EGU Session: Verification, diagnostic, and interpretability of AI models for weather forecasting
Philine Lou
Dear colleagues and friends,
We invite you to submit an abstract to our session at EGU 2026 in Vienna, where we explore current developments for AI-based weather forecast verification. As AI-based weather models move from research breakthroughs to operational landscapes, the scientific community faces a critical question: How do we rigorously verify these "black box" systems?
Session ITS4.4/NP5.2: Verification, diagnostic, and interpretability of AI models for weather forecasting Session Link: https://www.egu26.eu/session/55793
Conveners:
Zied Ben Bouallegue (ECMWF)
Jochen Broecker (University of Reading)
Philine Bommer (University of Edinburgh, TU Berlin)
Romain Pic (ETH Zurich)
While AI models often outperform traditional NWP in standard metrics like RMSE, a deep assessment of their physical realism and reliability is still ongoing. This session serves as a dedicated forum to move beyond surface-level metrics and investigate the internal logic and operational readiness of AI weather models.
We welcome contributions covering theoretical, methodological, or operational aspects, including:
Benchmarking & Intercomparison: Datasets and frameworks comparing AI vs. NWP.
Advanced Verification: Spatial methods, innovative scoring rules, and forecast realism.
Extreme Events & Fairness: Assessing predictability and artifacts in high-impact scenarios.
Explainable AI (XAI): Interpretability methods to deconstruct predictive reasoning.
Whether your work is focused on the statistical rigor of verification or the transparency of deep learning architectures, we want to hear from you.
Key Information:
Abstract Deadline: January 15, 2026 (13:00 CET)
Format: On-site in Vienna & Virtual options available.
Please feel free to reach out if you have any questions regarding the session. We look forward to your contributions and to a wide discussion in Vienna!
Best regards,
Philine Lou Bommer (On behalf of the session conveners)