Studying climate intervention scenarios with data science methods

Abstract

Stratospheric aerosol injection (SAI) is a proposed climate intervention strategy that may alleviate the most severe climate change impacts. The ways in which a possible SAI deployment may occur are vast and require social, environmental, and political consideration. To help scientists decide which of the many possible futures to study, scenario design methods are often used. SAI scenarios generally fall into one of two categories. The first, coordinated global deployment, is characterized by global cooperation among all or most of the world's population in deciding how to deploy SAI. This work explores a common tool used alongside coordinated global deployment climate model simulations to quantify the tool's sensitivity to internal variability. The second category, single actor deployment, is characterized by a single party or a small group independently deploying SAI to achieve their own climate goals. Under a range of these scenarios, atmospheric responses may motivate additional counter SAI deployments or retaliation from other parties. This work quantifies atmospheric responses shortly after a wide range of single actor deployments. Finally, this work concludes by introducing a novel machine learning emulator designed to explore SAI deployment scenarios for use alongside earth system models to inform future simulation design. This machine learning emulator is comprised of multiple interpretable machine learning models that are trained on simulations from the Community Earth System Model version 2 including a novel simulation titled, Climate Response After Stratospheric Sulfur dioxide injection (CRASSULA). Here the emulator design is described along with how this emulator can be used for scenario design. The work in this dissertation emphasizes the need to question how SAI scenarios are simulated, what assumptions are made, and what possible scenarios are being overlooked.