Heliophysics Hackweek 2020 Coronal Holes Team
Publishes Results at NeurIPS 2020

Full-disk image of the Sun captured at an extreme ultraviolet (EUV) wavelength of 193Å by the Atmospheric Imaging Assembly (AIA) onboard the Solar Dynamics Observatory (SDO). NASA image.

In August 2020, the Computational and Information Sciences and Technology Office (CISTO), with support from the NASA Center for Climate Simulation (NCCS) at NASA Goddard Space Flight Center and NVIDIA, hosted a virtual, global, two-week event, Heliophysics Hackweek 2020 (HW20). A year in the making, HW20’s main goals were to advance the use of artificial intelligence and machine learning (AI/ML) by scientists across and beyond NASA Goddard and, by doing so, increase cross-agency collaboration and open science.

Even in the midst of a global pandemic and after two postponements, 60 participants came together virtually during the two weeks of HW20 to train, learn, and code together in five research project teams, using a variety of online collaboration platforms. Participants committed a significant amount of time, energy, and creativity to the event, which was a success according to all involved.

Beyond increasing the application of AI/ML to NASA datasets by heliophysicists and collaborators, one of the event’s main goals was to foster ongoing collaboration and open science. As the five project teams made their final presentations, we learned that several participants planned to continue working together after the hackweek, some adding a few more team members to their mix, and some hoping to publish their hackweek coding results together. This hoped-for outcome made the difficult year of planning even more rewarding.

As an example, several members of the HW20 “Coronal Holes” research project team and a few other hackweek participants continued their collaboration and recently published their results at the prestigious 34th Conference on Neural Information Processing Systems (NeurIPS 2020). The team’s coauthored paper and poster, titled “SEARCH: SEgmentation of polAR Coronal Holes,” were published at the Machine Learning and the Physical Sciences Workshop at NeurIPS on December 11, 2020.

Ajay Kumar Tiwari at the Centrum Wiskunde & Informatica (CWI), Netherlands, had originated and submitted the idea for the SEARCH project in early 2020 to the HW20 “Space Weather” team (later renamed the “Coronal Holes” team). The idea was later selected as one of the five final hackweek team projects by NASA’s HW20 planning committee.

Benoit Tremblay from the Laboratory for Atmospheric & Space Physics at CU Boulder helped organize the Coronal Holes hackweek team, and a number of participants from within and beyond NASA joined to contribute to the team’s hackweek project: Andong Hu from CWI, Netherlands; Silvina Guidoni from American University; Michael Kirk, Emily Mason, and James Staeben from NASA Goddard; Matthew Penn from NVIDIA; and Tanmoy Samanta from George Mason University.

In the four months since the event, seven members of the Coronal Holes team worked on the SEARCH project, along with two additional HW20 members—Brent Smith from the Johns Hopkins University Applied Physics Lab (APL) and Linnea Wolniewicz from CU Boulder. This expanded, global team worked together virtually across several time zones to take the Coronal Holes hackweek project further and submit the coauthored SEARCH paper to NeurIPS 2020.

SEARCH collaborators. Clockwise from top left: Ajay Kumar Tiwari, Benoit Tremblay, Andong Hu, Linnea Wolniewicz, Matthew Penn, Tanmoy Samanta, Brent Smith, Silvina Guidoni, and Michael Kirk.

For the SEARCH paper, the nine collaborators used state-of-art network architecture to identify coronal holes (CHs) on the solar surface. The team used extreme ultraviolet (EUV) images from the NASA-operated Solar and Heliospheric Observatory (SOHO) and Solar Terrestrial Relations Observatory (STEREO) satellites. The team accessed and preprocessed the EUV images using a data reduction pipeline and combined them into synchronic maps—heliographic projections of the EUV solar corona—recorded by the Extreme UltraViolet Imager (EUVI) aboard the SOHO spacecraft.

Next, to reduce bias, the team used two unsupervised ML algorithms ranging from clustering to segmentation to identify polar CHs in synchronic maps. CHs are an important source for solar wind—the streams of plasma and particles from the Sun impacting Earth’s magnetic field. Identifying CHs with this method enables researchers to better understand the dynamics and evolution of the Sun’s coronal holes.

Following the team’s publication at NeurIPS 2020, Ajay Kumar Tiwari, the originator of this hackweek team project idea, observed: “It was such an exciting project to work on together at the hackweek, and a number of colleagues continued to work together afterward, pushing ourselves further with the promising results that we obtained at HW20.”

Tiwari continued, “The news of the paper being accepted at NeurIPS 2020 was so encouraging, and we all welcomed it. The whole experience of HW20 and the work afterwards was so satisfying…to collaborate with colleagues spanning the globe in different time zones was definitely a fruitful experience. For me personally, taking an idea from conception to completion was gratifying. I got a chance to work with some talented and wonderful people, forging new collaborations and friendships.”

CISTO senior scientist and HW20 NASA sponsor Dr. Mark Carroll observed, “The hackweek brought together a diverse group of participants. Our hope was that it would serve as a building block for future collaborations. The publication of this paper and presentation are exactly the results that we were hoping for when we decided to host a hackweek.”

The HW20 planners are pleased to hear about this ongoing collaboration and hope to make further announcements about ongoing collaborations and publications from other HW20 project teams in the near future. To further their research, all HW20 participants working on NASA-affiliated research are welcome to apply for NASA funding and access to NCCS high-end computing systems, NCCS services, and resources including analytics, data sharing, and visualization.

Related Links

Hackweek 2020 Teams Use Machine Learning and GPUs to Analyze Heliophysics Data,” NCCS News Highlight, 10/13/20.

“SEARCH: SEgmentation of polAR Coronal Holes,” Proceedings of the 34th Conference on Neural Information Processing Systems Machine Learning and Physical Sciences Workshop, 12/11/20; Tiwari, A. K., B. Tremblay, A. Hu, L. Wolniewicz, M. Kirk, S. Guidoni, E. B. Smith, M. Penn, T. Samanta, 2020:

Sean Keefe, NASA Goddard Space Flight Center