for statistical analysis.
Presentation
