Robin Guillaume-Castel

I am a climate scientist who uses explainable artificial intelligence (XAI) methods to understand and predict the Earth's climate system. My current research focuses on using AI not only for prediction, but also as a tool for scientific discovery and physical assessment. I develop XAI approaches to evaluate the physical realism of neural networks in weather and climate sciences, and to improve our understanding of the physical mechanisms that drive climate and weather variability. 

I am particularly interested in how neural networks learn climate processes, how their predictions can be interpreted in physical terms, and how AI can help understand previously unknown relationships within the climate system. More recently, my fork has been focused on understanding neural network predictions of heavy rainfall, linking large scale atmospheric dynamics to local precipitation events.

Research Interests

• Explainable artificial intelligence (XAI)
• Climate variability and predictability
• Heavy precipitation and extreme events
• Climate dynamics
• Sea surface temperature pattern effect
• Earth's energy budget
• Machine learning for Earth system science

Selected related work:

• Explainable AI shows that a neural network learns extratropical cyclones as predictors of heavy precipitation (preprint): https://egusphere.copernicus.org/preprints/2026/egusphere-2026-3362/egusphere-2026-3362.pdf
• Can AI models be trusted to predict heavy rainfall? (interview by UiB AI): https://www.uib.no/en/ai/181814/can-ai-models-be-trusted-predict-heavy-rainfall

While a major part of my current work focuses on explainable AI for climate and weather science, my broader scientific interests consist of climate dynamics in general. In particular, I am interested in how patterns and gradients of sea surface temperature (SST) influence atmospheric circulation and the global energy budget of the planet. My previous work has been focused on understanding how regional SST gradients shape global climate responses through their interaction with the Earth's energy budget, including their influence on the transient climate response.

Selected related work

• The role of SST pattern effects in transient global warming (Journal of Climate): https://journals.ametsoc.org/view/journals/clim/38/14/JCLI-D-24-0229.1.xml
• ENSO influences on internal variability of the Earth's energy budget (Geophysical Research Letter): https://agupubs.onlinelibrary.wiley.com/doi/pdf/10.1029/2025GL116952

BCCR Training Programme in Machine Learning:

In addition to my research, I contribute to building machine-learning expertise within the Bjerknes Centre community. Together with Sigrid Passano Hellan from NORCE, I co-designed and developed the BCCR Training Programme in Machine Learning, which provides structured training in machine learning and deep learning for geoscience researchers across the Bjerknes Centre. 

The programme is designed to give researchers with little or no prior experience in machine learning the knowledge and practical skills needed to understand these methods and apply them to their own scientific questions. 

Resources for this course are accessible here: https://bjerknes.uib.no/en/for-bjerknes-members/machine-learning-training