Awards & Nominations
Sun Trackers has received the following awards and nominations. Way to go!
Global Finalists Honorable Mentions
The Sun Tracker: a machine learning model that predicts Carrington-like Events using DSCOVR and WIND
High-Level Project Summary
Our algorithm was designed to monitor the activity of solar winds, targeting government agencies for environmental, climate and communication control. It is capable of analyzing solar wind emissions data detected by the DSCOVR spacecraft and the WIND mission, both developed by NASA. Furthermore, it can classify such emissions as dangerous to our technological societies or not so hazardous. The Sun Tracker algorithm helps to identify the intensity and velocity of those solar winds, being able to warn us of dangerous coronal mass ejections. Thus, preventing damage to earth telecommunications satellites, power lines, GPS systems and other electrical or electronic systems.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
What exactly did we develop?
- An easy applied algorithm that predicts electromagnetic storms from the sun;
- Predictions on the velocity of nonlinear protons from Earth-directed solar coronal mass ejections;
- An decentralized software open to stakeholders from all over the world, who want to contribute to our solar wind forecast.
Our Sun constantly “vomits” radiation and plasma in random directions. These Solar Flares and Coronal Mass Ejections (CMEs) are capable of heating our Earth and destroying our telecommunications satellites, power lines, GPS systems and other electrical or electronic systems.
The Sun Tracker Algorithm is a decentralized software that uses data from WIND and DSCOVR spacecraft to predict electromagnetic storms from the sun and the resulting velocity of non-linear protons from Earth-directed solar CME.
How does it work?
Our prototype model was made via the Decision Tree algorithm, on Orange. We evaluate the performance of our model through 2 model performance metrics. The R² of the built model was 0.795, while the MAE was 29.103 km/s. With such data, we can see the precision potential of our model. For the purpose of raising awareness of the MAE, it is pointed out that the observed speed of the solar winds varies between 258.72 and 770.727 km/s.
Although we only use 6 parameters and only 2022 data, our model can become more complex and assertive with more data input and greater infrastructure around the training process. NoSQL database structures such as MONGODB can be implemented free of charge to bring greater data handling capabilities.
What benefits does it have?
The great strength of the solution is its great potential for decentralization and global cooperation, in the act of feeding the database used in AI training. In addition, there is a strong potential for opening up the system to collaborating agents, so that they can feed the base with their own data. Such integration can be done via simple APIs. Likewise, the implementation of Webhook APIs allowed the backend associated with this solution to proactively communicate the collaborating agents when levels of solar wind speed exceed acceptable thresholds, thus maintaining a constant vigilance and a global potential alert for such events.
Combining data contribution with proactive alerting, it is possible to use common and recurring technologies in the market today to create a global collaborative network of surveillance and alert of solar wind events. Therefore, our project embraces the sharing economy, global cooperation and decentralization of means to achieve bigger and more important goals.
What do we hope to achieve?
We hope to predict adverse events that may cause economic, social, military, and other damages directly and indirectly linked to the planet's electrical and electrotechnical infrastructure, in addition to generating anticipated responses for the preparation of countermeasures.
What tools, coding languages, software did you use to develop your project?
We created our MVP using the Decision Tree algorithm, on Orange.
Space Agency Data
In order to predict the speed of solar winds, we used 6 parameters from Ang_dev, Bx, By, Bz, Vx, Vy data, with a database of 180K records covering all dates from 2022 to the end of August. We also used data obtained from the DSCVRand WINDplatforms, whose parameters have high reliability. Angular Deviation data were found with a different temporal cut from the others. The base ended up going to the year 2020 - as a possible consequence of a shadow on the probe or a technical failure due to a burn in the shadow - which we computed as an error and we did not use the data at the time of training the model. Such an event is shown as an aggregate risk of the solution.
Hackathon Journey
How would you describe your Space Apps experience?
For most of us, this was our first time participating in NASA's Space Apps Challenge, and if we could summarize our entire experience in only one word, it would be accomplishment! Why? Throughout this hackathon, we spent hours searching, studying, and creating this whole new idea for solving such an interesting challenge. Together, we went through some really difficult moments when our idea seemed so far away. With much responsibility and resilience we managed to figure it out. In only two days we did our best using AI, Machine Learning, and NASA's data, to bring out the greatest in this algorithm.
What did you learn?
We learned a lot about working as a team, especially because we had to divide tasks and rely on each other's work. Furthermore, all the knowledge we gained about space weather, solar activity, and how it affects our lives, as well as the prospects for future solar activity, is likely one of the most valuable parts.
What inspired your team to choose this challenge?
After having a look at all the challenges, the ones involving space weather and/or artificial intelligence stood out. Each of us did a brief first search on the topics, and after that, we reunited to decide the challenge. We chose to work on "Save the Earth from another Carrington Event!" as it looked like the most interesting one, not only because of its positive impact on our society, but also because of our previous strong interest in AI and Solar Flares.
What was your approach to developing this project?
We divided our project into a few parts: collecting, organizing, and compiling information from WIND and DSCOVR databases; coding and teaching our AI to analyze that data and process it; and creating a more visual approach to our algorithm using Orange software as a machine learning toolkit. Every member of our team was responsible for completing their portion of the project. And just like this, we created, in a simple and easy to visualize way, an algorithm capable of identifying dangerous solar winds directed at Earth, with potential to bring many global actors together in the important cause.
How did your team resolve setbacks and challenges?
Throughout the challenge, we were always supporting one another and reinforcing our interest in protecting our earth's electric systems from this massive danger. Our dream of somehow contributing to the big picture of our society is now full of accomplishments and new desires related to continuing on working in science with AI and machine learning for greater protection of our world.
Is there anyone you'd like to thank and why?
This challenge wouldn't be possible without each member of this team, whose interest, responsibility, and compromise made it possible. Thank you to our clever and creative writers, Mariana Vale Taveira, Roseno Gonçalves Lopes Filho, and Sara de Jesus da Costa França, and a special thanks to our group of programmers, Everton Tomazi, Priscilla Tomazi, and Leandro Vidal Costa Castelani, who worked day and night in order to make this solution possible.
We are beyond grateful to NASA for providing this global community of engaged innovators and changemakers. We’d especially like to thank the organizers and sponsors, for making this challenge happen, giving us this life-changing opportunity of working together.
References
Data:
ISTP Solar Wind Catalog Candidate Events
https://pwg.gsfc.nasa.gov/scripts/sw-cat/Catalog_events.html
CDAWeb Published Data - Index of /pub/data/wind/mfi/mfi_h2/2022
https://cdaweb.gsfc.nasa.gov/pub/data/wind/mfi/mfi_h2/2022/
WIND Near Real-Time Data
https://pwg.gsfc.nasa.gov/windnrt/
CDAWeb Published Data - Index of /pub/data/dscovr/h0/mag/2022
https://cdaweb.gsfc.nasa.gov/pub/data/dscovr/h0/mag/2022/
DSCOVR Space Weather Data Portal
https://www.ngdc.noaa.gov/dscovr/portal/#/
Resources:
WIND Understanding Interplanetary Dynamics
https://pwg.gsfc.nasa.gov/wind.shtml
Conway W. Snyder, Marcia Neugebauer, U. R. Rao. The solar wind velocity and its correlation with cosmic-ray variations and with solar and geomagnetic activity
https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/JZ068i024p06361
A. N. Fazakerley, L. K. Harra, L. van Driel-Gesztelyi. AN INVESTIGATION OF THE SOURCES OF EARTH-DIRECTED SOLAR WIND DURING CARRINGTON ROTATION 2053
https://iopscience.iop.org/article/10.3847/0004-637X/823/2/145/meta
M Calisto, I Usoskin, E Rozanov. Influence of a Carrington-like event on the atmospheric chemistry, temperature and dynamics: revised
https://iopscience.iop.org/article/10.1088/1748-9326/8/4/045010/meta
Consuelo Cid, Elena Saiz, Antonio Guerrero, Judith Palacios, Yolanda Cerrato. A Carrington-like geomagnetic storm observed in the 21st century
https://www.swsc-journal.org/articles/swsc/abs/2015/01/swsc140015/swsc140015.html
Consuelo Cid, Elena Saiz, Antonio Guerrero, Judith Palacios, Yolanda Cerrato. Searching for Carrington-like events and their signatures and triggers
https://www.swsc-journal.org/component/article?access=doi&doi=10.1051/swsc/2016001
Solar storm Risk to the north American electric grid
The Geomagnetic Storm of 1989
https://www.youtube.com/watch?v=mEqE-g128kk
Project video:
1859 Carrington-Class Solar Storm Pummeled Earth's Magnetic Field | Video
https://www.youtube.com/watch?v=dVS4Q4VgDxk&ab_channel=VideoFromSpace
Tools:
Orange
Tags
#artificialintelligence #machinelearning #algorithm #sun #solarwind #DSCOVR #WIND #earthprotection