Identifying Fundamental Upstream Drivers of Ion Escape at an Unmagnetized Planet Using Machine Learning Methods

Abstract:

Understanding the dependence of Martian ion escape with upstream solar radiation and solar wind conditions is crucial for the studies of long-term atmospheric loss and evolution of Mars, and more generally, unmagnetized planets. However, quantifying the dependence of the Martian ion escape rate on multiple upstream drivers (solar Extreme UltraViolet flux, solar wind, and interplanetary magnetic field) from observations can be challenging. With limited data and correlated upstream parameters, it is difficult to distinguish the effect of each individual parameter on the ion escape. Besides, manually constructing any empirical model of the ion escape rate as a function of multiple upstream parameters risks introducing subjective bias. In order to avoid these limitations in traditional data analysis methods, we apply machine learning methods to study the Martian ion escape variation with multiple upstream drivers. We train neural network models using data from the NASA Mars Atmosphere and Volatile EvolutioN and ESA Mars Express missions, with positions and multiple upstream parameters as input and planetary ion densities and velocities as output. The successfully trained model will return spatial distributions of the escaping ion density and velocity with any upstream condition input, and thus can be used to quantify the ion escape rate dependence on multiple upstream drivers and to study the underlying mechanism of these variations. We will also compare the results from machine learning methods with those from traditional data analysis methods to assess the advantages and limitations of both methods in studying Martian ion escape variation.