HSI

Predicting GOES X-ray Flux from STIX Data

Developing a machine learning pipeline to forecast solar flare signatures observed by GOES, using STIX spectrogram data from Solar Orbiter.

Basic domain knowledge

To understand this project, it helps to know a few basics about the Sun and how we observe it. The Sun often releases sudden bursts of energy called solar flares, which send out radiation across space. Satellites measure these flares from different positions: for example, the GOES satellites near Earth record the X-ray radiation that reaches us, while the Solar Orbiter spacecraft carries the STIX instrument to capture detailed X-ray signals closer to the Sun. By comparing these two perspectives, scientists can better understand how solar activity develops and how it affects space weather around Earth.

Initial situation

The Solar Orbiter mission provides high-resolution X-ray spectrograms through its STIX instrument. At the same time, the GOES satellites continuously measure X-ray flux at Earth. While both datasets are crucial for understanding solar flares, they differ in resolution, energy ranges, and observing positions. A robust method for aligning and relating these observations was missing.

Problem statement / Project goal

The project aimed to explore whether STIX data could be used to predict the GOES X-ray flux profile. This required aligning datasets across different instruments, normalizing energy channels, correcting for light-travel time differences, and designing machine learning models capable of handling sequential flare data. The minimum objective was to normalize and structure the data; the maximum objective was to build a working predictive model.

System architecture diagram
Data visualization STIX vs GOES
Training curve example
Model vs GOES comparison

Solution developed and its benefits

We built a modular pipeline consisting of data import, normalization, alignment, and model training. GOES data were retrieved from the NOAA archive and matched to STIX flares based on timestamps. To improve precision, distances between the Sun, Earth, and Solar Orbiter were considered, and a time-delay correction was applied. The normalized data were stored in a database and made available through an API, ensuring flexibility and scalability.

  • Automated data import from NOAA and STIX flare lists
  • Normalization across different energy ranges
  • Light-travel time correction between Earth and Solar Orbiter
  • Training a MLP neural networks
  • Evaluation using RMSE and correlation metrics

While the prediction error margin remains relatively large, the project demonstrates the feasibility of using STIX data as a basis for forecasting GOES flux curves. The modular architecture allows easy extension and integration into existing infrastructures.

Outlook

The second part of this project will be carried out in Spring 2026, where the focus will be on refining the machine learning models and extending the analysis to a larger dataset. This will provide deeper insights into the potential of combining multi-satellite observations with AI techniques for improved space weather forecasting.

Key terms

  • Programming languagesPython
  • Machine LearningMultilayer Perceptron (MLP)
  • DatabaseMongoDB
  • Data SourcesNOAA GOES Archive, Solar Orbiter STIX

Customer

Internal research project
Conducted at FHNW as part of the Bachelor Thesis module.

Team

Alain Burger

Timon van den Nobelen

Advisors

Samuel Krucker
samuel.krucker@fhnw.ch

Paolo Massa
paolo.massa@fhnw.ch