Predicting Equatorial Spread F at JICAMARCA Sector Via Supervised Machine Learning

Abstract The nighttime equatorial spread‐F (ESF) occurrence is essential to cause disturbances of the trans‐ionospheric radio wave signals. Previous research has successfully modeled the probability or growth rate of ESF, but remains deficient in distinguishing their typology and duration, which are...

Full description

Saved in:
Bibliographic Details
Main Authors: Shunzu Gao, Chao Xiong, Hongtao Cai, Qian Pan, Weijia Zhan, Hong Zhang, Yuhao Zheng
Format: Article
Language:English
Published: Wiley 2025-03-01
Series:Space Weather
Subjects:
Online Access:https://doi.org/10.1029/2024SW004214
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract The nighttime equatorial spread‐F (ESF) occurrence is essential to cause disturbances of the trans‐ionospheric radio wave signals. Previous research has successfully modeled the probability or growth rate of ESF, but remains deficient in distinguishing their typology and duration, which are also critical for predicting their influences on radio wave propagation. In this paper, an ensemble learning approach is proposed to predict the occurrence and typology of ESF, based on over one solar cycle observations of coherent scattering radar (CSR) located at Jicamarca. The training data set cover from 2000 to 2016, and the observations in 2017 are chosen as testing data set. The model inputs include local time (LT), day of year (DoY), height, solar flux index (P10.7), geomagnetic activity related indices (Kp, SYMH, AE), the true height of the ionospheric F layer (HF), the vertical plasma drift velocity of the ionospheric F layer, and the occurrence rate of ESF from the previous 3 hr. By using the Shapley Additive exPlanations (SHAP) tool analysis, it shows that the LT is the most influential parameter, and the P10.7 contributes more than the geomagnetic indices, to predict the occurrence of ESF. An important advantage of our model is that it can present different types of ESF by reconstructing their duration versus height distributions. Comparison results between prediction and observation suggest that the model predicts all bins with an accuracy reaching 81.2%, and it is able to predict the daily occurrence of ESF with accuracy of about 90.1%.
ISSN:1542-7390