Shruti Nath: Climate model emulators – tools for agile exploration of future climate uncertainties

Climate models lay the foundation on which we can explore future climate scenarios and the impacts of anthropogenic greenhouse gas (GHG) emissions. They are process-driven models which solve a set of governing, physical equations over a discretized grid to prognose and diagnose a wide range of climate variables. 

Running climate models typically takes months and carries high computational costs for generating and storing realizations. This makes exploring uncertainties surrounding future emission trajectories, alongside other uncertainties – such as that due to natural climate variability – time consuming and computationally expensive. 

Climate model emulators are lightweight statistical tools that train on climate models to then approximate key relationships and properties of impact-relevant climate variables in a functionally identical manner. 

In this talk I will mainly discuss climate model emulator developments conducted on a monthly level and give examples of use-cases within extreme event attribution. I will furthermore talk about data compression techniques used when extending climate model emulator frameworks to multivariate emulations and give concrete examples of impacts that can be explored with such monthly, multivariate emulations.

Shruti Nath has a background in mathematics and completed her PhD jointly at the Berlin based NGO Climate Analytics and the Institute of Atmospheric and Climate Sciences at ETH Zürich. Her PhD involved lightweight emulator development, where she explored a range of statistical methods to approximate the uncertainty dimensions surrounding future climate scenarios. Dr. Nath is currently based at the University of Oxford as a Postdoctoral researcher.