GC51A-01 Advancing Emulation for Understanding Nature-Society Interactions (Invited)

Accurately modeling Earth systems in ways that can inform decision-making requires understanding the interactions between natural systems and human activities, and assessing how these interactions change under potential intervention strategies. Emulators can play an important role in helping to incorporate additional components of Earth systems and their interactions in integrated model analyses, without adding substantial additional computational burden. We build upon a recent review of climate model output emulation to assess the advantages and disadvantages of various emulation techniques for specific applications, based on differences in methods (from simple linear regression to advanced machine learning emulators), complexity (including interpretability and computational costs to create them), and outputs (variables and their spatial and temporal frequency). Using physical intuition and fundamental principles about the climate system, we have also identified when and where some of the simplest approaches like linear methods can be accurate, and where they break down, and how more complex emulation techniques such as response functions and Gaussian process approaches relate to one another. Here, we present a strategic approach to emulation, combining simple and advanced methods, that leverages physical intuition while minimizing unnecessary computational complexity. Outputs include projecting future temperature, precipitation, and air quality, including uncertainty and variability. We show selected concrete examples of this approach in specific real-world applications for decision-making and outreach, identifying situations where simple methods are useful, and where they fall short. Finally, we show how emulation can enable and enhance detailed projections of the impacts of up-to-date scenarios that include current and proposed policies.