Neural Network for Soving the General Dynamics Equation for Aerosols

JIANI YANG, Matthew Ozon, Richard Flagan, John Seinfeld, California Institute of Technology

     Abstract Number: 214
     Working Group: Aerosol Physics

Abstract
Atmospheric aerosols represent a fundamental part of the atmosphere, acting on its physical and chemical properties. Aerosol particles play a predominant role in climate by acting as cloud condensation nuclei (CCN) and ice nucleating particles (INP). Furthermore, they modify the global radiation budget through scattering or absorbing (solar) radiation. However, the radiative forcing effect of aerosols remains hindered by large uncertainties in climate models. The impact of aerosols on radiation is strongly dependent on the size distribution of particles. To better understand the atmospheric physics and chemistry processes contributing to aerosol formation and evolution, we propose an estimation method. The general dynamics equations in terms of number and mass distributions are coupled with kinetic equations to model the time evolution of aerosol population and the chemistry of volatile compounds. From these models and measurement data, we estimate the parameters (along with their uncertainties) dictating the evolution of aerosol systems, e.g. mass concentrations, using artificial intelligence based on data assimilation techniques. This work is a proof of concept for a controlled environment. We demonstrate the applicability of our method using simulated data that resemble real experimental setups. The results of this research would benefit enhancing model prediction and uncertainty quantification of climate change through better estimation of parameters and their uncertainties in climate models.