Aerosol classification using machine learning algorithms

Abstract

Aerosols are particles that are omnipresent in the atmosphere. They vary in size, shape and composition. They can be naturally occurring or might be produced artificially. However, proper classification and characterization of aerosols have been still in progress and this creates uncertainty in climatological studies. In this paper, an aerosol classification scheme has been presented based on the measurements done using a CIMEL sunphotometer in Thessaloniki, Greece from 1998 to 2017. The study has been mainly upheld by the direct measurements of Single Scattering Albedo (SSA) at 440nm and Fine Mode Fraction (FMF) at 500nm. These parameters have been used to establish testing and training datasets. Machine learning algorithms have been used to validate the classified data. Various performance metrics have been evaluated. Also, the best-fit algorithm for classifying aerosol data has been found out.