The goal of the SWx TREC Deep Learning Laboratory (DLL) is to provide interdisciplinary resources to advance our knowledge of space weather using cutting edge software and hardware. If you are interested in working with the DLL on a project or want to know more about our resources, please contact Julie Barnum at  


The DLL currently conducts research in solar flare prediction and GNSS scintillation prediction.


The DLL has two Systems 76 workstations for machine learning research: one with an NVIDIA Titan RTX GPU card and 20 TB of local storage, and one with dual NVIDIA Titan V GPU cards and 85 TB of local HDD storage.



Founder Thomas Berger
Director Julie Barnum
Deputy Director Wendy Hawley Carande


Varad Deshmukh
Allison Liu
Kiera van der Sande

MISA seminar series

The Machine Intelligence for Space Applications (MISA) seminar series—hosted by the SWx TREC DLL—serves to communicate machine learning (ML) topics as they apply to the broader heliophysics and space weather science and prediction community. Presentations include a wide variety of subjects, including, but not limited to, DLL-specific research and hardware updates, non-DLL research introductions and updates, ML innovations and advancements, paper overviews, and community conference/meeting highlights. The MISA seminars will take place the first Tuesday of every month at 2:00 PM MT, occasionally changing due to holidays, meetings, conferences, etc. Talks are currently online-only (Zoom) and will be recorded and posted for the later viewing.

Boulder MISA (B-MISA) group mailing list

A group mailing list has been compiled of people in the Boulder area who are interested in the application of machine learning, including deep learning and more classical algorithms, to space weather science and prediction. Boulder Machine Intelligence for Space Applications (B-MISA) members can post information about relevant seminars and other local opportunities. Reach out to Julie Barnum if you'd like to be added to the mailing list.

Solar flare prediction

TREC solar flare prediction efforts employ novel topological data analysis (TDA), computational geometry (CG), feature engineering, long-short-term memory (LSTM), and convolutional neural network (CNN) architectures to predict solar magnetic eruptions. The systems are trained on the 20 PByte NASA Solar Dynamics Observatory (SDO) database using both magnetic field data from the Helioseismic and Magnetic Imager (HMI) instrument and chromospheric and coronal images from the Atmopheric Imaging Assembly (AIA) instrument.


  1. Deshmukh, V., T. E. Berger, E. Bradley, J. Meiss, “Leveraging the mathematics of shape for solar magnetic eruption prediction”, J. Space Weather & Space Clim., 10, 13, 2020.
  2. Newman, W. D., W. H. Carande, “Long-short-term memory model for solar flare predictions”, submitted to Frontiers in Astron. and Space Sci., 2020.

GNSS scintillation prediction

Global navigation satellite system (GNSS) atmospheric scintillation interference prediction and classification. This effort uses support vector machine (SVM), long short-term memory (LSTM), and spatio-temporal neural network architectures to predict the onset of ionospheric scintillation (bubbles) that can severely degrade GNSS signal reception in both equatorial and high latitude regions. Moreover, the algorithms developed must also be able to separate and identify ionospheric scintillation from other disturbance signatures observed on GNSS signals. These other sources maybe due to intentional or unintentional radio frequency interferences, multipath reflections, and satellite oscillator anomaly.


  1. Jiao, Y., J. Hall, Y. Morton, “Performance evaluation of an automatic GPS ionospheric phase scintillation detector using a machine-learning algorithm,” Navigation, J. Institute of Navigation, 64(3):391-402, DOI: 10.1002/navi.188, Summer 2017.
  2. Jiao, Y., J. Hall, Y. Morton, “Automatic equatorial GPS amplitude scintillation detection using a machine learning algorithm,” IEEE Trans. Aero. Elec. Sys., 53(1): 405-418, DOI:10.1109/TAES.2017.2650758, Online ISSN 1557-9603, 2017.