Emslie, Gordon

Solar eruptive events are the most powerful releases of energy in the solar system,
releasing in excess of 10^25 J of energy over some tens of minutes (equivalent to
over a billion megaton bombs). The accompanying coronal mass ejections (CMEs)
can have a mass of some 10 billion tons and can travel at speeds exceeding 1000
km/s. A solar eruptive event can cause severe disruptions in the space environment
around the Sun: high fluxes of high-energy charged particles (so-called solar energetic
particles – SEPs) move outwards along the interplanetary magnetic field lines, and
hazardous doses of extreme ultraviolet (EUV) and X-ray radiation are produced.
The Heliophysics Group at MSFC is planning a launch of the Hi-C EUV imaging
sounding rocket instrument in tandem with a launch of the FOXSI hard X-ray imager in
March 2024, in connection with the Solar Flare Sounding Rocket Campaign. The Hi-C
and FOXSI instruments will be launched near-simultaneously from Poker Flats, AK,
and will both focus on a specific region of the Sun that has been identified as having a
high probability of producing flaring activity during the ten minutes or so of data
collection that is possible on a suborbital rocket launch. The launch countdown will be
moved beyond its T-10-minute hold point based on sufficiently encouraging evidence
that flare activity is imminent. We are collaborating with this group to improve the
identification and near-real-time forecasting of solar flare events, in order to maximize
the potential success of the rocket missions.
EUV emission, produced by atomic species in the 10^5-10^6 degree solar
atmosphere, the so-called “transition region” between the hot 10^6+ degree corona
and the relatively cool (~10^4 K) solar chromosphere (the “surface” of the Sun), is
characterized by a set of intense, relatively narrow, spectral lines, each produced by a
given atomic species that is formed over a fairly narrow range in temperature.
Restructuring of the solar atmosphere during the early phases of a flare can cause
significant changes in both density and temperature, with correspondingly dramatic
changes in the intensity of spectral lines. This offers a real-time, “surgical” probe of the
details of the evolving solar atmosphere and hence the likelihood of imminent flaring
activity.
Funded by a three-year grant from the NASA EPSCoR program through the University
of Kentucky, faculty and students at WKU will apply convolutional neural network
(CNN) machine learning techniques to analyze datacubes derived from the extreme
ultraviolet observations of the Solar Dynamics Observatory (SDO) Atmospheric
Imaging Assembly (AIA; all data are available in the public domain) in order to detect
patterns in 4-D space (x,y,𝜆,t) that are precursors to flare activity, and thus provide a
timely prediction of activity to the rocket launch teams. Our algorithms will be trained
and validated using historical SDO AIA data, consisting of solar maps in seven
discrete EUV spectral lines (each representing a different temperature in the solar
atmosphere), and produced every twelve seconds. The reliability of any algorithm will
be tested simply by using a “blind” study in which a mix of data sets, some of which
produced high levels of subsequent activity and some of which did not, will be
analyzed by investigators who do not have advance knowledge of the subsequent
behavior. Once we have sufficient confidence in the predictive power of the algorithms,
we will use this technique, applied to real-time SDO AIA data on the planned launch
day, to inform the Solar Flare Sounding Rocket Campaign, first in a planned dry run in
March 2023 and then again for the actual launch in March 2024.
In addition to its practical value of a reliable indicator for use in this Campaign, the
development of such a ten-minute flare predictive capability would add considerably to
our understanding of how flares and solar eruptive events evolve, in turn leading to a
more robust predictive capability .

Computational Methods:

In this project, we will develop neural network models for nowcasting solar flares using
solar images for the last ten years. The dataset has over 183,960,000 raw images
(with 4096 x 4096 pixels for each image). A pre-data selection will be applied to select
about 3,500,000 images or 2500 solar flare events with over 300,000 training data
samples . Then, novel neural network models using 2D and 3D imagery data will be
developed. Depending on the test performance, additional imagery data might be
needed. In addition, other data modalities are also used to enhance performance
besides the imagery data.
The potential neural network structure might be used in the project, including
Convolutional Neural Networks (CNNs), Long-Short Term Memory (LSTM), and
Long-term Recurrent Convolutional Networks (LRCNs). In addition, physics-based
neural network methods might also be applied. Due to the challenge and unique
requirement of the project, none of the UKY or commercially available or
in-development methods can be used on this project.

Software:

The following Python libraries might be used for this project:
● Basic scientific computing and image processing libraries, such as Numpy ,
Scipy , scikit-learn , OpenCV
● giotto-tda , available at https://giotto-ai.github.io/gtda-docs/latest/index.html ,
is a open-source, high performance topological machine learning toolbox in
Python built on top of scikit-learn
● Scientific computing for neural network development, such as PyTorch ,
Tensorflow , and Keras
● Solar data-specific libraries, such as Sunpy and aiapy .
In addition, basic project management and version control software are also needed,
such as Anaconda/Miniconda, rClone, git, etc.

Collaborators:

Dr. Paolo Massa

Dr. Ivan Novikov

Dr.Gongbo Liang, Texas A&M University -San Antonio

Students: 

David J Boils, Undergrad, Added on 07/26/2022 on LCC Resources

Jarret Packwood, Added on LCC Resources on 12/20/2023 

Grants:

Publications: