Climate Change Attribution of Extreme Events Using AI-based Weather Models

Climate change is altering the frequency and intensity of extreme events globally, such as heat waves, droughts, flooding, and tropical storms. It is, therefore, of primary importance to quantify the influence of climate change on the properties of specific extreme events. The scientific community has developed different methodologies to answer that question. One extended approach, commonly known as the pseudo-global-warming (PWG) approach, consists of removing the anthropogenic climate change signal from event-constrained initial conditions of a physics-based weather model. However, performing state-of-the-art atmospheric model simulations of every single extreme event requires significant computational resources. Artificial Intelligence (AI)-based weather models offer a chance to speed up this process due to its much lower computational cost.

In this study, we apply the PGW approach to the AI-based FourCastNet model forecasts of a selection of extreme events. To account for the uncertainty associated with the influence of climate change, we remove the signal in thermodynamic variables from several CMIP6 models independently from the AI model's initial conditions. At the same time, we also test the sensitivity to the initialisation date. Using this approach, we quantify the impact of climate change on a heatwave event and tropical storm. Our results indicate that climate change significantly contributed to the intensity and characteristics of these two extreme events. We also discuss the strengths as well as limitations of the AI-based model in the simulation and attribution of extreme events by comparing our results to reanalysis and to more classical attribution methods. This research contributes to our understanding of the impacts of climate change on extreme events and highlights the potential of AI-based models for climate attribution studies.