Fast Dst computation by applying deep learning to Swarm satellite magnetic data

Abstract

Dst (Disturbance Storm Time) is an hourly index of magnetic activity computed from the measured intensity of the globally symmetrical equatorial electrojet (Ring Current) obtained by a series of near-equatorial geomagnetic observatories. We selected and trained an Artificial Neural Network (ANN) to give the estimation of the Dst index through the magnetic data measured by the Swarm three-satellite mission. From November 2014 to December 2019, we selected a balanced number of quiet and disturbed days, to get the most uniform set of Dst index values as possible. We then collected a big data collection of Swarm magnetic signals, confined to three very narrow belts of low-to-mid latitude: this choice allows it to better resemble the geographic distribution of the geomagnetic observatories contributing to the calculation of Dst. We also extended the analysis to mid latitude locations to increase the number of satellite samples. Once we determined by means of simulations the best network topology, we trained the network and tested its capabilities. The outcomes show that the ANN is able to give a reliable fast estimation of the Dst index directly from Swarm satellite magnetic data.