Bulletin of the American Physical Society
66th Annual Meeting of the APS Division of Plasma Physics
Monday–Friday, October 7–11, 2024; Atlanta, Georgia
Session JO05: Space Plasmas
2:00 PM–5:00 PM,
Tuesday, October 8, 2024
Hyatt Regency
Room: Hanover C
Chair: Emily Lichko, University of Chicago
Abstract: JO05.00007 : Iterated vs One-Step optimized system science models for Dst Index with Neural Networks*
3:12 PM–3:24 PM
Presenter:
Maria J Quezada Roco
(Universidad de Chile)
Authors:
Maria J Quezada Roco
(Universidad de Chile)
Juan A Valdivia
(Universidad de Chile)
Jose Antonio Rogan
(Universidad de Chile)
Max Ramirez
(Universidad de Chile)
Sylvain Blunier
(Eso)
To characterize the dynamics within the solar wind-magnetosphere-ionosphere system, various geomagnetic indices (GI) have been developed. One notable index, SYM-H, focuses on evaluating the equatorial component in the eastern region to assess ground-level effects of the ring current with high temporal resolution. Analysis of these GI time series reveals the magnetosphere’s variable response to solar wind variations. In this context, machine learning approaches, designed specifically for system science discovery, play a critical role. Our approach builds upon strategies like those in [1], incorporating identified Robust Solar Wind Drivers.
Our methodology involves using diverse seeds to initiate neural network training and employing different initializers in each layer of the NN, fostering diverse model behaviors. Through comparative analysis of prediction strategies—both one-step and iterative—we gain insights into how different architectural designs adapt during various storm phases.
Additional observations suggest the magnetosphere functions as a dynamically driven multiscale system with interconnected subsystems. Accurate modeling requires a comprehensive understanding of subsystem coupling. Our method proposes critical considerations for developing forecasting models for these coupled subsystems. It also introduces an interdisciplinary approach that sheds light on neural network dynamics.
*We acknowledge the support from ANID project 21242519.We extend our appreciation for the parched support received from FONDECYT project 1240655/1240663/1240697.
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700