A Climate Intervention Dynamical Emulator (CIDER) for scenario space exploration

Abstract. Stratospheric Aerosol Injection (SAI) is a form of proposed climate intervention to reflect incoming solar radiation, offsetting some of the impacts of greenhouse gas warming. Due to the characteristics of stratospheric circulation, the lifetime of such aerosols, and the differential impacts that different aerosol patterns can produce on surface climate, many possible scenarios of SAI implementations might exist, ranging from steady, cooperative deployments across one or more injection latitudes to highly dynamic, uncoordinated deployments involving multiple independent actors with different aims. However, a full exploration of this scenario space is constrained by the computational cost of fully coupled climate model simulations that are usually used to evaluate the impacts of potential scenarios. Here, we describe the development and evaluation of the Climate Intervention Dynamical EmulatoR (CIDER), a climate emulator that can be used to quickly simulate the response to a SAI deployment on both a regional and a global scale for a set of variables (temperature, precipitation, evaporation, and sea ice fraction) as the injection rates vary in magnitude, latitude, and time. CIDER is trained on a large but finite set of pre-existing Earth System Model (ESM) simulations, but it can emulate novel, out-of-sample scenarios at a small fraction of a cost of one ESM simulation. Because CIDER does not include a representation of how SAI affects the diurnal and seasonal cycles, nor how it affects internal variability, it is not meant to substitute for ESMs, nor to directly inform more detailed impact analyses of SAI. Nevertheless, it can be used to quickly understand the broad impacts of different SAI strategies and produce large sets of different SAI implementations, making it a valuable tool for educational and communication purposes, for rapid identification of scenario parameters prior to simulation in a full ESM, and for coupling with Integrated Assessment Models (IAMs). In this paper, we describe CIDER and its workflow, as well as the process we used to train on existing simulations. We then evaluate the emulator's performance on a novel scenario, simulated using the same climate model used for the training set, but not included in the set, showing that CIDER is capable of emulating outside-the-box scenarios with a high degree of fidelity. The novel scenario we use is an example of a multi-actor, uncoordinated SAI deployment, and thus rather different from the balanced, coordinated scenarios used in the training set and typically simulated for SAI. The code and underlying training set are open source and available for the community to reproduce our results and improve upon them.