Employing machine learning techniques for systematic and statistical studies of collisionless shocks

Using Magnetospheric Multiscale (MMS) data from Earth orbit, an automated clustering algorithm has been employed to classify dayside MMS data into four distinct regions: magnetosphere, magnetosheath, solar wind, and ion foreshock, as detailed in Toy-Edens et al. [JGR 2024].  Applied to eight years of MMS data, over 25,000 bow shock crossings were identified from all four MMS spacecraft. Using that event database, we highlight a series of results including: new, 3-dimensional, parameterized boundary model fits for the bow shock; statistical characteristics of the quasi-parallel and quasi-perpendicular bow shock; and a new 4-point timing algorithm to systematically determine bow shock normal directions. We detail new results concerning the accuracy and performance of the shock normal results, showing that this new approach works remarkably well. We also highlight some new results of kinetic shock behavior and compare those directly to results from state-of-the-art simulations of and corresponding predictions for collisionless shocks. We end with a discussion of future work, in which we hope to train a parameterized generative model for collisionless shock crossing data as a function of upstream plasma characteristics. Our hope is to be able to apply that model to collisionless shocks beyond 1 au, validating its performance with shock observations from other systems (including Venus, Mars, Jupiter, etc.), and ultimately apply it to solar and other astrophysical shocks that are beyond our reach for in situ observations.