Correcting for Biases in Filter-Based Aerosol Light Absorption Measurements at the ARM Southern Great Plains Site

JOSHIN KUMAR, Theo Paik, Nishit Shetty, Patrick Sheridan, Manvendra Dubey, Allison Aiken, Rajan K. Chakrabarty, Washington University in St. Louis

     Abstract Number: 87
     Working Group: Remote and Regional Atmospheric Aerosol

Abstract
Measurement of absorption of solar radiation by aerosols is vital for assessing direct radiative forcing, which affects local and global climate. Low-cost and simple-to-operate filter-based instruments, such as Particle Soot Absorption Photometer (PSAP) that collect aerosols on a filter and measure light attenuation through the filter are widely used to infer aerosol light absorption. However, filter-based absorption measurements are subject to unquantifiable artifacts associated with the presence of the filter medium and the complex interactions between the filter fibers and accumulated aerosols. Various correction algorithms (Bond et al., 1999; Virkkula et al., 2005; Li et al., 2020) have been introduced to correct for the filter-based absorption coefficient measurements toward predicting the particle-phase absorption coefficient (Babs). Since previously developed correction algorithms have a fixed analytical form, fundamentally, they are unable to predict particle-phase absorption coefficients with a high degree of accuracy universally: different corrections for rural and urban sites across the world.

In this study, we have analyzed three months of high-resolution ambient data collected in parallel using a PSAP and 3-wavelength photoacoustic spectrometer; both instruments were operated at the Department of Energy’s Southern Great Plains user facility in Oklahoma. We implemented the following algorithms to predict particle-phase Babs values from PSAP data and estimate their accuracy – (1) the Virkkula (2010) correction algorithm, (2) the Revised Virkkula algorithm with updated coefficients, and (3) a Random Forest Regression (RFR) machine learning algorithm. The RFR algorithm outperformed predictions by both the Revised Virkkula and Virkkula (2010) algorithms. The wavelength averaged Root Mean Square Error (RMSE) values for predicting Babs using RFR, Revised Virkkula, and Virkkula (2010) algorithms were 0.37, 0.67, and 3.18 Mm-1, respectively.

To further test the potential of the proposed machine learning model, we trained and tested the RFR algorithm on a dataset of laboratory-generated combustion aerosols. The RFR model used the size distribution, uncorrected Tricolor Absorption Photometer based Babs, and Nephelometer Bscat as input variables and predicted particle-phase Babs values within 5% of the reference Babs.

Presently we are working on the application of the RFR algorithm to FIREX-AQ data from NASA DC-8 aircrafts to correct the artifacts of the filter-based absorption measurement instruments using machine learning. Preliminary analysis shows that RFR is capable of accommodating inferences from various instruments and is more accurate in predicting Babs as compared to the traditional equation fitting-based correction algorithms.