# Mapping Saturn using deep learning

## TL;DR

This paper introduces PlanetNet, a deep learning model that efficiently maps spectral and cloud features on Saturn, enabling detailed analysis of atmospheric phenomena like storms from large datasets.

## Contribution

The paper presents a scalable deep learning approach for planetary spectral mapping, surpassing traditional methods in accuracy and efficiency.

## Key findings

- Mapped Saturn's 2008 storm with detailed spectral features
- Identified regions of vertical upwelling and cloud variation
- Enhanced understanding of storm dynamics and composition

## Abstract

Clouds and aerosols provide unique insight into the chemical and physical processes of gas-giant planets. Mapping and characterizing the spectral features indicative of the cloud structure and composition enables an understand-ing of a planet's energy budget, chemistry and atmospheric dynamics. Current space missions to Solar System planets produce high-quality datasets, yet the sheer amount of data obtained often prohibits detailed 'by hand' analyses. Current techniques rely mainly on two approaches: identifying the existence of spectral features by dividing the fluxes of two or more spectral channels, and performing full radiative transfer calculations for individual spectra. The first method is not sufficiently accurate and the second is not easily scalable to the entire planetary surface. Here we have developed a deep learning algorithm, PlanetNet, that is able to quickly and accurately map spatial and spectral features across large, heterogeneous areas of a planet. We use PlanetNet to delineate the major components of the 2008 storm on Saturn, enhancing the scope of the area previously studied and indicating regions that can be probed more deeply with radiative transfer models. Our spectral-component maps indicate compositional and cloud variations of the vast region affected by the storm, showing regions of vertical upwelling, and diminished clouds at the centre of compact substorms