Scholarlink

<p>Equatorial plasma bubbles (EPBs) are ionospheric irregularities that degrade Global Navigation Satellite System (GNSS) signal quality through rapid total electron content (TEC) fluctuations and scintillation. This study presents a novel approach for EPB detection with Doppler Index and classification using an Xtreme Gradient Boosting (XGBoost) model based on multi-feature GNSS parameters. Doppler index (DI), vertical TEC (VTEC), elevation angle, timestamp and Ionospheric Pierce Point (IPP) were extracted from GNSS observations at the low-latitude IGS GNSS station HYDE during 12 geomagnetically active days spread across 2022, 2023 and 2025. Class imbalance was addressed using SMOTE and SMOTE-Edited Nearest Neighbors (SMOTEENN). To test the robustness, EPB events over the years 2015, 2023, 2024 and 2025 were considered, excluding the days from the train dataset. Results indicate that XGBoost with SMOTEENN has an accuracy of 96.67% and an improved minority-class F1-score of 0.637, outperforming XGBoost + SMOTEENN with single-parameter ROTI (Rate of TEC Index) (0.510) and DI (0.512) models. Validation using DI-ROTI relationships and geomagnetic indices confirms the physical consistency of the GNSS EPB classification results.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine learning based classification of ionospheric equatorial plasma bubbles using GNSS-derived doppler index

  • 期刊论文

  • 发表日期：

    2026年5月2日

查看全文
  • Bandaru Srilakshmi Sowmya,
  • Supraja Reddy Ammana,
  • Manthati Ipsitha,
  • Vinodh Kumar Minchula,
  • Venkata Ratnam Devanaboyina

摘要

Equatorial plasma bubbles (EPBs) are ionospheric irregularities that degrade Global Navigation Satellite System (GNSS) signal quality through rapid total electron content (TEC) fluctuations and scintillation. This study presents a novel approach for EPB detection with Doppler Index and classification using an Xtreme Gradient Boosting (XGBoost) model based on multi-feature GNSS parameters. Doppler index (DI), vertical TEC (VTEC), elevation angle, timestamp and Ionospheric Pierce Point (IPP) were extracted from GNSS observations at the low-latitude IGS GNSS station HYDE during 12 geomagnetically active days spread across 2022, 2023 and 2025. Class imbalance was addressed using SMOTE and SMOTE-Edited Nearest Neighbors (SMOTEENN). To test the robustness, EPB events over the years 2015, 2023, 2024 and 2025 were considered, excluding the days from the train dataset. Results indicate that XGBoost with SMOTEENN has an accuracy of 96.67% and an improved minority-class F1-score of 0.637, outperforming XGBoost + SMOTEENN with single-parameter ROTI (Rate of TEC Index) (0.510) and DI (0.512) models. Validation using DI-ROTI relationships and geomagnetic indices confirms the physical consistency of the GNSS EPB classification results.