High-Level Project Summary
While the major highlight for this challenge was on to predict intense CMEs like the Carrington Event, our group understood that the focus needs to be on more grounding tasks. We tried to come up with a solution to correct and re-calibrate the noisy data from the FC instrument of DSCOVR spacecraft. The more accurate the data from DSCOVR is, the better the prediction models would be. As such, we will be showcasing two solutions that can be integrated to: 1. Remove Noise and 2. Predict Solar Intensities.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
Project shall be explained in 3 phases.
- Data Preparation
- De-noising
- Prediction
Data Preparation[1] : The following datasets were used to study the data collected by the space probes, DSCOVR and WIND:
- DSCOVR_H1_FC: DSCOVR spacecraft data measuring proton velocity, solar wind density, solar wind temperature.
- DSCOVR_H0_MAG: DSCOVR spacecraft's data measuring magnetic field.
- WI_H1_SWE: WIND spacecraft data measuring proton velocity, solar wind density, solar wind temprerature.
- WI_H2_MFI: WIND spacecraft's data measuring magnetic field.
The basic idea was to compare the difference in the data between DSCOVR and WIND spacecraft. Since, both WIND and DSCOVR are orbiting the Earth-Sun L1 point, their observations would be very similar most of the times. Now, the tiny offset that occurs between the two spacecrafts could be corrected with the help of dynamic time warping[2].
However, there is a time interval difference in the data capturing in between DSCOVR_H0_MAG dataset and WI_H2_MFI dataset. To better plot the time-series data, we must normalize these values. In order to achieve that, we have used the WIND dataset to convert the date time variable into an integer and round it off to nearest DSCOVR dataset time interval at that position. Wherever necessary, a mean of data grouped by the time interval was performed. As a result, both the datasets could be merged as following:
In the above image, the B1F1 parameter from DSCOVR and BF1 parameter from WIND are shown merged due to time series normalization. Now that we have our data, we can use dynamic time warping. In our case, we used the fastdtw[3] library of Python.
De-noising: The output from Dynamic Time Warping provides us with a 1-to-1 mapping that deals with the time delay in data collection. The difference that remains in the magnetic values in between the two spacecrafts' data would be a result due to the current issues in the DSCOVR Fraday Cup. This raises the question, 'How can we deal with this noisy data ?". One solution could be to pass the DSCOVR magnetic field values through a de-noising auto encoder and train them to give out the "clean" data of WIND magnetic values. This way, we'll be able to develop a function that could automatically clean the magnetic field values and give out precise outputs for other parameters.
Prediction:
System User Interface:
Space Agency Data
TYPES OF DATA
DSCOVR_H0_MAG : Info
Parameters Used:
Date
Magnetic field magnitude (B1F1)
(x,y,z) coords for Magnetic field vector in GSE cartesian coordinates.
Time Interval: Year 2018
DSCOVR_H1_FC : Info
Parameters Used:
Date
Data Query Flag (DQF)
Solar winds' velocity (V_GSE)
Solar wind density (Np)
Solar wind proton temperature (THERMAL_TEMP)
Time Interval: Year 2016-2019
WI_H2_MFI: Info
Parameters Used:
Date
Magnetic field magnitude (BF1)
(x,y,z) coords for Magnetic field vector in GSE cartesian coordinates.
Time Interval: Year 2018
Hackathon Journey
Five people from different parts of the world united for one of the biggest hackathon experiences. Alfaxad, Ved, Rishabh, Shashank and Erick. This is our first time to meet in this hackathon but the passion, dedication and collaboration is something to long for. Learning from each other, we realized that we all united because of the inspiration that NASA has given us. From our childhoods we have all been fascinated by achievements by adventurous people who chose to defy the odds and start the age of space exploration. To us NASA space apps is an opportunity to be a part of the adventure, an opportunity to contribute to the work that other people have done. It makes us feel as a part of something bigger than ourselves, some of us long to have an opportunity to watch a live rocket launch and describe it as one of the happiest moments of our lives. But until that day comes, we can as well unite build friendships, solve interesting challenges and be a part of the adventure. We also believe that diversity leads to best solutions to problems as diverse people bring in different perspectives and we made that our advantage.
With the same mindset, we set to solve "save the earth from another carrington event". We were excited to tackle a challenge that has a large scale impact and push ourselves to work with tools and frameworks we were not familiar with before just to make this possible. We resolved all setbacks by taking initiative, rapid research, testing, and collaboration. We were all willing to assist each other and sometimes one would stay up and working while other took power naps. It is wonderful to see how much productivity one can have during a hackathon. We also had time to connect, discuss hobbies, interests and careers. Honestly, this is an experience we long to have each year and it exceeded our expectations. We would personally like to the the NASA space apps team, volunteers, local leads, and collaborators who have worked so hard to make this event possible.
References
Tags
#AI #SAVER #DSCOVR #WIND #SPACE