Machine learning emulators to predict the atmospheric response to stratospheric aerosol injection scenarios

Stratospheric aerosol injection (SAI) is a proposed climate intervention strategy to lessen global warming as an additional approach operating alongside emissions reductions, adaptation strategies, and large-scale carbon dioxide removal. There are still many unknowns to how the Earth system may respond to SAI, and the range of possible SAI scenarios is large and therefore computationally expensive to model. Here, we discuss early stages of designing and training machine learning emulators to predict the atmospheric response to a range of SAI scenarios. These emulators are trained on various climate model simulations (i.e. pre-industrial control, SSP2-4.5, and scenarios from the Assessing Responses and Impacts of Solar climate intervention on the Earth system with stratospheric aerosol injection simulations) as well as a new simulation called Climate Response After Stratospheric Sulfate Injection (CRASSULA). The CRASSULA simulation is designed specifically for training these machine learning emulators and is carried out with the Community Earth System Model 2 with Whole Atmosphere Community Climate Model 6 in a middle-atmosphere configuration. This suite of machine learning emulators can be used to explore a large combination of SAI scenarios, help quantify the atmospheric impact and associated risks, and assist in the design of future climate model simulations of SAI scenarios.