A machine learning approach for predicting atmospheric aerosol size distributions

摘要

An accurate model and parameterization of aerosol concentration is needed to predict the performance of electro-optical imaging systems. Current models have been shown to vary widely in their ability to accurately predict aerosol size distributions and subsequent scattering properties of the atmosphere. One of the more prevalent methods for modeling particle size spectra consists of fitting a modified gamma function to measurement data, however this limits the distribution to a single mode. Machine learning models have been shown to predict complex multimodal aerosol particle size spectra. Here we establish an empirical model for predicting aerosol size spectra using machine learning techniques. This is accomplished through measurements of aerosols size distributions over the course of eight months. The machine learning models are shown to extend the functionality of Advanced Navy Aerosol Model (ANAM), developed to model the size distribution of aerosols in the maritime environment.