The Role of AI and the HPC-Edge Continuum in Urgent Science

This panel, moderated by Manish Parashar of the University of Utah, examines how AI and the HPC-edge continuum can support urgent science: time-constrained computing for emergencies such as wildfires, earthquakes, tsunamis, floods, and pandemics. Nicola Ferrier of Argonne National Laboratory presents the NSF-funded Sage Grande project, which deploys edge nodes with GPUs and vision language models that have detected volcanic activity and flooding in Hawaii, and discusses early experiments with sandboxed agents that observe, reason, and act. Miyoshi of RIKEN describes 30-second-refresh heavy rain forecasting using phased array weather radar and the Fugaku supercomputer, demonstrated during the Tokyo Olympics and Osaka Expo, plus AI-accelerated storm surge and tsunami prediction and a vision of a data assimilation-AI agent as controller for hazard workflows. Marisol Monterrubio of Barcelona Supercomputing Center covers urgent computing for earthquakes, including an ensemble of 50 seismic source simulations completed in 90 minutes on MareNostrum 5 during the Mexican national drill. Abani Patra of Tufts University discusses digital twins of offshore wind turbines and seagrass ecosystem modeling, emphasizing data harmonization, virtual sensing, surrogate models, and rigorous combination of data and model uncertainty. The discussion addresses agentic AI for cascading disasters, data provenance and ownership across the continuum, quantity-of-interest surrogates versus full simulation reproduction, and investment priorities, with panelists stressing upskilling scientists and connecting responders with the research community.