Dr. Abigail Azari

University of California, Berkeley

Integrating Data Science and Planetary Science – Building Statistical Views of Planetary Space Environments

Abstract: The best estimations of data size returned from the earliest generation of interplanetary missions launched in the 1970s are on the scale of a hundred gigabytes. Today, modern missions, are collecting multiple terabytes. These modern missions, including the MAVEN mission to Mars and the Cassini mission to Saturn, are transforming the field of planetary science rapidly to an observationally data-rich field, allowing for the applicability and at times necessary use, of machine learning and other data science methods to analyze returned datasets. However, similar to other fields in the geosciences, the datasets returned from spacecraft often are spatiotemporal in nature, and our scientific interest focuses on identifying rare events. These issues are compounded from the fact that in many cases, planetary mission datasets present the first, and sometimes the only observations of extreme environments for the foreseeable future.

In this presentation I present solutions to these challenges with use cases including: 1) the development of an interpretable machine learning model to identify extreme plasma intensification events around Saturn, and 2) the design of a searchable database, most commonly used in industrial data pipelines, to separate and quantify the spatio-temporal nature Mars’ space-environment from influences including the interplanetary magnetic field orientation. If time permits, this presentation will also include a summary of a current project to apply Bayesian statistics to quantify the influence of the solar magnetic field direction in the upper atmosphere of Mars. Both of these projects aim to further understanding of how plasma moves around, and through planetary environments. Through the use of the data intensive methods implemented in these projects, these applications provide system-wide perspectives and allow us to further understand current, and past planetary environments. Finally, this presentation summarizes recommendations submitted to the recent Planetary and Astrobiology Decadal Survey for the successful integration of machine learning for planetary sciences’ challenges and needs.