Scholarlink

<p>This manuscript is part of the Heliophysics Summer School Machine Learning Special Collection. Predicting solar flares remains a major challenge in space-weather forecasting. Large eruptive flares can trigger coronal mass ejections and energetic particle events that threaten satellites, endanger astronauts, and degrade High Frequency (HF) communications. Current flare alerts rely on Soft X-ray measurements from the X-ray Sensor onboard Geostationary Operational Environmental Satellite (GOES-XRS) crossing fixed flux thresholds, and therefore are issued only once the flare is already in its impulsive phase. This study focuses on solar-flare nowcasting using the Hot Onset Precursor Event (HOPE), a recently identified pre-flare phenomenon. In this study we extend existing HOPE trigger algorithms by applying a simple multilayer perceptron model to estimate flare peak magnitude and peak time. The method consists of detecting HOPE plasma temperature and emission measure signatures, and then using the time-histories of these variables to predict flare magnitude and time. The system is optimized to detect <InlineEquation ID="IEq1"> <EquationSource Format="MATHML"><math> <mrow> <mo>&gt;</mo> <mi mathvariant="normal">M</mi> <mn>5.0</mn> </mrow> </math></EquationSource> <EquationSource Format="TEX">$\mathrm{&gt;M5.0}$</EquationSource> </InlineEquation> flares. The model was trained using 180 flares from GOES-14 through GOES-19 XRS instrument dataset, and tested over 46 independent flares. A 1-week proof-of-concept validation was performed during which all four <InlineEquation ID="IEq2"> <EquationSource Format="MATHML"><math> <mrow> <mo>&gt;</mo> <mi mathvariant="normal">M</mi> <mn>5.0</mn> </mrow> </math></EquationSource> <EquationSource Format="TEX">$\mathrm{&gt;M5.0}$</EquationSource> </InlineEquation> flares were successfully detected, with an average lead time of 17.9 minutes relative to the NOAA R3 radio-blackout alert. These results show that HOPE-based machine learning models can deliver early warnings for high-impact flares.</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

Advancing Solar Flare Nowcasting with Machine Learning Based Detection of Hot Onset Precursor Events

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月22日

查看全文
  • Anant Telikicherla,
  • Jessica Hamilton,
  • A. A. Shmies,
  • Malik H. Walker,
  • Fatima Yousuf

摘要

This manuscript is part of the Heliophysics Summer School Machine Learning Special Collection. Predicting solar flares remains a major challenge in space-weather forecasting. Large eruptive flares can trigger coronal mass ejections and energetic particle events that threaten satellites, endanger astronauts, and degrade High Frequency (HF) communications. Current flare alerts rely on Soft X-ray measurements from the X-ray Sensor onboard Geostationary Operational Environmental Satellite (GOES-XRS) crossing fixed flux thresholds, and therefore are issued only once the flare is already in its impulsive phase. This study focuses on solar-flare nowcasting using the Hot Onset Precursor Event (HOPE), a recently identified pre-flare phenomenon. In this study we extend existing HOPE trigger algorithms by applying a simple multilayer perceptron model to estimate flare peak magnitude and peak time. The method consists of detecting HOPE plasma temperature and emission measure signatures, and then using the time-histories of these variables to predict flare magnitude and time. The system is optimized to detect > M 5.0 $\mathrm{>M5.0}$ flares. The model was trained using 180 flares from GOES-14 through GOES-19 XRS instrument dataset, and tested over 46 independent flares. A 1-week proof-of-concept validation was performed during which all four > M 5.0 $\mathrm{>M5.0}$ flares were successfully detected, with an average lead time of 17.9 minutes relative to the NOAA R3 radio-blackout alert. These results show that HOPE-based machine learning models can deliver early warnings for high-impact flares.