EGU2020-1848
https://doi.org/10.5194/egusphere-egu2020-1848
EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Neural network applications in geomagnetic storm prognosis based on the pre-storm occurrence of magnetic islands in the solar wind

Mikhail Fridman
Mikhail Fridman
  • Higher School of Economics, Laboratory of Complex Systems Modeling and Control, Moscow, Russian Federation (lapinbank@gmail.com)

So far, the problem of a short-term forecast of geomagnetic storms can be considered as solved. Meanwhile, mid-term prognoses of geomagnetic storms with an advance time from 3 hours to 3 days are still unsuccessful (see  https://www.swpc.noaa.gov/sites/default/files/images/u30/Max%20Kp%20and%20GPRA.pdf).

 This fact suggests a necessity of looking for specific processes in the solar wind preceding geomagnetic storms. Knowing that magnetic cavities filled with magnetic islands and current sheets are formed in front of high-speed streams of any type (Khabarova et al., 2015, 2016, 2018; Adhikari et al., 2019), we have performed an analysis of the corresponding ULF variations in the solar wind density observed at the Earth's orbit from hours to days before the arrival of a geoeffective stream or flow. The fact of the occurrence of ULF-precursors of geomagnetic storms was noticed a long time ago (Khabarova 2007; Khabarova & Yermolaev, 2007) and related prognostic methods were recently developed (Kogai et al. 2019), while the problem of automatization of the prognosis remained unsolved.

 A new geomagnetic storm forecast method, which employs a Recurrent Neural Network (RNN) for an automatic pattern search, is proposed. An ability of self-teaching and extracting deeply hidden non-linear patterns is the main advantage of Deep Neural Networks (DNNs) with multiple layers over traditional Machine Learning methods. We show a success of the RNN method, using either the unprocessed solar wind density data or Wavelet analysis coefficients as the input parameter for a DNN to perform an automatic mid-term prognosis of geomagnetic storms.  

Adhikari, L., et al. 2019, The Role of Magnetic Reconnection–associated Processes in Local Particle Acceleration in the Solar Wind, ApJ, 873, 1, 72, https://doi.org/10.3847/1538-4357/ab05c6
Kogai T.G. et al., Pre-storm ULF variations in the solar wind density and interplanetary magnetic field as key parameters to build a mid-term prognosis of geomagnetic storms. “GRINGAUZ 100: PLASMA IN THE SOLAR SYSTEM”, IKI RAS, Moscow, June 13–15, 2018, 140-143, ISBN 978-5-00015-043-6. https://www.researchgate.net/publication/327781146_Pre-storm_ULF_variations_in_the_solar_wind_density_and_interplanetary_magnetic_field_as_key_parameters_to_build_a_mid-term_prognosis_of_geomagnetic_storms
 Khabarova O. V., et al. 2018,  Re-acceleration of energetic particles in large-scale heliospheric magnetic cavities, Proceedings of the IAU, 76-82, https://doi.org/10.1017/S1743921318000285 
Khabarova O.V., et al. Small-scale magnetic islands in the solar wind and their role in particle acceleration. II. Particle energization inside magnetically confined cavities. 2016, ApJ, 827, 122, http://iopscience.iop.org/article/10.3847/0004-637X/827/2/122
Khabarova O., et al. Small-scale magnetic islands in the solar wind and their role in particle acceleration. 1. Dynamics of magnetic islands near the heliospheric current sheet. 2015, ApJ, 808, 181, https://doi.org/10.1088/0004-637X/808/2/181

Khabarova O.V., Current Problems of Magnetic Storm Prediction and Possible Ways of Their Solving. Sun&Geosphere,  http://sg.shao.az/v2n1/SG_v2_No1_2007-pp-33-38.pdf , 2(1), 33-38, 2007

Khabarova O.V. & Yu.I.Yermolaev, Solar wind parameters' behavior before and after magnetic storms, JASTP, 70, 2-4, 2008, 384-390, http://dx.doi.org/10.1016/j.jastp.2007.08.024

How to cite: Fridman, M.: Neural network applications in geomagnetic storm prognosis based on the pre-storm occurrence of magnetic islands in the solar wind, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-1848, https://doi.org/10.5194/egusphere-egu2020-1848, 2019