Accelerating predictive capabilities by 10x

Motivation: 

To understand complex natural systems and their impacts on energy infrastructure, we need high-resolution models that deliver predictions at regional to neighborhood scales actionable for decision makers.

Approach: 

AI foundation models cut the computational time for DOE’s Energy Exascale Earth System Model to reach a steady state necessary for high-confidence predictions.

Impacts: 

• Speed model time-to-solution by 10x
• Reduce typical spin-up time from 5 full days on the Frontier supercomputer to a few hours
• Accelerate discovery through increased productivity
• Reduce uncertainty through large ensembles

S&T Challenge:  Aligns with “Predicting U.S. Water for Energy”