Dst index estimation from cosmic ray and solar wind data during solar cycle 24 using artificial neural networks

Abstract

This research employs artificial neural networks (ANNs) to estimate the disturbance storm time (Dst) index using cosmic ray (CR) and solar wind (SW) data from solar cycle 24. A dataset of 1361 events (2008–2019) was compiled and divided into training and testing subsets. Validation was performed using a dataset of 116 events from the year 2005. The flexibility and robustness of the model was further verified with three geomagnetic storms (GSs) representing different conditions during solar cycle 23. The CR parameters, SW parameters, and geomagnetic indices were sourced from the Forbush Effects and Interplanetary Disturbance (FEID) Catalogue and NASA's OMNI database. The ANN model utilized a feedforward neural network with Levenberg-Marquardt backpropagation, achieving optimal performance with a single hidden layer of six neurons, which minimized the root mean square error (RMSE). The model's overall correlation coefficient was 87%, indicating strong predictive capability. The results demonstrate the ANN model's efficacy in capturing complex interactions between solar and cosmic-ray phenomena and their impacts on geomagnetic activity, providing a robust tool for space weather studies. Additionally, the strong correlation observed between the Dst index and cosmic ray data suggests that incorporating these data in future studies could be highly valuable.