Research PaperNeural network classification of substorm geomagnetic activity caused by solar wind magnetic clouds
Introduction
The application of advanced neural network technology to fundamental problems of heliogeophysics allows us to determine the cause-effect relationships between dynamics of magnetosphere and interplanetary medium parameters. As previously described by Barkhatov et al., [2017, 2018, 2019a,b,c,d,e,f], the studied AL index dynamics containing information on the characteristics of magnetic clouds (MC) in which the magnetosphere is plunged. The Solar wind plasma and interplanetary magnetic field provide the sources of energy for magnetospheric substorms. The IMF Bz component is conventionally considered to be the most effective for generating substorms. The Solar wind velocity (V, km/s) and plasma density (N, cm−3) vary insignificantly on the substorm time scales; therefore, they are usually not considered as independent energy sources for substorm activities. Resent studies by Barkhatov et al. (2017); Vorobjev et al. (2018a); Vorobjev et al. (2018b) show that accumulation of the Solar wind kinetic energy impacts substorm formation. The neural network technology successfully solves the problems of cause-effect relationships between multiple parameters [Barkhatov et al., 2002; Barkhatov et al., 2005; Barkhatov et al., 2006; Barkhatov, 2013; Barkhatov et al., 2014a,b; Barkhatov et al., 2019a,b,c,d,e,f; Manakova Yu. et al., 2016]. In Barkhatov et al., [2017] the application of Elman intellectual neural network successfully recovered the variations of AL index of magnetic activity for isolated substorms. It was shown that the recovery of the AL index is most efficient when the integral parameter Σ[NV2] is included in the input sequences. This parameter takes into account the preceding pumping of Solar wind kinetic energy into the magnetosphere.
In the present study we applied the neural network classification to analyze the high-latitude geomagnetic activity, which is formed as result of the interplanetary magnetic cloud body interacts with the Earth's magnetosphere. The IMF integral components are used as the input parameters for the ANN as their effectiveness has been previously confirmed. We utilize Kohonen ANN, which implements machine vision algorithms for self-learning similar to popular projects, such as Google Deep Mind (https://deepmind.com/), or Microsoft Project Brainwave (https://www.microsoft.com/en-us/research/project/project-brainwave/).
The study aims to classify the events; each class includes both the characteristic parameters of interplanetary magnetic clouds and dynamics of substorm activity. This classification can be used to refine the models describing the impact of Solar plasma flows from specific sources on the Earth's magnetosphere.
Section snippets
Data used and processing algorithm
We reviewed observations of thirty-three interplanetary magnetic clouds (IMC) between 1998 and 2012 (see Table in [Barkhatov et al., 2019a,b,c,d,e,f]). The Solar wind parameters were analyzed for each IMC observation: plasma density (N), velocity (V), and vector B components (Bx, By, Bz) of the interplanetary magnetic field (IMF) in the GSM coordinate system, as well as the magnetic activity indices Dst and AL. Data with 1 min resolution were obtained from http://cdaweb.gsfc.nasa.gov. The
Determination of the optimal number of classes
The most logical and simple way to classify the substorm events is by intensity of magnetic activity in the auroral zone. Here, we determined the optimal number of classes using one parameter No. 14 (intensity of AL index) for all 33 events by fast learning method. In case of fast learning, each event is offered to ANN only once while the conventional training implies each event is presented to ANN several times. It can assume in advance that here should be expected at least 3 classes of
Classification of the cause-effect patterns of the events
The principal classification experiments in this study were carried out using combinations of various parameters. The first combination of parameters corresponds only to the causes of the events. It works with ANN No 1, classifies parameters related to IMC and provides the classes of causes. The second combination of parameters corresponds only to the consequences of the events. It works with ANN No 2, classifies parameters related to geomagnetic reaction of magnetosphere and provides the
Summary and conclusions
In the study, a classification of the cause-and-effect relationship of substorm activity with the characteristics of large-scale magnetic fluxes of the magnetic cloud type upon their impact on the Earth's magnetosphere is performed. For this, we used the data of auroral activity during the periods of the 33rd magnetic storms recorded from 1998 to 2012. Each of the considered storms contains from a few to several tens of substorms. The classification of the created causal relationship images is
Acknowledgement
This work was supported by the RFBR, grant No.18-35-00430 (E.A. Revunova, O.M. Barkhatova) and the Ministry of Education and Science of the Russian Federation, grant No.5.5898.2017/8.9 (N.A. Barkhatov, S.E. Revunov).
References (24)
Development of methods for predicting the geomagnetic state of the magnetosphere based on the search for fundamental laws of solar-terrestrial connections
Vestnik of Minin University
(2013)- et al.
Determination of magnetic cloud parameters and prediction of magnetic storm intensity
Geomagn. Aeron.
(2010) - et al.
The method of artificial neuron networks as a procedure for reconstructing gaps in records of individual magnetic observatories by the data from other stations
Geomagn. Aeron.
(2002) - et al.
Artificial neural network technique for predicting the critical frequency of the ionospheric F2 layer
Radiophys. Quantum Electron.
(2005) - et al.
Prediction of the maximum observed frequency of the ionospheric HF radio channel using the method of artificial neural networks
Geomagn. Aeron.
(2006) - et al.
The classification algorithm for MHD wavelet-skeleton spectral patterns of geoeffective plasma flows in the solar wind
Vestnik of Minin University
(2014) - et al.
Orientation of the Solar wind magnetic clouds affects the seasonal variation of geomagnetic activity
Cosmic Res.
(2014) - et al.
Dynamics of Solar wind parameters affects the formation of substorm activity
Geomagn. Aeron.
(2017) - et al.
Studying the relationship between high-latitude geomagnetic activity and parameters of interplanetary magnetic clouds with the use of artificial neural networks
Geomagn. Aeron.
(2018) - et al.
Establishing the orientation of shock wave plane of solar wind magnetic cloud for conclusions about the level of auroral substorm activity
JP J. Heat Mass Transf.
(June 2019)