Does Progress in Earth System Modeling Require More Data, More Compute, or Both?

Explore the fundamental resource requirements driving progress in Earth system modeling through this 37-minute conference talk examining whether advancement depends more on data availability, computational power, or both. Discover how autoregressive modeling serves as the prevailing paradigm in Earth system science, manifesting in two primary forms: traditional time-stepping of physical equations and training large deep learning models with real-world and artificial atmospheric trajectories. Learn about the versatility of these models in addressing diverse tasks including weather forecasting, climate projection, climate downscaling, and extreme event quantification. Examine emerging modeling approaches such as denoising diffusion that better handle dataset limitations like inadequate spatial or temporal sampling. Understand the critical role of data and computational resources as the foundation for all Earth system models, and identify where resource scarcities currently impede forward progress. Gain insights into potential pathways for overcoming these limitations and advancing the field of Earth system modeling through strategic resource allocation and methodological innovations.