Solar Radiation Management Pessimization

Deploying Solar Radiation Management (SRM) to cool the planet comes with serious risks, including the possibility of altering the climate in ways that are harmful to humans. Some models suggest that, if done modestly and responsibly, stratospheric aerosol injection (one proposed method of SRM) could benefit humanity overall. But a key question remains: are these models reliable, given the uncertainties in how climate models are built?

Climate models with a high level of accuracy are computationally expensive—they require a lot of power, time, and computer memory to run—limiting their ability to represent a full range of relevant risks. This project proposes using machine learning to optimize uncertain parameters within these climate models to help produce worst-case SRM outcomes, a process we call “pessimization.” This will allow potentially dangerous mechanisms to be further investigated and ultimately mitigated or ruled out.

One of the most important risks of SRM is its potential to disrupt rainfall patterns and agriculture. Therefore, our first step will be to search for scenarios in which deploying SRM enhances precipitation changes caused by global warming over the largest land area. The proposed work represents the first time that a modern optimization algorithm is applied to identify risks associated with SRM. Critically, our methodology has the capacity to identify risks and physical mechanisms that are currently completely unknown, so that resources can be directed toward studying and hopefully mitigating them. We may also find that certain worst-case SRM outcomes aren’t as severe as feared, which could support the case for a scenario in which SRM is deployed.