2018 Workshop: Next generation CEDAR science

We heard from an excellent set of speakers that helped progress the conversation around the use of data science in geospace. Ryan McGranaghan opened the session with a scene-setting presentation. These slides as well as contributed discussion topics from each of the session conveners can be found here: NextGenCEDAR Discussion Slides (pdf)

Central Points:

• Powerful use cases allow us to observe and begin to realize the potential of data science to progress geospace understanding
• Radically interdisciplinary collaboration is necessary to achieve success, and communication across interdisciplinary communities is a challenge
• Innovative new data science solutions should *complement* more traditional approaches in geospace
• Data science is a broad field, encapsulating the entire data lifecycle (from data storage and management through analysis and visualization). It is *NOT* simply machine learning
• We must invite the most capable/advanced people from data science community (specifically data curation, machine learning, and visualization) to participate in our community. This raises the question of how to get those people invested and interested in geospace science?

Speaker summaries:

First, we heard from Liz MacDonald about the 'The spark of crowdsourced opportunities & outcomes for Heliophysics’. She raised several though-provoking questions, including: 1) Is citizen science a new frontier for data science in geospace? 2) Why would we expect the same approaches to science that have guided progress for the past 10, 20, and 30 years (her point: we should focus on ‘doing new things’)? and 3) Can new discovery be created at the ‘gaps’ between fields (i.e., at the edges)? She presented compelling evidence about the potential of citizen science from the discovery of STEVE (Strong Thermal Emission Velocity Enhancement) aurora. Key questions from the subsequent discussion included: 1) How do we remove/avoid jargon for geospace research? and 2) Are there ‘right’ problems for citizen science? MacDonald NextGenCEDAR (pdf)

Next, Christine Gabrielse began her presentation by addressing the four V’s of ‘big data’: 1) volume; 2) variety; 3) veracity (i.e., uncertainty); and 4) velocity. She illustrated that ‘variety’ may be the niche that geospace currently best aligns with. She highlighted the data variety challenge through an analysis of mesoscale structures in the magnetosphere-ionosphere system and the progress that heterogeneous data fusion can achieve. Based on her excellent research, she suggested that our community needs to embrace data variety in the near-term. Conversation after her talk centered around the challenge of making data ‘usable’. Gabrielse NextGenCEDAR (pdf)

Chen Zhou gave an exciting overview of his research into using deep learning methods (specifically generative adversarial networks, GANs), achieving progress toward the challenge of incomplete and sparse data in geospace as a parallel to image completion problems. Chen made a prescient point that collaboration with computer scientists is critical to successfully applying these advanced machine learning methods. Zhou NextGenCEDAR (pdf)

Liam Kilcommons followed Chen’s presentation by providing a clear and compelling overview of an under-utilized, yet important, method for examining the robustness of data-driven discovery in geospace: cross-validation. He highlighted k-fold cross-validation using Super Dual Auroral Radar Network (SuperDARN) and Advanced Magnetospheric and Planetary Electrodynamics Response Experiment (AMPERE) data. His presentation was a glowing example of how new geophysical understanding can be obtained through data-driven methods; a testament to a theme of the session that innovative new data science solutions can complement more traditional approaches in geospace. Kilcommons NextGenCEDAR (pdf)

Piyush Mehta then used the ionosphere-thermosphere system as a case study to apply a new data assimilative framework. He raised important issues such as the distinction between theoretical, empirical, data-driven, and computational models and the idea of ‘predictability’ of the ionosphere. His presentation emphasized four points critical to data assimilation and areas of importance discussed throughout the CEDAR 2018 workshop: 1) model errors; 2) time step evolution; 3) initialization; and 4) calibration. The post-presentation discussion centered around the best approaches to capturing nonlinearities of geospace in the data assimilative model. Mehta NextGenCEDAR (pdf)

Finally, Eric Sutton concluded the contributed talks with a fascinating look at new approaches to thermospheric neutral density specification. He framed his new approach through a clarification between chaotic (e.g., the troposphere) and strongly-driven systems (e.g., the ionosphere-thermosphere). His new assimilative approach demonstrated accuracy with respect to the Challenging Minisatellite Payload (CHAMP) and Gravity Recovery and Climate Experiment (GRACE) observations. In the discussion-oriented ethos of the session, Eric raised intriguing questions: 1) How do estimated drivers represent reality? and 2) How to disentangle solar and geomagnetic influences? Sutton NextGenCEDAR (pdf)