Abstracts Details


First Name * : Ishita
Last Name * : Gulati
Affiliation * : Newcastle University
Abstract Type * : Poster
Title * : Using Recursive Neural Network (RNN) for Detection of Geomagnetic Storms in the EUREF Network
Author(s) * : Ishita Gulati, Federico Angelini, Handong Li, Scott Stainton, Martin Johnston and Satnam Dlay
Abstract Session * : Solar flares and coronal mass ejections
Abstract * : Deep inside the Sun’s corona, begins the solar activity, which emits electromagnetic radiation in the form of radio waves and is capable of producing magnetized plasma clouds and energetic particles. The turbulence in the fluid motion at various surfaces on the Sun, called the sunspots, release sudden and intense magnetic energies known as solar flares. These flares and their associated Coronal Mass Ejections (CMEs) make their way through the outer layers of the Sun, passing through the Interplanetary Magnetic Field (IMF), finally inflowing towards the Earth’s ionosphere, magnetosphere, and the atmosphere. This endures the Global Navigation Satellite Systems (GNSS) and various technologies associated with it. Current research in space science focuses on the accurate prediction of solar events to alarm users in advance of the incoming storms. This prediction accuracy, however, is not available at present and substantial research is being undertaken to make this possible. In this paper, we try to achieve the prediction-based anomaly detection as prediction is highly dependent on the reliable detection which can then be used for the classification of geomagnetic storms. As we approach the end of the current solar cycle SC-24, it is critical to understand and analyze the trends in the previous solar cycles, or the periods of high activity and their causes in the current/last solar cycle. This paper presents a comparative study between the actual experimental data-set and that obtained by using the RNN model which will then form a base for future predictive modelling and analysis. This can be achieved by providing geomagnetic indices such as Kp, speed of the solar wind and Disturbance Storm Time (Dst), as time sequences into the Recurrent Neural Network (RNN). The recursive nature of the RNN can be used to model the information dynamically, which makes it a good choice for this study. For instance, the Long Short-Term Memory (LSTM) architecture, well known for its performance in modelling multivariate temporal sequences is used to extract the features to classify the nature of the storm as quiet, moderate and/or high-activity period. This input data at varying spatial and temporal instances during the period of strong geomagnetic activity can be used to detect the intensity of storms, by analysing behavioural aspects of the training data. Once the intensity of the storm has been detected, it can be used to predict the Slant and Vertical Total Electron Content (STEC) and (VTEC) values, by applying the ionospheric thin shell model at 350 km above the Earth’s surface, which can be used to forecast the occurrence of next storm activity.