Summary
The program objective is to explore AI-assisted modeling of complex processes related to climate. The specific goals of this effort are to:
- Explore the use of third wave AI methods to enhance models of complex interconnected processes. In particular, to develop hybrid AI models of the climate and Earth system that capture missing physical, chemical, or biological processes with sufficient computational efficiency to explore decadal scale effects and characterize tipping points and bifurcations.
- Develop methods to assimilate diverse data into models and estimate the “value of new data” to enhance confidence in target-specific forecasts relative to state-of-the-art (SOTA) techniques.
Climate change, whether natural or human-driven, has huge potential impacts on geopolitical and economic stability, food and water security, and DoD missions and operations.
Current climate models of highly complex, underlying physical processes are computationally intensive and provide limited actionable guidance to policy makers, especially on the risks and causes of sudden tipping-points, runaway feedback loops, and the strategic implications of potential adversarial activity. Third wave AI methods (e.g., neuro-symbolic hybrid AI models that can incorporate context and can extract causal factors and internal structure) have the potential to improve the accuracy of climate forecasts as well as improve the predictability of tipping points and to provide actionable guidance on new data when predictability remains poor. Faster learned models, particularly when used in conjunction with full-scale physics models for validation, will enable policy makers to better explore the climate impacts and risks of policy decisions.
Quantifying climate risks is essential to prepare for a range of scenarios such as those related to DoD planning and decision support (e.g., arctic strategy/defense, regional destabilization, global power/economic realignment, base/force locations, and extreme weather threats) and to identify new potential high-value observations (e.g., stratospheric vs. ocean surface vs. deep ocean vs. arctic, etc.) to enhance confidence in forecasts.
The 18-month AIE program will develop hybrid AI methods in Phase 1 (12-months) using open and rich datasets; and in Phase 2 (6 months), the program will apply the developed methods to quantify the “value of new data” for target-specific forecasts and establish the benefits of new methods relative to SOTA.