Machine Learning Integration of PTAAM and SP2 Measurements for Enhanced Aerosol Absorption Characterization

Quantifying the light absorption of atmospheric aerosols remains one of the more critical challenges in climate science. Black carbon (BC) and mineral dust (MD) dominate aerosol light-absorption globally, yet their mass absorption cross-sections (MAC)—the fundamental measure linking particle mass to light absorption—vary by orders of magnitude across the literature. This inconsistency stems partly from measurement artefacts inherent to existing techniques. Filter-based methods suffer from systematic errors, photoacoustic approaches introduce thermal biases, and single-particle instruments like the SP2 (Single Particle Soot Photometer) require assumptions about particle morphology that may not hold in the real ambient environments.

This project proposes a hybrid strategy that integrates two complementary measurement platforms with machine learning to address these limitations. Photo-Thermal Aerosol Absorption Monitor (PTAAM) offers high sensitivity while remaining insensitive to scattering effects, whereas the SP2 provides detailed microphysical information about individual particles. The methodological novelty lies not merely in combining these tools, but in developing advanced algorithms—particularly graph neural networks (GNNs)—to extract physically meaningful patterns from their joint data streams.

The work encompasses three interconnected objectives: first, calibrating the SP2 for dust and iron oxide detection through rigorous laboratory work with size-selected aerosols; second, establishing size- and wavelength-resolved absorption spectra using a newly developed PTAAM system; third, constructing machine learning models that fuse these measurements to produce more reliable optical property estimates. Validation occurs through both controlled laboratory experiments and field campaigns in contrasting environments.

By reducing uncertainties in aerosol light-absorption measurements, this study promises to improve climate model predictions and remote sensing retrievals—bridging fundamental aerosol physics with practical applications in understanding aerosol-radiation interactions.