High-Level Project Summary
Radio amateurs and scientists have been exploring the ionosphere since the 1930s. Information about electron density in the ionosphere can be very important for space weather observations and forecasts. In today's world, the ability to image the Earth's ionosphere is no longer enough, now it is required (and all technical possibilities for this exist) to predict changes in the ionosphere. Our team decided to use the most modern tool - Machine learning - to solve these problems.
Link to Final Project
Link to Project "Demo"
Detailed Project Description
Compared to the original definition of the challenge, we not only create and display an image of the Earth's ionosphere, but also solve the problem of change prediction, which is much more relevant to today's challenges. For this we have chosen the most modern tool, machine learning.
Most of the existing models for predicting ionospheric change based on machine learning do not take into account the equatorial ionization anomalies and contain a temporal resolution of about 2 hours. They do not segment data very well for training, testing, and validation, and they do not use batch normalization.
We present a solution that eliminates all these shortcomings , and improves temporal resolution from 2 to 1 hour.
Global ionospheric TEC (electron density in the ionosphere) data are publicly available from many associated ionospheric analysis centers, such as CODE. But also on the basis of the data provided by civilian scientists and radio amateurs on the propagation of radio waves, both from station to station and from the ISS to the station, we can make assumptions about the electron density in the ionosphere. This process is well studied at present, work on it has been carried out since the 1930s.
In our work we calculate electron density as a function of altitude, and various ionospheric layers. As the basis we will take the hypothesis, which was stated by Prof. Sean Victor in his book "Radio and Microwave Wireless Sys". It also introduces the value of the maximum useful frequency. And theoretically showed how to determine the maximum and minimum number of electrons in the ionosphere in the experiment shown in Figure 1. After consulting the mentors, we got the following information: after the end of the satellite's lifetime, some satellites are still working. And it is possible to adjust the signal emission to the one that suits us best for the study of the ionosphere.
this case, the transmitter will be a satellite and the receiver will be an amateur radio station (figure 2).
ikewise, George A. Hajj and Larry J. Romans in their article Ionospheric electron density profiles obtained with the Global Positioning System: Results from the GPS/MET experiment showed how with the Abel inversion technique, the electron density profile and the height of the tangent point at a particular instant during the occultation can be solved for.
If this is combined with the work of Eugene Bang,Jiyun Lee entitled Methodology of automated ionosphere front velocity estimation for ground-based augmentation of GNSS. And take from their work the Computation of Ionospheric Pierce Point (IPP) Velocities.
hus, taking all of the above into account, we will be able to obtain data on electron density in the ionosphere. Next, we apply machine learning techniques . The detailed methodology has already been described by Lei Liu , Y. Jade Morton , and Yunxiang Liu in their article entitled ML Prediction of Global Ionospheric TEC Maps. We use the convLSTM architecture, which is capable of learning functions from a spatiotemporal sequence. It has been successfully applied in many areas of multidimensional spatiotemporal forecasts. In this study, the convLSTM layer is used as the core module for predicting global TEC maps. The detailed structure of the convLSTM module can be found in Figure 5.
Here, two prediction strategies are implemented to predict global TEC maps:
Residual prediction: The output is the TEC residual between the map at time and map 24 hours ago,
Direct prediction: TEC maps are predicted directly by the ML model.
You can read more about other models and their testing in the article. From the article: This best-performing convLSTM model also shows more accurate prediction compared to c1pg, which is a 1-day predicted global TEC product released by CODE analysis center. Moreover, prominent structures, such as typicalEIA and TEC enhancement over the southern hemisphere, are successfully reproduced by the convLSTM model. Our statistical evaluation shows that the convLSTM model significantly outperforms both the c1pg and persistence models under all levels of solar activities for all lead times tested and at all latitudes. A slight performance degradation is observed for convLSTM under high solar activities compared to the ones under low solar activities. When compared to c1pg and persistence models, the convLSTM model shows better performance for all lead times tested under geomagnetic quiet conditions and for lead times below 8 hr of storm conditions.
Space Agency Data
Hackathon Journey
Our team participated in the hackathon for the first time. Before the hackathon we did not know each other, but the desire to solve complex problems brought us together.
Our team includes an engineer in the field of radio electronics, a tester, a designer, and a project manager. We analyzed the problem deeply, and we did our research. And together we thought about how we could "exceed expectations". What is the ultimate goal of people using an ionospheric map?
And we realized how we could bring something new and better to the user experience. Users today not only want a map of the ionosphere, but also a prediction of future changes. And today we have all the tools to implement such a solution.
References
https://www.mdpi.com/journal/atmosphere/special_issues/Ionospheric
http://ionosphere.meteo.be/ionosphere/
https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=6431636
https://www.itu.int/dms_pub/itu-r/opb/hdb/R-HDB-32-1998-PDF-E.pdf
https://www.waves.utoronto.ca/prof/svhum/ece422/notes/20c-ionosphere.pdf
https://www.unoosa.org/pdf/icg/2015/icg10/25.pdf
https://www.windy.com/?50.480,30.999,5,m:fdUahNw
https://www.diva-portal.org/smash/get/diva2:1447441/FULLTEXT01.pdf
https://www.researchgate.net/publication/360629859_A_New_Global_Ionospheric_Electron_Density_Model_Based_on_Grid_Modeling_Method
https://donnees-data.asc-csa.gc.ca/dataset/98466021-2q1w-5g2d-677zwru214wx68
https://agupubs.onlinelibrary.wiley.com/doi/10.1002/rds.20066
https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/97RS03183
https://www.sciencedirect.com/science/article/abs/pii/S0273117708006698?via%3Dihub
https://www.mdpi.com/2072-4292/12/15/2373/htm