Correcting filter-based aerosol light absorption measurement biases in a coastal urban-industrial region

This study focuses on filter-based aerosol light absorption measurement biases and their correction algorithms in the coastal urban-industrial area of Houston-Galveston region. Known as a major petrochemical hub in the United States, this area is dominated by industrial flaring emissions. The aerosols in this region are mainly composed of organic carbon and sulfates. We observed that conventional filter-based instruments, despite their cost-effectiveness and simplicity of use, overestimate aerosol light absorption by approximately four times in comparison to reference particle phase instruments, such as photoacoustic spectrometers. To mitigate these unquantifiable measurement biases, we applied and compared different analytical correction algorithms including the widely used Bond-Ogren (Citation2010) and Virkkula (Citation2010), as well as a customized Random Forest Regression (RFR) machine learning algorithm. Our analysis revealed that RFR significantly improved correction efficacy, reducing the wavelength-averaged Root Mean Square Error (RMSE) by approximately 50% compared to traditional analytical methods. We performed SHapley Additive exPlanations (SHAP) analysis to identify the key parameters that influence the accuracy of our RFR correction algorithm. We find that at longer visible wavelengths, dark-brown carbon from flaring emissions in the sampling region exacerbates biases in filter-based measurements. This study underscores the importance of employing advanced correction algorithms for correcting filter-based aerosol light absorption measurements, especially in complex urban settings influenced by industrial emissions.