A neural network inversion system for atmospheric remote-sensing measurements

Abstract

A neural network inversion system is being developed to retrieve physical properties of the atmosphere. The neural network is being trained with radiative transfer simulations, atmospheric measurements, and theoretical understandings about the physical properties and their signatures in satellite measurements. The learning and adjusting process will be very fast and automated. This study seeks to improve future remote-sensing algorithms by bridging visual understanding within the human brain and the retrieval techniques developed by researchers in scientific community. With the new inversion technique of remote-sensing measurements, we will greatly reduce the time and mass storage of conventional inversion methods.