RG: Machine Learning

Chairs

Table 1: RG Chairs.

Name	Affiliation	Email
1. Hyunju Connor	NASA GSFC	[1]
2. Matthew Argall	UNH	[2]
3. Xiangning Chu	LASP, CU Boulder	[3]
4. Bashi Ferdousi	AFRL	[4]
5. Valluri Sai Gowtam	NASA GSFC/CUA	[5]

GEM/CEDAR 2025

The Machine Learning-based Geospace Environment Modeling (MLGEM) group is excited to announce two joint sessions at the upcoming CEDAR-GEM Joint Workshop. These sessions aim to bring together AI innovators and domain experts across the heliophysics community, spanning the Sun, solar wind, magnetosphere, ionosphere, thermosphere, mesosphere, and ground-based observations.

Session 1: General Contributions Session

This session welcomes contributions demonstrating the use of AI and machine learning in addressing a broad range of geospace and atmospheric science challenges.

Session 2: Storm Challenge Session

This special session invites diverse AI/ML-based studies of heliophysics systems—including, but not limited to, the Sun, solar wind, magnetosphere, ionosphere, thermosphere, mesosphere, and ground-based systems— focused on the following challenge storms:

1. January 4, 2023 (Minimum Sym-H: -74 nT at 09:04 UT)

2. May 6, 2023 (Minimum Sym-H: -108 nT at 05:11 UT)

3. May 11, 2024 (Minimum Sym-H: -518 nT at 02:14 UT)

The goal is to examine storm impacts from a system-of-systems perspective, identify gaps in current understanding and ML approaches, and explore future improvements. We encourage preliminary results and works-in-progress. While comparisons with observations are appreciated, they are not required for participation. These sessions aim to foster interdisciplinary collaboration between the GEM and CEDAR communities, share best practices, and discuss challenges and opportunities in applying machine learning to geospace and atmospheric modeling. We hope to generate lively discussions around the role of AI in advancing our understanding of geospace dynamics.

MLGEM: Session 1: 10:00am- 12:00pm on Wednesday, Jun 25, 2025 | Zoom link: https://cua.zoom.us/j/84478187920 Session 1: General Contributions Session - 06/25/2025 (Wed, 10:00 AM - 12:00 PM) Order Speaker Topic Area Title Introduction (2 min) 1 10:02-10:13 Daniel Da Silva Dayside Magnetosphere Data-Mining Similar Scenarios for Uncertainty Quantification of Solar Wind Predictions at L1 2 10:13-10:24 Gonzalo Cucho-Padin Dayside Magnetosphere Machine learning-based modeling of the Northern Earth's cusp 3 10:24-10:35 Connor O'Brien Dayside Magnetosphere Magnetosheath Control of the Cross Polar Cap Potential: Correcting for Measurement Uncertainty Using Machine Learning Online 4 10:35-10:46 Mikhail Sitnov Nightside Magnettosphere The 26 February 2008 substorms: Data-mining picture 5 10:46-10:57 Savvas Raptis Nightside Magnetosphere Characterizing Earth's Plasma Sheet through Multi-Spacecraft Observations and Machine Learning 6 10:57-11:08 Brendan Powers Nightside Magnetosphere Machine Learning Applications for Injection and Bursty Bulk Flow Identification 7 11:08-11:19 Chuanfei Dong Inner Magnetosphere Integration of Kinetic Effects in Multi-Moment Multi-Fluid Models through Machine Learning 8 11:19-11:30 Qusai Al Shidi Inner Magnetosphere Reduced Order Probabilistic Emulator of RAM-SCB: Orthogonal Autoencoders 9 11:30-11:41 Jonathan Mellina Inner Magnetosphere Investigating Tidal Effects in Earth's Plasmasphere Using a Machine Learning Model 10 11:41-11:52 Simon Mackovjak Ionosphere Automation of all-sky images classification 11 11:52-12:00 Gregory Riggs Ionosphere Statistical ULF Wave Latitude Distribution: A Key to Understanding Off-Equatorial Radiation Belt Electron Radial Diffusion Session 2: General Contributions (Cont.) + Storm Challenge Session - 06/25/2025 (Wed, 2:00 PM - 4:00 PM) https://cua.zoom.us/j/88222028213 Introduction (2 min) 1 2:02-2:14 Andong Hu Ground Long-Term and Short-Term Dst Forecasting of the Gannon Event Using Advanced ML Approach Invited Online 2 2:14-2:26 Lucas Jia Ground Imbalanced Regression Artificial Neural Network Model for Auroral Electrojet Indices (IRANNA): Can We Predict Strong Events? 3 2:26-2:38 Piyush Mehta Thermosphere Storm-Time Prediction of Thermospheric Density with Reduced Order Probabilistic Emulator Invited 4 2:38-2:49 Sai Gowtam Valluri (11 min) Ionosphere ML-based auroral particle precipitation and electrodynamics models 5 2:49-3:01 Jubyaid Uddin Ionosphere ML-AIM model for Calculating High Latitude Ionospheric Electrodynamics 6 3:01-3:13 Ben Martinez Ionosphere High Latitude Ionospheric Prediction and Forcing Deconvolution: A Tree-Based Machine Learning Approach using PFISR and SD-WACCM-X 7 3:13-3:25 Grant Stephens Nightside Magnetosphere Data mining reconstruction of extreme geomagnetic storms 8 3:25-3:37 Donglai Ma Inner Magnetosphere Machine Learning interpretation of radiation belt dynamics Invited 9 3:37-3:48 Xiangning Chu (11 min) Inner Magnetosphere Imbalance regression is critical for space weather 10 3:48-4:00 Evan McPherson Inner Magnetosphere Imbalanced Regressive Model of Electron Fluxes in the Earth's Outer Radiation Belt

mini-GEM 2024

The Machine learning-based Geospace Environment Modeling (MLGEM) session is scheduled for 3:30-5:00 p.m. on Sunday, December 8, during the mini-GEM workshop. We will present AI model results of various heliophysics systems for selected challenge storms (see below) and discuss strategies to integrate these models into an Artificial Intelligence Modeling Framework for Advancing Heliophysics Research (AIMFAHR).

CHALLENGE STORM LIST:

January 4, 2023 (minimum Sym-H :-74nT at 09:04UT)

May 6, 2023 (minimum Sym-H : -108 nT at 05:11 UT)

May 11, 2024 (minimum Sym-H : -518nT at 02:14 UT)

SESSION DETAILS:

Location: Rock Creek Salon B, the Westin Washington, D.C. Downtown (999 9th St NW, Washington D.C., 20001)

Virtual Meeting Link: https://cua.zoom.us/j/81902025946

Schedule (Sunday, December 8):

1. 3:30 – 3:40pm: Bayane Michotte de Welle – Magnetopause Reconnection Lines and Polar Cap Potentials

2. 3:40 – 3:50pm: Gonzalo Cucho-Padin – Cusp

3. 3:50 – 4:00pm: Mikhail Sitnov – Inner Magnetosphere Magnetic fields and Currents

4. 4:00 – 4:10pm: Brianna Isolar – Inner Magnetosphere Electric Fields

5. 4:10 – 4:20pm: Valluri Sai Gowtam – Global Aurora Precipitation

6. 4:20 – 4:30pm: Jubyaid Uddin – Field-Aligned Currents and Ionospheric Potentials

7. 4:30 – 4:40pm: Hongfan Chen – Geomagnetic Field Perturbation

8. 4:40 – 4:50pm: Raman Mukundan - Geomagnetic Field Perturbation

9. 4:50 – 5:05pm: Discussion on AI Modeling Framework for Advancing Heliophysics Research (AIMFAHR)

a. Review of results and suggestions for future challenge-storm studies.

b. Strategies for coupling AI models of different systems

c. Usefulness of AI models trained with physics-based simulation results (surrogate models) in AIMFAHR.

d. Open discussion on additional topics.

About the ML-GEM Resource Group

Machine learning-based Geospace Environment Modeling (ML-GEM) is a new resource group selected by the GEM Steering Committee. Its two primary goals are advancing system-of-systems science in Sun-Earth interaction from a data-driven perspective and developing an ML-based Geospace Environment Modeling by integrating community-wide ML efforts.

Goals and Objectives

1. Invite community-wide ML efforts: Extend invitations to a diverse community, including GEM, CEDAR, SHINE, and ML communities, fostering broader participation and collaboration.

2. Share ML advancements: Disseminate the latest ML techniques, including their advantages and disadvantages, and lessons learned. This knowledge-sharing will support community-wide geospace environment modeling efforts.

3. Facilitate integration discussions: Stimulate discussions on integrating diverse ML models across geospace systems, such as solar wind, magnetosheath, cusp, ring current, radiation belt, plasmasphere, and ionosphere.

4. Develop ML-based models: Initiate the development of an ML-based geospace environment model and encourage the strategic creation and inclusion of ML components in ML-GEM for enhanced performance.

5. Explore systematic responses: Investigate system-of-systems science in the solar wind – Earth interaction by leveraging ML-GEM and other ML models.

6. Compile a catalog of ML-ready dataset: Cleaning and preparing datasets constitute a substantial portion of machine learning model development. We will curate a list of existing ML-ready datasets in heliophysics for various scientific purposes.

GEM-2024 Activities

ML-GEM chairs have scheduled 4 sessions for the upcoming 2024 GEM Summer workshop held in Fort Collins, Colorado during June 23-28.

1. ML-GEM stand-alone session (10:30am - 12:00pm on 06/27 Thursday): All ML efforts across the GEM research areas are invited.

Zoom Link: https://unh.zoom.us/j/92641406400?pwd=ghwL2kuHUr4ZUCcZZVd8rVFk1Q5XGn.1, Password: 452277.

8min/presentation. We recommend 6min presentation, 1min question, and 1min transition.

ML-GEM Standalone Session Schedule: 10:30am - 12:00pm on Thursday, Jun 27

	Name	Affiliation	Title	Time	Topic
1	Mikhail Sitnov	APL/JHU	Data mining of the cislunar magnetotail	10:30	Tail/nightside
2	Savvas Raptis	APL/JHU	Modeling Earth's Plasma Sheet with Machine learning	10:38	Tail/nightside
3	Xiaofei Shi	UCLA	Data mining-based reconstruction of the transition region in the nightside magnetosphere validated through ELFIN isotropic boundaries	10:46	Tail/nightside
4	James Edmond/Jimmy Raeder	UNH	Clustering of Plasma Region Observations using Self-Organizing Maps	10:54	Dayside
5	Connor O'Brien	BU	PRIME-SH: A Data-Driven Probabilistic Model of Earth’s Magnetosheath	11:02	Dayside
6	Sai Gowtam Valluri	UAF	Machine Learning-based Particle Precipitation Model (ML-PPM)	11:10	Ionosphere
7	Matthew Blandin	UAF	Residual Convolutional Neural Networks for Global Geomagnetic Field Predictions	11:18	GIC
8	Jonathan Mellina	LASP	Tidal Effects in Earth's Plasmasphere Using a Machine Learning Model	11:26	Plasmasphere
9	Wen Li	BU	Integrate Deep Learning into Physics-Based Simulation: Modeling Global Electron Precipitation Driven by Whistler Mode Waves	11:34	Inner Mag
10	Evan McPherson	LASP	Imbalanced Regressive Model of Electron Fluxes in the Earth's Outer Radiation Belt	11:42	Inner Mag
11	Xiangning Chu	LASP	Imbalanced regressive neural network model for whistler-mode hiss waves: spatial and temporal evolution	11:50	Inner Mag
12	Drew Turner	APL/JHU	Unsupervised clustering and a generative model for Earth's bow shock using MMS data	11:58	Dayside

2. ML-GEM joint session with the Inner Magnetosphere Focus Groups (3:30 - 05:00 PM on 06/27 Thursday): ML efforts particularly in the inner magnetosphere research area are invited. We will discuss how to align ML efforts with current science associated with the SCIMM/RB/CP Focus Groups. An outcome will be to identify an open question, pair ML and data analysts, and create a plan to pursue answers for the next GEM Workshop.

Zoom Link: https://unh.zoom.us/j/93264767597?pwd=tjaJrNendpMdMGP8aaYbRKrIGnCVbO.1. Password: 435443

The discussion-only session with inner MAG focus groups.

MLGEM-RB-SCIMM-CP Joint Session: 3:30 - 5:00pm on Thursday, Jun 27

	Name	Affiliation	Title	Time	Topic
1	FG/RG Chairs		Science questions for MLGEM, RB, SCIMM, and CP fields
2	Everyone		Discussion
			- Identify an open question, pair ML and data analysis, and create a plan to pursue answers for the next GEM Workshop.

3. ML-GEM discussion session (10:30am - 12:00pm on 06/28 Friday): We will showcase currently available AI models in the heliophysics community and discuss how to unify them into a single AI modeling framework for understanding and predicting the geospace environment.

Zoom Link: https://unh.zoom.us/j/99110890924?pwd=LsvPTw5LtrTwQVeprH8uQZCbLp4DEF.1. Password: 034895

The session will host three talks from the standalone session (10:30 to 10:54am) and then move to the discussion sessions (10:54am -12:00pm).

ML-GEM Standalone/Discussion Session: 10:30 - 12:00pm on Friday, Jun 28

	Name	Affiliation	Title	Time	Topic
1	Jacob Bortnik	UCLA	Using interpretable AI to discover the drivers of acceleration vs depletion radiation belt events	10:30	Inner Mag
2	Qusai Al Shidi	WVU	Reduced Order Probabilistic Emulator of RAM-SCB: Towards Non-linearity with Deep Learning	10:38	Inner Mag
3	Man Hua	UCLA	Machine‐learning based identification of the critical driving factors controlling storm‐time outer radiation belt electron flux dropouts	10:46	Inner Mag
4	Hyunju Connor	NASA/GSFC	Overview of ML-GEL goals, plan, and discussion topics	10:54
5	Everyone		Presentation of current ML models (2min/model) : Submit slide format [6]	11:00
			Discussion
			1. Integrate the ML models into a unified AI framework of Heliophysics Modeling
			2. Select a common testing dataset (year, storms, substorms) for the framework validation
			3. Determine action items for the next ML-GEM meeting

4. ML-GEM tutorial session (1:30 - 3:00pm on 06/28 Friday): A hands-on tutorial on the Long Short-Term Memory (LSTM) technique that models time-series data. This tutorial will use the LSTM model and the SuperMAG geomagnetic field data, published in Blandin et al. (2022; https://doi.org/10.3389/fspas.2022.846291).

Zoom Link: https://unh.zoom.us/j/97969090948?pwd=KRF3SEW7E5GELUn8DXozlvTYbKsvMa.1. Password: 074062.

ML-GEM Tutorial Session: 1:30 - 3:00pm on Friday, Jun 28

	Name	Affiliation	Title	Time	Topic
1	Hyunju Connor	NASA/GSFC	Long Short-TERM Memory Model of Alaska geomagnetic field variation	1:30
2	Matt Blandin	UAF	Introduction of the LSTM code	1:40
3	Everyone	-	Hands-on Activities: Follow the instructions below to get started.	1:50

Instructions for the ML-GEM Tutorial Session:

Welcome to the ML-GEM tutorial session! We will be using the Python programming language and the Google Colab platform for our exercises. There’s no need to download or install anything. Just follow these simple steps:

Step 1: Download the tutorial Jupyter notebook and the data files via the provided link:https://drive.google.com/drive/u/1/folders/1CflXlSi5KZ4Iy0f4MkWRrR_j8t6mmLO1

Step 2: Sign in to your Google Drive account.

Step 3: Create a new folder named “LSTM_tutorial” in your Google Drive.

Step 4: Upload the Jupyter notebook and the data files into the “LSTM_tutorial” folder.

Step 5: Open the tutorial file by double-clicking on it. It will launch in the Google Colab platform.

Step 6: By default, the code will execute using the “CPU” option. To change this, locate the dropdown menu next to the “Connect” button in the top-right corner. Select “Change runtime type,” then choose the “T4 GPU” option.

You’re now fully prepared for the tutorial. If you run into any issues during the setup process, don’t hesitate to reach out. One of the co-chairs will be available to assist you. We’re excited to have you join us for the tutorial!