Collaborative Research: CAIG: Understanding radiative feedbacks, ocean heat uptake, and energy conservation to improve ML-climate emulations

About This Grant

Models used to simulate weather and climate rely on sophisticated algorithms to represent the physics of the atmosphere, ocean, land surface, and cryosphere. These models have been quite successful but they have two important shortcomings: first, they are computationally intensive, typically running on world-class supercomputers and generating terabytes of data which are challenging to host and serve. Second, they do not take advantage of the large amounts of observational data collected over decades using satellites, weather balloons, ocean moorings, and other observing systems. A new approach addresses these shortcomings by developing "climate emulators" which use machine learning to extract statistical relationships from observations and various types of physics-based computer simulations. Climate emulators have tremendous potential but it is unclear how well they capture the underlying physics of weather and climate and are thus able to generalize beyond their training sets. For instance an emulator which has learned statistical relationships in a cold climate might not perform well in a warmer climate or vice versa. Work performed under this award uses multiple emulators to simulate the response of the climate system to patches of warmer surface temperatures in different regions. The patch methodology is well established and thus allows evaluation of emulators against traditional physics-based climate models. The work also addresses long-standing questions in coupled climate dynamics, such as the effect of surface temperature fluctuations in one region on surface temperature in other widely separated regions, the effect of regional surface temperature variations on the global energy balance, and the extent to which precipitation in the Southwestern US can be predicted from knowledge of surface temperature variations over the tropical Pacific Ocean. A key issue in the work is the ability of climate emulators to conserve energy, as energy conservation would dramatically increase their value and adoption for both scientific and practical applications. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.