Artificial Intelligence to Enhance Mission Science Output for In-situ Observations: Dealing with the Sparse Data Challenge

The goal of space exploration is to better understand the universe encompassing our small planet and its life. Space research includes remote (similar to observational astronomy) and in-situ observations made at the immediate location of the spacecraft. Observations of the first type usually deal with high-resolution and high cadence data (e.g., of the Sun’s disc images), often referred to as the so-called Big Data (e.g., Armstrong and Fletcher, 2019). In the last decade there has been great progress with processing this data due to a revolution in machine learning provided by a new generation of multi-layer artificial neural networks (e.g., LeCun et al., 2015). In this WP we would like to draw the community’s attention to another challenge of Sparse Data sampling in case of in-situ observations (e.g., Sitnov et al., 2020 and refs. therein). For instance, in the case of the Earth’s magnetosphere, there are fewer than a dozen dedicated probes beyond the low-Earth orbit available for magnetospheric observations at any given time. As a result, we still poorly understand the global structure and evolution of the magnetosphere, its magnetic field, electric currents, plasma pressure and high-energy plasma populations. We still don’t understand the mechanisms of the main activity processes, magnetic storms and substorms (McPherron, 2016; Wolf et al., 2017). The 2013 Heliophysics Decadal Survey called for a future Magnetospheric Constellation (MagCon) mission consisting of dozens of spacecraft to address such data sparsity challenges in such an enormous volume of space. Considering such future missions and the inescapable data sparsity challenge, along with the advance of instruments, orbits and probes, it is also critical to advance the application of Artificial Intelligence (AI) technologies to the analysis of space physics datasets, such as machine learning (ML), data mining (DM) and data assimilation (DA) (e.g., LeCun et al., 2015; Kubat, 2015; Kalnay, 2006).