Speaker
Description
In this study we demonstrate how neural network (NN) models can be used to improve our understanding of OMPS LP measurements, optimize forward modeling, and enhance OMPS LP retrievals. Retrieving atmospheric profiles from limb observations requires modeling how solar light scatters through the atmosphere. Operational retrieval codes make approximations for computationally expensive processes like multiple scattering (MS) to enable their application to large volumes of measurements. In recent years, machine learning techniques such NNs have been demonstrated to significantly reduce the computational time of modeling methods. Here we present an alternative MS approximation method for limb simulations based on NNs. We also demonstrate an alternative NN retrieval algorithm for inverse problems where traditional methods fail, like retrieving NO2 from OMPS LP. We discuss the optimal training strategies for the NNs as well as the accuracies of the trained models.