Description:

Weather forecast products and decision support tools based on novel approaches incorporating machine learning, artificial intelligence, and other technologies could positively impact applications across the weather, water, and climate enterprise. However, navigating these products from research-to-operations can be challenging. Particularly for AI-based tools there are both social (perceived reliability, trustworthiness, skepticism towards novel approaches) and technical (complexity of continuous integration/deployment, training/serving skew, data engineering) challenges which may be new or different than ones that traditional weather forecast products encounter during R2O.

In this session we seek diverse perspectives from AI/ML practitioners building and operationalizing products with weather and climate applications. What are the critical social and technical challenges that one may expect to encounter when launching an AI-based weather product? Are there particular tools, technologies, or approaches that can help mitigate some of the technical challenges related to producing machine learning models, or are there specific gaps in tooling for weather-related problems that the community ought to be aware of and invest in closing? How should teams building these products approach developing trust with their end users, be they in the private or public sectors? Are there resources (especially in terms of knowledge- or practitioner-based communities) that practitioners should be aware of which can help them overcome R2O challenges?

We are particularly interested in perspectives from both practitioners in the private and public sector, who may have very different requirements or hurdles in operationalizing AI-based products. Specific problems of interest:

• From the cluster to the cloud: the technologies, tools, and platforms accelerating the productionization of research products into operations.
• Verification and validation: the processes, procedures, and communities which develop trust for novel AI-based forecast or decision support processes, including testbeds, field experiments, or monitoring/verification services.
• Quantifying uncertainty: techniques for effectively capturing and communicating probabilities or uncertainties associated with model outputs and providing effective guidance for end users.
• Success stories: examples of lab groups, companies, or public agencies which have rolled out new AI-based decision support tools, and the key factors that played into those successes.
• Cautionary tales: examples of lab groups, companies, or public agencies which attempted, but ultimately failed, to roll out new AI-based decision support tools and the key challenges which stood in the way of success.