Automated solar flare spectral line characterisation for the DKI Solar Telescope

Lead Research Organisation: University of Glasgow
Department Name: School of Physics and Astronomy

Abstract

The Daniel K Inouye Solar Telescope (DKIST) is a 4m solar optical and IR telescope under construction in Hawaii, which will start science operations in early 2010. This telescope will produce up to 54Tb per day of imaging and spectropolarimetric data. The telescope field-of-view covers a small part of the disk at high resolution - corresponding to 30km on the Sun at a wavelength of 500nm. This project uses photospheric and chromospheric spectropolarimetric data, in which the four Stokes parameters are recorded at many points across a spectral line at each pixel in an image, encoding information about the atmospheric magnetic field as well as its density and temperature structure. We focus in particular on the spectropolarimetry of solar flares, in which substantial changes in the chromospheric and sometimes also photospheric spectropolarimetric data are observed. The overall goals of the project are 1) automated characterisation of Stokes profiles in these regions and 2) fast model-fitting of intensity and Stokes' profiles using a library of forward-modeled Stokes profiles.

Publications

10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
ST/R504750/1 01/10/2017 30/09/2021
2039671 Studentship ST/R504750/1 01/10/2017 30/09/2021 John Andrew Armstrong
 
Description National Center for Atmospheric Research (NCAR) Solar Spectropolarimetry Summer School
Amount £2,000 (GBP)
Organisation NCAR National Center for Atmospheric Research 
Sector Academic/University
Country United States
Start 10/2018 
End 10/2018
 
Description National Solar Observatory (NSO) Daniel K. Inouye Solar Telescope (DKIST) Workshop
Amount £1,089 (GBP)
Organisation National Solar Observatory (NSO) 
Sector Public
Country United States
Start 06/2019 
End 06/2019
 
Title Database of Inverted Solar Flare Images 
Description I used the RADYNVERSION code to produce an inversion from every observation of a solar flare from the Swedish 1-m Solar Telescope (SST) from 09/06/2014. This tells us about the atmospheric parameters such as temperature and electron density that produce the observed spectral lines. 
Type Of Material Database/Collection of data 
Year Produced 2019 
Provided To Others? No  
Impact This database provides a wealth of data to analyse to determine the physics happening in the flaring chromosphere. 
 
Title Database of Solar Flare 09/06/2017 
Description We received data from the Swedish 1-m Solar Telescope (SST) which I then converted to a sensible format alongside generating appropriate metadata. 
Type Of Material Database/Collection of data 
Year Produced 2020 
Provided To Others? No  
Impact This dataset contains the most energetic flare of the last solar cycle meaning there is plenty of opportunities to use this data to study a really explosive event. 
 
Title Slic Training/Validation Dataset 
Description This is data from the Hinode Solar Optical Telescope which has been prepared for training and use in convolutional neural network classification tasks including being assigned a label for the feature that it includes. This consists of around 13000 images containing one of five features of the solar atmosphere: flare ribbons, filaments, prominences, sunspots and quiet (the lack of the other four features). 
Type Of Material Database/Collection of data 
Year Produced 2019 
Provided To Others? Yes  
Impact This database was used to train a convolutional neural network to address mammoth cataloguing problems in solar physics and the model is subsequently freely available. 
URL https://github.com/rhero12/Slic/releases/tag/1.1.1
 
Title Solar Image Classification Using Convolutional Neural Networks 
Description We trained a convolutional neural network to be able to catalogue images of the solar chromosphere and as a demonstration that this is a worthwhile task as automating these processes are incredibly useful. 
Type Of Material Data analysis technique 
Year Produced 2019 
Provided To Others? Yes  
Impact We concluded that this model represents a starting point where this kind of model can be implemented in data pipelines to speed up feature identification and also help researches locate data they are looking for. Furthermore, it can be useful for providing a starting point for other deep neural networks used in solar physics. 
URL https://github.com/rhero12/Slic/releases/tag/1.1.1
 
Description Craft Prospect ScotDIST CDT Placement 
Organisation Craft Prospect
Country United Kingdom 
Sector Private 
PI Contribution I undertook a mandatory internship at Craft Prospect as part of my ScotDIST CDT funding. I worked primarily on the development of deep neural networks for classification of images of the Earth from space. The main goal was to get these to be able to run in real-time on the hardware on-board CubeSats.
Collaborator Contribution My partners made their expertise freely available to help me in a field I was not familiar with.
Impact There is a paper currently in preparation for the work.
Start Year 2019
 
Description ScotDIST CDT 
Organisation University of Edinburgh
Department School of Physics and Astronomy
Country United Kingdom 
Sector Academic/University 
PI Contribution I ran a workshop at the UK-wide data intensive science CDT welcome meeting in Edinburgh in November 2018 focusing on unsupervised classical machine learning techniques for other students to attend.
Collaborator Contribution Several workshops/events have been run to introduce those within the CDT to industry and data science methods. This collaboration has also provided the funds for technical equipment such as graphics cards and computers. There have been several meetings where we all discuss our research and some interesting discussions have spawned from that. The CDT also provides the stipend, tuition fees and travel/maintenance/technical budget for my research project.
Impact Both papers that I am an author on have resulted directly from this collaboration as they provide my PhD funding: Fast Solar Image Classification Using Deep Learning and its Importance for Automation in Solar Physics (10.1007/s11207-019-1473-z) and RADYNVERSION: Learning to Invert a Solar Flare Atmosphere Using Invertible Neural Networks (10.3847/1538-4357/ab07b4).
Start Year 2017
 
Description ScotDIST CDT 
Organisation University of St Andrews
Department School of Mathematics and Statistics
Country United Kingdom 
Sector Academic/University 
PI Contribution I ran a workshop at the UK-wide data intensive science CDT welcome meeting in Edinburgh in November 2018 focusing on unsupervised classical machine learning techniques for other students to attend.
Collaborator Contribution Several workshops/events have been run to introduce those within the CDT to industry and data science methods. This collaboration has also provided the funds for technical equipment such as graphics cards and computers. There have been several meetings where we all discuss our research and some interesting discussions have spawned from that. The CDT also provides the stipend, tuition fees and travel/maintenance/technical budget for my research project.
Impact Both papers that I am an author on have resulted directly from this collaboration as they provide my PhD funding: Fast Solar Image Classification Using Deep Learning and its Importance for Automation in Solar Physics (10.1007/s11207-019-1473-z) and RADYNVERSION: Learning to Invert a Solar Flare Atmosphere Using Invertible Neural Networks (10.3847/1538-4357/ab07b4).
Start Year 2017
 
Title Craft Prospect Neural Network (cpnn) Library 
Description This software was the culmination of my internship with Craft Prospect and includes scripts to build classifier deep neural networks alongside applying distortions to images. There is also scripts to do statistical analysis of the trained deep neural networks using the confusion matrix. 
Type Of Technology Software 
Year Produced 2019 
Impact The software was used to demonstrate that a deep neural network can run in real-time for large field-of-view classification on-board a CubeSat. 
 
Title Slic (Solar Image Classification) 
Description The software is a convolutional neural network for detecting features in the solar chromosphere. It can perform these detections in real-time with processing of over 1300 images taking 7 seconds. 
Type Of Technology Software 
Year Produced 2019 
Open Source License? Yes  
Impact Slic has a lot of people talking about automating these processes in solar physics. 
URL https://link.springer.com/article/10.1007%2Fs11207-019-1473-z
 
Description DKIST first light announcement 
Form Of Engagement Activity A press release, press conference or response to a media enquiry/interview
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Media (as a channel to the public)
Results and Impact Press release on the first light observations by the DKIST, highlighting University of Glasgow's role, STFC's support, and our research activities that will benefit from DKIST observations.
Year(s) Of Engagement Activity 2020
URL https://www.gla.ac.uk/news/headline_708045_en.html
 
Description Explorathon 2019 
Form Of Engagement Activity Participation in an activity, workshop or similar
Part Of Official Scheme? No
Geographic Reach Regional
Primary Audience Public/other audiences
Results and Impact We had a stall at the Explorathon event which consisted of table top activities to convey our research to a broad audience. We focused on the many colours of the Sun and how we can probe different layers of the solar atmosphere by looking at different wavelengths of light. We demonstrated this with spectral lamps and spectrometers. Looking at the lamp would show the spectral lines of the element in the bulb and the key message was that we see these spectral lines which are formed in certain temperatures at different heights. We had a few hundred visitors to our booth which sparked discussion from the importance of studying the Sun to how to get involved in the future.
Year(s) Of Engagement Activity 2019
URL https://www.explorathon.co.uk/events/explorathon-unlocked-saturday/
 
Description PubhD Glasgow 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach Local
Primary Audience Public/other audiences
Results and Impact The format of PubhD is each research having a 30 minute slot with 10 minutes dedicated to explanation of research and 20 minutes dedicated to questions/scrutiny from the audience. The research has 10 minutes and a flip chart to describe their research along with the usefulness of it. The audience were very engaged and brought about lots of interesting discussion. The organiser was pleased with the turnout (higher than usual) and said it would be good to have us back again.
Year(s) Of Engagement Activity 2020
URL http://glasgowskeptics.com/event/pubhd-solar-flare-special/