The direct interaction between large-scale interplanetary disturbances emitted from the Sun and the Earth's magnetosphere can lead to geomagnetic storms representing the most severe space weather events. In general, the geomagnetic activity is measured by the Dst index. Consequently, its accurate prediction represents one of the main subjects in space weather studies. In this scenario, we try to predict the Dst index during quiet and disturbed geomagnetic conditions using the interplanetary magnetic field and the solar wind parameters. To accomplish this task, we analyzed the response of a newly developed neural network using interplanetary parameters as inputs. We strongly demonstrated that the training procedure strictly changes the capability of giving correct forecasting of stormy and disturbed geomagnetic periods. Indeed, the strategy proposed for creating datasets for training and validation plays a fundamental role in guaranteeing good performances of the proposed neural network architecture.
Prominence of the training data preparation in geomagnetic storm prediction using deep neural networks
Cristoforetti, M;Gobbi, A;
2022-01-01
Abstract
The direct interaction between large-scale interplanetary disturbances emitted from the Sun and the Earth's magnetosphere can lead to geomagnetic storms representing the most severe space weather events. In general, the geomagnetic activity is measured by the Dst index. Consequently, its accurate prediction represents one of the main subjects in space weather studies. In this scenario, we try to predict the Dst index during quiet and disturbed geomagnetic conditions using the interplanetary magnetic field and the solar wind parameters. To accomplish this task, we analyzed the response of a newly developed neural network using interplanetary parameters as inputs. We strongly demonstrated that the training procedure strictly changes the capability of giving correct forecasting of stormy and disturbed geomagnetic periods. Indeed, the strategy proposed for creating datasets for training and validation plays a fundamental role in guaranteeing good performances of the proposed neural network architecture.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.