This session focuses on machine learning (ML) methodology applied in the meteorological context. The topics include model architectures, training strategies, uncertainty quantification, evaluation and validation schemes underpinning reliable ML for the Earth system. We aim to bring together contributors from meteorology, climate science, computer science, and applied mathematics who are advancing the theoretical and methodological foundations of ML for weather and climate. We also welcome approaches applied to weather extremes across time scales, with a strong emphasis on uncertainty quantification and probabilistic prediction in operational settings. In particular, we encourage studies that bridge AI-based forecasts with impact-based forecasting, risk assessment, and decision support, including applications to floods, droughts, heatwaves, storms, atmospheric rivers, and compound or cascading hazards.

We invite contributions on topics including, but not limited to:

* Novel model architectures with potential to be applied in meteorology/climatology.
* Novel applications of ML architectures for geophysical data.
* Training strategies and objectives
- including e.g. loss functions, self-supervision, pre-training and fine-tuning, transfer learning, and data augmentation, ...
* Integration of physical knowledge
- physics-informed and hybrid models, constraints and regularisation, stability and robustness, ...
* Uncertainty quantification and reliability
- probabilistic ML, ensembles, Bayesian approaches, decision-relevant evaluation, ...
* Evaluation strategies and evaluation studies
- Intercomparison of different architectures, comparison with physical methods, benchmark strategies.
* Interpretability, explainability and fairness
- methods to understand, diagnose and stress-test ML models...
* Human aspect -- how AI changes our work, organisations, and culture?
* ML and hybrid approaches for extreme event prediction
* Evaluation of AI forecasts for rare and high-impact events
* Integration of AI methods into operational workflows: Case studies demonstrating operational feasibility and societal benefits
* Translation of probabilistic AI forecasts into impact-based warnings and user-oriented products