Conveners
TDM13 - Scientific outlooks for analysis of space weather data in the age of AI
- Stefaan Poedts (KU Leuven)
- Panagiotis Gonidakis (Centre for Plasma Astrophysics (CmPA), Department of Mathematics, KULeuven, Belgium)
- Ekaterina Dineva (KU Leuven, CmPA)
- George Miloshevich (KU Leuven)
Description
There is a growing demand in the space weather community for the analysis of large datasets, due to availability of data from space missions and numerical simulations. To keep pace, the community is using off-the-shelf algorithms, adapting models from computer science. These models offer valuable opportunities to improve space weather services, e.g. forecasting capabilities. Other important applications include automatic detection and segmentation of areas of interest, where machine learning algorithms provide greater flexibility and efficiency compared to traditional methods. Furthermore, such algorithms can be implemented in ground-based facilities and spacecraft for on-board automation and maximizing the retrieval of scientifically interesting data. Given the volume of publications of AI in space weather, the community should come up with better practices for standardization of data and methods to facilitate unbiased comparisons between the models. Importantly, transitioning AI from research prototypes to tools used by space weather centers requires trust, uncertainty quantification, validation and explainability. In this TDM, we raise the following questions: 1.) What are the fundamental challenges of producing trustworthy forecasting of space weather events using AI? 2.) What are the best practices for automatic detection and annotation of space weather events? 3.) Can AI not only detect events but also assess their importance and trigger high-resolution data capture for selective downlink.
