Description:

In recent years, interest in AI and ML-based weather forecast and decision support tools has grown across public and private sectors. However, many organizations need help operationalizing these tools and face challenges ranging from social (reliability, trustworthiness, skepticism) to technical (integration, training/serving skew, data engineering, MLOps, cloud vs. HPC infrastructure).

We seek wide perspectives from AI/ML practitioners involved in developing and operationalizing weather and climate applications. Key questions include:  What are the critical social and technical challenges when launching AI-based weather products? Are there specific AI/ML based tools, technologies, or approaches to overcome these challenges? How can trust be developed with end users? What resources can help practitioners overcome research-to-operations (R2O) challenges?

We are particularly interested in insights from public or private sector practitioners who have successfully implemented AI/ML products to empower other developers. Key areas of interest include:

1. Cluster to cloud: technologies, tools, and platforms accelerating research product production.
2. Verification and validation: processes and communities fostering trust in AI-based forecast or decision support systems.
3. Quantifying uncertainty: techniques for capturing and communicating model output probabilities and providing guidance for end users.
4. Model confidence: methods to quantify ML models' robustness to a changing climate.
5. Success stories: examples of successful AI-based decision support tool rollouts and key factors in their success.
6. Cautionary tales: instances of unsuccessful AI-based decision support tool rollouts and key challenges faced.

Conveners: Thanh Vo, NOAA/NESDIS/OSAAP Affiliate and ISS, Inc. and Aaron Hill, NOAA/OAR Affiliate, Colorado State University