Space weather events can impact satellite communications, astronaut health, and the electric power grid. It is thus of utmost importance that we develop efficient, reliable tools to determine when space weather events, such as solar flares, will occur and how strong they will be. The SWx TREC Deep Learning Laboratory has developed several state-of-the-art machine learning projects to improve solar flare prediction through the use of deep learning models, generative adversarial network data augmentation, and explainable artificial intelligence techniques. In particular, we compared two generative adversarial networks (GANs) to super-resolve the Solar and Heliospheric Observatory’s Michelson Doppler Imager (SOHO/MDI) magnetogram data to match the quality of the Solar Dynamics Observatory’s Helioseismic and Magnetic Imager (SDO/HMI) magnetogram data. We find that both GANs are able to preserve key features of the original SOHO/MDI magnetogram data while achieving better resolution to match the SDO/HMI data. In the future, we will use the combined, augmented dataset in a Long Short-Term Memory model for solar flare prediction to see if training on the expanded dataset results in improved predictive power compared to training on the SDO/HMI dataset alone. In addition to data augmentation, we have used Local Interpretable Model-Agnositc Explanations (LIME) on our existing solar flare prediction model to provide more insight into specific predictions. This is an important step in building trust in our model and understanding what features are driving the model’s predictions. In this presentation, we will discuss these recent projects as well as future work that the SWx TREC Deep Learning Laboratory will tackle in order to advance the field of machine learning in space weather, including: improved hardware, better visualization capabilities, cutting edge models, software tools, and community resources.

With the COVID-19 pandemic still active, the Boulder Solar Alliance Research Experience for Undergraduates (BSA REU) decided to keep the program remote for a second consecutive year. Our coordination team took lessons learned from the 2020 virtual BSA REU program and adapted the research experience to suit a virtual environment, especially with respect to increased technological support. The primary changes, as well as the reasons for implementing them, are outlined below. Due to the virtual nature of the program, all of the projects relied more heavily on coding. In response, the BSA REU team invested more time and resources in programming tutorials and weekly programming help sessions in Python, IDL, and MATLAB. The participants also faced unequal access to high-quality hardware resources in a remote environment. As a result, students received a technology stipend to help them upgrade their computer and internet resources. Additionally, with an increase in the focus on programming, a higher number of projects in 2021 involved machine learning and data science techniques compared to previous years. However, many of the students were unfamiliar with machine learning (ML) concepts. The coordination team provided an introductory ML lecture and tutorial during boot camp and hosted a weekly ML sub-group meeting to provide support and resources for students involved in ML projects. Finally, without being able to present results in person, it was important to provide an interactive online experience for the poster presentation session. To make the final poster presentation more engaging in a virtual environment, we used Gather Town, an online service where participants create avatars that can interact with the virtual environment. In this presentation, we will discuss how the adjustments to the BSA REU program in a virtual environment, including those listed above, and how we think REU programs can adapt to future remote and hybrid options. We will also discuss what elements of a remote program can be carried forward into an on-site program to enhance the on-site experience.

Space weather forecasting remains a national priority in the United States due to the impacts of events like solar flares to life on Earth. High energy bursts of radiation originating from solar flares have the potential to disrupt critical infrastructure systems, including the power grid and GPS and radio communications. The rise of machine learning and the development of higher-quality instruments has greatly improved solar flare prediction models over the past decade. However, the magnetogram data used for solar flare forecasting—taken by the Solar and Heliospheric Observatory/Michelson Doppler Interferometer (SOHO/MDI) and the NASA Solar Dynamic Observatory/Helioseismic and Magnetic Imager (SDO/HMI) instruments—are incompatible due to differences in the cadence, resolution, and size of the data. Furthermore, many studies only focus on data from a single instrument which disregards decades worth of potential training data that is necessary to understand solar cycles. In this work, we show Generative Adversarial Networks (GANs) can be used to super-resolve the historic lower-quality SOHO/MDI data set to match SDO/HMI quality to create a standardized magnetogram data set. The implementation of a Pix2Pix GAN produced some undesirable artifacts in the synthetic image while image translation methods CycleGAN and CUT preserved solar features present in the data more accurately, even in the absence of paired data. The resulting combined, higher-quality data set will be used to improve the predictive power of current solar flare forecasting models.