AI is transforming weather and climate science as AI-based weather and climate models now rival their physics-based counterparts on various standard benchmarks while producing forecasts at a fraction of the computational cost. However, many important questions remain beyond overall prediction skills. How reliable are their forecasts for extreme events, particularly for those outside their training range? Can they capture spatio-temporal and inter-variable dependence, and are their forecasts physically consistent? How well do they quantify uncertainty? What strengths of physics-based models do they still lack?
Answering these questions is essential for understanding how AI-based and physics-based numerical models differ, identifying where AI models could be further improved, and building trust in using AI weather and climate models for scientific discovery and operational forecasting. However, the mathematical and statistical tools to address these questions remain underdeveloped.
This workshop brings together researchers from applied mathematics, statistics, atmospheric and Earth sciences, and computer science to foster interdisciplinary discussions on these challenges and deepen our understanding of AI weather and climate models.
More details about the workshop can be found at: https://zhongwei-zhang.github.io/Workshop_Heidelberg_2027/