High-Level Project Summary
We developed a methodology to strengthen existing satellite software that attempts to predict solar flares that could disrupt life on earth. We related DSCOVR and Wind satellite data to each other and used that to find patterns that may indicate when solar flares will appear.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
- We wanted to analyze magnetic flux data over varying time periods and monitor for spikes in flux that showed indicators of a solar flare
- Initially tried using Dynamic Time Warping to correlate data from Wind and DSCOVR satellites
- Pivoted to a different method of correlation, via conversion of epoch time to Unix time
- Calculated amplitude of flux and plotted that against the recalculated Unix time to correlate data from both satellites while avoiding time shift issues in data.
- Laid out the groundwork for future ML model
- Used python, NASA's CDF data interpreters, and cdaweb.gov data
Space Agency Data
We used Nasa datasets for our machine learning algorithm. The particular datasets we pulled the magnetic flux was provided by DSCOVR and wind. The datasets allowed us to determine the solar wind temperature, density, and velocity needed to predict the next solar flare.
Hackathon Journey
Our journey was full of twists. In many cases we had to pivot from our original methodology to something that was faster, simpler, or more accurate. We learned a lot about NASA data files and data types and about how to manipulate and interpret physics-related data. We resolved challenged by talking to mentors, strategizing and whiteboarding difficulties. We'd like to thank our advisors @mike_from_smithsonian and @GOJITOM on discord.
References
- https://www.spaceweatherlive.com/en/solar-activity/top-50-solar-flares/year/2022.html
- https://www.epochconverter.com/
- https://datascientyst.com/convert-unix-time-to-date-pandas/
- https://towardsdatascience.com/dynamic-time-warping-3933f25fcdd
- https://cdaweb.gsfc.nasa.gov/pub/data/wind/mfi/mfi_h2/2022/
- https://cdaweb.gsfc.nasa.gov/pub/data/dscovr/h0/mag/2022/
Tags
#machinelearning #ml #ai #ann #solarflares #nasa #jpl #rockets