Estimation of aerosol acidity at a suburban site of Nanjing using machine learning method

Abstract

Aerosol acidity is found to exert negative effects on ecosystem diversity and architectural appearance. Current analytical technology is unable to measure in-situ aerosol acidity (i.e., pH value) of ambient fine particle due to the absence of appropriate pH electrodes. Thermodynamic modeling methods including ISORROPIA II and Extended Aerosol Inorganics Model Version IV (E-AIM V) are mostly used in the estimation of in-situ aerosol acidity with the inputs of water soluble ions worldwide. This study proposes a flexible method with the aid of multilayer perceptron (MLP) neural network analysis to estimate in-situ aerosol acidity of ambient fine particle (< 2.5 μm in aerodynamic diameter or PM_2.5) with the inputs of water soluble ions (i.e., Cl⁻, NO₃⁻, SO₄²⁻, Na⁺, NH₄⁺, K⁺, Mg²⁺, Ca²⁺), gaseous air pollutants (i.e., CO, NO₂, SO₂) and meteorological parameters (i.e., humidity and temperature). The dataset consists of ambient fine particles collected across four individual sampling periods in the autumn and winter of 2019 and 2020 at a suburban site of Nanjing. The pH values of ambient fine particle were found to be ranging from 2.0 to 4.0 estimated by E-AIM model. Levels of pH estimated by MLP neural network analysis agreed well with pH values estimated by E-AIM model with R² value of 0.98.