Machine Learning Modeling of Narrow SAID‐Like Spikes in Duskside Subauroral Flows From 11‐Year DMSP Observations
Space Weather · 2026-11 · 24 (11)
DOI: 10.1029/2026sw005299 ↗Kurz gefasst
Fachartikel in der Zeitschrift Space Weather, online erschienen am 06.11.2026. Verfasst von Haobo Hu und 6 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
Subauroral ion drifts (SAID) are narrow, intense westward plasma flows. However, their narrow latitudinal distribution and possible overlap with broader subauroral flows pose significant challenges for accurate modeling. This study proposes a dual‐modal model that takes magnetic latitude, magnetic local time, day of year, solar F10.7 index, and geomagnetic AE, Kp, and SYM‐H indices as inputs, and predicts SAID‐like westward‐flow velocities across subauroral latitudes in the afternoon‐to‐dusk sector. The model utilizes 11 years (2009–2019) of defense meteorological satellite program observations, with all 2016 events withheld for independent testing. The architecture employs a learned dynamic gating network to fuse an extreme gradient boosting (XGBoost) model, which provides a stable background oriented velocity estimate, with a Fourier‐feature multi‐layer perceptron (MLP) model designed to capture localized velocity spikes. Evaluation results show that the dual‐modal model yields a mean absolute error of 431 m/s for high‐speed samples (>1,000 m/s) on the 2016 test set, lower than standalone XGBoost (512 m/s), Fourier MLP (504 m/s), spatial CNN (792 m/s), and long short‐term memory (LSTM) (925 m/s). For lower‐speed samples, the LSTM yields the lowest errors, indicating that no architecture is uniformly optimal. Parametric dependence analysis demonstrates that the modeled event‐level peak velocities reproduce the principal observed variations with location, season, F10.7, AE, Kp, and SYM‐H, although the modeled mean peaks are generally lower. The model provides an empirical representation of narrow, high‐speed duskside subauroral flows and a candidate input for future Ionosphere‐Thermosphere coupling studies.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Hu, H., Pan, X., Li, S., Liu, J., Liu, X., Jiao, Y., & Cao, T. (2026). Machine Learning Modeling of Narrow SAID‐Like Spikes in Duskside Subauroral Flows From 11‐Year DMSP Observations. Space Weather, 24(11). https://doi.org/10.1029/2026sw005299
Vancouver
Hu H, Pan X, Li S, Liu J, Liu X, Jiao Y, et al. Machine Learning Modeling of Narrow SAID‐Like Spikes in Duskside Subauroral Flows From 11‐Year DMSP Observations. Space Weather. 2026;24(11). doi:10.1029/2026sw005299
BibTeX
@article{Hu2026,
author = {Hu, Haobo and Pan, Xinshuang and Li, Shuhan and Liu, Jing and Liu, Xuanqing and Jiao, Yu and Cao, Tianyu},
title = {Machine Learning Modeling of Narrow SAID‐Like Spikes in Duskside Subauroral Flows From 11‐Year DMSP Observations},
journal = {Space Weather},
year = {2026},
volume = {24},
number = {11},
doi = {10.1029/2026sw005299}
}
EUVIMED