AI Emulator for Climate Engineering
Image credit: Alex Wikner, The CeTD Group
Solar radiation modification (SRM) has been proposed to reflect more sunlight back to space to cool the planet, potentially reducing some of the impacts from climate change. Earth System Models (ESMs) have been the primary tool to understand the physical climate responses to SRM interventions. However, ESMs are computationally expensive, making it difficult to reduce uncertainties surrounding the effects of SRM, especially on weather extremes.
Human decision is particularly uncertain when modeling SRM because there are many degrees of freedom to consider in designing interventions. This includes how and how much, where, when, and for what desired climate target SRM might be deployed. These uncertainties are particularly high for weather extremes, which are rare but have the most significant impacts for society. Reducing these uncertainties requires models that are accurate and fast, yet the most accurate ESMs are slow and computationally demanding.
Recent advances in artificial intelligence (AI) weather and climate emulators, which are large neural networks trained on ESM outputs evolving in space and time, present an exciting opportunity to reduce uncertainties at a fraction of the time and cost. In collaboration with Argonne National Laboratory, this project will build, train, and test the first AI climate emulator for SRM. The emulator will be trained with output from the Energy Exascale Earth System Model (E3SM) under different idealized SRM perturbations. The tests will evaluate the emulator’s fidelity to interpolate between various SRM scenarios and represent changes in regional weather extremes.
