Application of machine learning for modeling and characterizing electron and proton auroras on Mars

Auroras on Mars are known since their first discovery in 2005 by Mars Express and subsequently have been observed by Mars Atmosphere and Volatile Evolution (MAVEN) since 2014. Since 2021, Emirates UV spectrometer (EMUS) onboard the Emirates Mars Mission (EMM) has been observing Martian auroras with an unprecedented frequency. These auroras are seen as FUV and EUV emissions of H, O, CO, and CO2 and are categorized based on their morphologies and the particles that are responsible for these emissions. Electron precipitation on the nightside causes discrete and diffuse auroras whereas solar wind protons penetrating the Mars atmosphere cause proton auroras on the dayside. EMUS also detected new discrete auroras extending thousands of km into the nightside with a sinuous morphology. The variety and abundance of Mars aurora occurrences make them an important tool for gaining new insights into solar wind interaction with Mars's magnetosphere. Mars aurora research therefore involves characterizing aurora occurences in terms of solar activity, seasonal variability, IMF orientation, crustal magnetic fields, and energies of precipitating particles. In this work, we present applications of machine learning for modeling proton auroras as well as automatically detecting discrete electron auroras, leveraging a plethora of MAVEN and EMUS observations. We also focus on explainability of these ML models, commonly perceived as “black-boxes”, and approaches to analyze and validate correlations learned by these models. We discuss in detail the characteristics of proton and electron auroras thus revealed by these models and present future directions for such applications on Mars and other planets.