Irkutsk, Russian Federation
Irkutsk, Russian Federation
Irkutsk, Russian Federation
Irkutsk, Russian Federation
Ekaterinburg, Russian Federation
Ionosphere has an important impact on the quality of radio communication, radar, and global positioning. One of the essential characteristics describing the state of the ionosphere is its critical frequency fₒF2. Its prediction provides effective modes of operation of technical radio equipment as well as enables calculation of the corrections needed to improve the accuracy of its functioning. Different physical and empirical models are generally used for fₒF2 prediction. This paper proposes an empirical prediction technique based on machine learning methods and observational history. It relies on a regression approach to the prediction based on the known daily quasi-periodicity of ionospheric parameters related to solar illumination. Algorithmically, this approach is implemented in the form of convolutional neural networks with two-dimensional convolution. The input data for the analysis is presented as two-dimensional solar time — date matrices. The model was trained on data from the mid-latitude ionosonde in Irkutsk (RF) and tested using data from several mid-latitude ionosondes: Arti (RF), Warsaw (Poland), Mohe (China). It is shown that the main contribution to the prediction value of fₒF2 is made by the data on the nearest few days before the prediction; the contribution of the remaining days strongly decreases. This model has the following forecast quality metrics (Pearson correlation coefficient 0.928, root mean square error 0.598 MHz, mean absolute error in percent 10.45 %, coefficient of determination 0.861) and can be applied to fₒF2 forecast in middle latitudes.
ionosphere, machine learning, neural networks, fₒF2
1. Abadi M., Barham P., Chen J., Chen Z., Davis A., Dean J., Devin M., Ghemawat S. TensorFlow: A system for large-scale machine learning. Proc. OSDI. 2016, pp. 265-283. DOI: 10.5281/ zenodo.4724125.
2. Barkhatov N.A., Revunov S.E., Urjadov V.P. Artificial neural network technology for forecasting critical frequency of ionospheric layer F2. Proceedings of Higher Educational Institutions. Radiophysics. 2005, vol. 48, iss. 1, pp. 1-15. (In Russian).
3. Bilitza D., Mckinnell L.-A., Reinisch B., Fuller-Rowell T. The International Reference Ionosphere (IRI) today and in the future. J. Geodesy. 2011, vol. 85. DOI:https://doi.org/10.1007/s00190-010-0427-x.
4. Boulch A., Cherrier N., Castaings T. Ionospheric activity prediction using convolutional recurrent neural networks. 2018. DOI:https://doi.org/10.48550/arXiv.1810.13273.
5. Breiman L. Bagging Predictors. Technical Report. 1994, No. 421.
6. Bring J. How to standardize regression coefficients. The American Statistician. 1994, vol. 48, no. 3, pp. 209-213. DOI:https://doi.org/10.2307/2684719.
7. Consultative Committee on International Radio (CCIR) Atlas of Ionospheric Characteristics Report 340. International Telecommunication Union, Geneva, Switzerland, 1967.
8. Galkin I.A., Reinisch B., Vesnin A.M., Huang X. Assimilation of sparse continuous groundbased ionosonde data into IRI using NECTAR model morphing. The 1st URSI Atlantic Radio Science Conference (URSI AT-RASC). Las Palmas, 2015, pp. 1-8. DOI:https://doi.org/10.1109/URSI-AT-RASC.2015.7303112.
9. Goodfellow I., Bengio Y., Courville A. Deep Learning. MIT Press, 2016, 800 p.
10. Hargrivs J.K. Upper atmosphere and solar-terrestrial connections. Introduction to Physics of the Near-Earth Space Environment. Leningrad, Gidrometeoizdat, 1982, 351 p. (In Russian).
11. Kingma D.P., Ba J.A. A Method for Stochastic Optimization. International Conference on Learning Representations. 2014. DOI:https://doi.org/10.48550/arXiv.1412.6980.
12. Lundberg S., Lee, S.-I. A Unified Approach to Interpreting Model Predictions. arXiv. 2017. DOI:https://doi.org/10.48550/ARXIV. 1705.07874.
13. Opitz D., Maclin R. Popular ensemble methods: An empirical study. J. Artificial Intelligence Res. 1999, vol. 11, pp. 169-198. DOI:https://doi.org/10.1613/jair.614.
14. Ratovsky K.G., Potekhin A.P., Medvedev A.V., Kurkin V.I. Modern Digital Ionosonde DPS-4 and its capabilities. Solar-Terr. Phys. 2004, iss. 5, pp. 102-104. (In Russian).
15. Rush C., Fox M., Bilitza D., Davies K., McNamara L., Stewart F., PoKempner M. Ionospheric mapping - an update of foF2 coefficients. Telecommun. J. 1989, vol. 56, iss. 3, pp. 179-182.
16. Salimov B.G., Khmelnov A.E. Prediction of the critical frequency of the ionosphere foF2 using a neural recurrent LSTM network. Proc. Conference “Lyapunovskie Chteniya”. ISDC SB RAS, Irkutsk, 2020, pp. 60-61. (In Russian).
17. Sivavaraprasad G., Lakshmi Mallika I., Sivakrishna K., Venkata Ratnam D. A novel hybrid machine learning model to forecast ionospheric TEC over low-latitude GNSS stations. Adv. Space Res. 2022, vol. 69, iss. 3, pp. 1366-1379. DOI:https://doi.org/10.1016/j.asr.2021.11.033.
18. Smirnov V.F., Stepanov A.E. New capabilities in studies of the high-latitude ionosphere: DPS-4 digisonde first results on measurements of localization and dynamics of large-scale ionospheric structures in Yakutsk. Solar-Terr. Phys. 2004, no. 5 (118), pp. 105-106. (In Russian).
19. Yu S., Ma J. Deep learning for geophysics: Current and future trends. Rev. Geophys. 2021, vol. 59, iss. 3, e2021RG000742. DOI:https://doi.org/10.1029/2021RG000742.
20. URL: http://irimodel.org/IRI-2016 (accessed June 27, 2019).
21. URL: https://ckp-rf.ru/catalog/ckp/3056 (accessed October 3, 2018).
22. URL: https://omniweb.gsfc.nasa.gov/form/dx1.html (accessed August 31, 2020).
