Observational Requirements in the Context of AI prediction Systems - a PCAPS ORCAS Task Team

The PCAPS ORCAS task team is part of the WMO's World Weather Research Programme's PCAPS (Polar Coupled Analysis and Prediction for Services) project. PCAPS builds upon the foundational work of the Polar Prediction Project and its flagship activity, the Year of Polar Prediction, to improve the actionability, impact, and fidelity of environmental forecasting for human and environmental well-being in the Arctic and Antarctic regions. PCAPS ORCAS is a community effort that aims to enhance forecasting capabilities by exploring the potential of new AI techniques. Outcomes from this initiative will contribute to strengthening observing systems, including satellite and field campaign data, to provide better initialisation and validation for sea-ice forecasts. 

Recent advances in artificial intelligence are transforming sea-ice forecasting, with AI models demonstrating comparable or superior performance to traditional physics-based approaches while requiring significantly fewer computing resources. These advantages could enable more frequent and timely predictions, benefiting stakeholders. However, the effective development and validation of these AI systems depend heavily on high-quality observational data. AI models are generally trained on reanalysis datasets, and data from observational campaigns--though vital for process understanding--has seen limited integration into these products. Such observations are essential to evaluate the physical realism of AI models and build trust in their predictive capabilities.

The PCAPS ORCAS task team systematically evaluates the observational requirements necessary for next-generation AI-based sea-ice prediction systems. This effort combines historical campaign data analysis with collaborative AI model assessments, focusing particularly on extreme events captured during major observational campaigns such as MOSAiC. We examine how different types of observational data contribute to model initialisation and validation while assessing the physical consistency of AI predictions compared to traditional forecasting systems. 

This approach identifies critical gaps in current observing systems and will inform the design of future field campaigns and observation networks, including those proposed for Antarctica InSync and the upcoming fifth International Polar Year. Our recommendations for strengthening polar observing systems specifically address the unique requirements of AI-based prediction systems while maintaining physical consistency in forecasts. These insights are essential for the polar science community as we work to improve the accuracy and reliability of sea-ice predictions in a rapidly changing Arctic and Antarctic environment.