Why NASA and IBM made a new AI model that predicts the Sun's damaging flares

Why NASA and IBM made a new AI model that predicts the Sun's damaging flares

🎙 IBM Research 👥 120K 📅 August 20, 2025 ⏱ 18 min 👁 2K 📄 interview 🧭 2026-08-17
Available in: English (current) Français

Keywords

Suryasolar flaresfoundation modelNASAIBM

Summary

In this interview, Mike Murphy from IBM Research talks with Johannes Schmude, the technical project lead for the Surya model, a new AI foundation model for solar physics developed in collaboration with NASA. Schmude explains the motivation behind the model, which aims to predict solar flares and other solar events that can impact Earth’s technology and astronauts. He discusses the challenges of building a foundation model for a new scientific domain, including decisions about temporal scales, spatial resolution, and whether to use deterministic or probabilistic approaches. The model is trained on data from NASA’s Solar Dynamics Observatory (SDO) and uses a forecasting pretext task rather than masking. Schmude highlights that the model can generate visual predictions of solar flares, which is novel. The team also released Surya Bench, a benchmark dataset and evaluation suite. Schmude shares surprises, such as the finding that injecting domain knowledge like solar rotation often hurt performance, and that a purely data-driven approach worked better. He discusses potential applications in space weather forecasting, protecting satellites, and ensuring astronaut safety. The model is open-sourced on Hugging Face, and the team plans to continue research and support adoption.

190 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides valuable insights into the development of a novel AI foundation model for solar physics. It explains the scientific and AI challenges, such as the difficulty of predicting solar flares from limited observational data and the decision-making process for model design. The argumentation is coherent and credible, supported by the interviewer’s questions and the expert’s detailed responses. However, the discussion is high-level and lacks quantitative performance metrics or comparisons with existing methods, which would strengthen the scientific value.

Scientific Rigor, Source Quality, Title Accuracy

The video is produced by IBM Research, and the interviewee is the technical project lead, lending authority to the claims. The sources cited include the IBM Research blog and the Hugging Face model repository, which are appropriate for verifying the model’s existence and details. The title accurately represents the content, focusing on the motivation and development of the Surya model. The video does not include any external sources or references to peer-reviewed literature, which limits the scientific rigor.

173 words

Title / Content Match

The title accurately reflects the content, which focuses on the motivation and development of the Surya model for predicting solar flares.

Quality & Reliability

8/10

The video is an interview with the technical project lead, providing credible insights into the development and capabilities of the Surya model. The claims are plausible and align with the open-source release and benchmark. However, the video is promotional in nature and lacks independent verification or detailed technical evidence.

Key Moments

Cited Sources

Concurring Sources

External References

Contribution & Novelties

The video presents the Surya model as a pioneering AI foundation model for solar physics, capable of generating visual predictions of solar flares. This is a novel contribution compared to traditional binary classification approaches. The model is open-sourced, along with a benchmark, to foster community adoption and further research. The discussion also highlights the counterintuitive finding that purely data-driven approaches outperformed models with injected domain knowledge, which is a valuable insight for AI for science.

Pour aller plus loin :

122 words

Radar Profile

The radar profile shows high scores in quality and reliability, reflecting the credible source and expert interview. The quantity of information is moderate, and the technical level is accessible to a general audience. The overall profile suggests a well-produced, informative video with strong scientific backing.

Reliability 8/10

💬 No comments were provided for analysis.