Machine Learning Modeling of Earth's Plasma Sheet Using Multi‐Spacecraft Observations

Accurate modeling of Earth's plasma sheet is essential for global simulations and understanding magnetospheric dynamics, yet empirical models often fail to capture its complex spatial structure. Leveraging decades of in situ observations from Geotail and the Magnetospheric Multiscale mission, we employ gradient boosting and neural networks, benchmarked against linear regression and empirical baselines, to model plasma sheet ion density, temperature, and ion‐to‐electron temperature ratios. Our findings reveal that predictive accuracy is highly sensitive to data set curation, particularly the classification of the plasma sheet region and the accuracy of the plasma moments used for training. The presented data‐driven models successfully reproduce global asymmetries that empirical baselines miss, including dawnside density enhancements, duskward ion temperature elevations, and variations driven by charge‐dependent drifts. To test generalization, we evaluate the models against Time History of Events and Macroscale Interactions During Substorms observations during a severe geomagnetic storm and demonstrate that incorporating high‐density intervals, often removed as outliers, enables the model to capture storm‐time plasma sheet density enhancements, reducing the mean absolute error by ∼30% relative to the empirical baseline for this event. The remaining discrepancies arise primarily from the sparsity of extreme events in historical records, suggesting that hybrid physics‐based and data‐driven modeling is a promising path toward improved storm‐time prediction.

Authors

Institutions

Publication Details

Journal
Journal of Geophysical Research Machine Learning and Computation
Published
2026-11-19
DOI
https://doi.org/10.1029/2026jh001589
Primary Topic
Ionosphere and magnetosphere dynamics
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Machine Learning Modeling of Earth's Plasma Sheet Using Multi‐Spacecraft Observations

V. G. Merkin, Simon Wing, Kareem Sorathia, Connor O'Brien et al.
Journal of Geophysical Research Machine Learning and Computation
Ionosphere and magnetosphere dynamics
article

Machine Learning Modeling of Earth's Plasma Sheet Using Multi‐Spacecraft Observations

V. G. Merkin, Simon Wing, Kareem Sorathia, Connor O'Brien, S. Ohtani, Louis Richard, Savvas Raptis, A. P. Devanandan
article en

Abstract

Accurate modeling of Earth's plasma sheet is essential for global simulations and understanding magnetospheric dynamics, yet empirical models often fail to capture its complex spatial structure. Leveraging decades of in situ observations from Geotail and the Magnetospheric Multiscale mission, we employ gradient boosting and neural networks, benchmarked against linear regression and empirical baselines, to model plasma sheet ion density, temperature, and ion‐to‐electron temperature ratios. Our findings reveal that predictive accuracy is highly sensitive to data set curation, particularly the classification of the plasma sheet region and the accuracy of the plasma moments used for training. The presented data‐driven models successfully reproduce global asymmetries that empirical baselines miss, including dawnside density enhancements, duskward ion temperature elevations, and variations driven by charge‐dependent drifts. To test generalization, we evaluate the models against Time History of Events and Macroscale Interactions During Substorms observations during a severe geomagnetic storm and demonstrate that incorporating high‐density intervals, often removed as outliers, enables the model to capture storm‐time plasma sheet density enhancements, reducing the mean absolute error by ∼30% relative to the empirical baseline for this event. The remaining discrepancies arise primarily from the sparsity of extreme events in historical records, suggesting that hybrid physics‐based and data‐driven modeling is a promising path toward improved storm‐time prediction.

Journal of Geophysical Research Machine Learning and ComputationVol. 3(6)
Swedish Institute of Space Physics (SE), University of New Hampshire (US), Johns Hopkins University Applied Physics Laboratory (US), University of Oxford (GB)
Ionosphere and magnetosphere dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon