Decadal Vision 2026-2035

The E3SM Decadal Vision describes a 10-year effort to advance E3SM into a trustworthy hybrid physics-AI Earth system prediction capability for seasonal-to-decadal timescales. Building on a decade of progress, including a skillful coupled Earth-energy model and a world-leading global storm-resolving model on exascale computers, the project is now focused on improving predictions of energy-relevant Earth system behavior across months to years. The vision emphasizes understanding the sources of predictability, especially for phenomena that affect U.S. energy and water systems.

A central theme of the document is the importance of predicting extreme events and other high-impact conditions such as hurricanes, cold-air outbreaks, windstorms, heat waves, droughts, floods, ice storms, and wildfires. These events are closely tied to power outages, infrastructure damage, and shifts in energy supply and demand. E3SM aims to provide better forecasts of the frequency, intensity, and geographic distribution of these hazards to support planning, resilience, and decision-making across the energy sector.

The plan is guided by two overarching science questions. First, it seeks to identify the mechanistic sources of predictability for DOE-relevant Earth system phenomena at seasonal, annual, and decadal scales. Second, it asks how model design choices, including higher resolution, change predictable signals and internal variability. To answer these questions, E3SM will use large ensembles, improved initialization and data assimilation, calibrated model development, and process-based evaluation to distinguish predictable signals from noise and to better understand where skill comes from.

A major computational focus of the vision is the expanded use of AI across the full modeling workflow. The plan highlights six transformative areas, including AI emulators, machine-learned parameterizations, AI-based data assimilation, autotuning calibration, workflow automation, and AI pattern recognition. Together with ultra-high-resolution physics-based simulations and continued leadership in DOE high-performance computing, these capabilities are intended to create a discovery loop that improves both prediction skill and scientific understanding.