A Counterfactual Emulator for Circulation-Driven Extremes in Southeast Asia

Climate extremes jeopardize human health and the environment. Recent unprecedented extremes suggest a complex interplay between anthropogenic warming and internal variability of the climate system, with large-scale circulations exhibiting considerable uncertainty in response to climate change. Therefore, understanding the influence of large-scale atmospheric and oceanic circulations on extreme events in a changing climate is crucial for climate adaptation and risk assessment. Traditional physical climate models, while powerful, require extensive computational resources to explore the broad spectrum of potential future circulation states and their implications for the infrequent occurrence of extreme events. This study takes Southeast Asia as an example to demonstrate the influence of Madden–Julian Oscillation (MJO) on extreme precipitation and droughts in a changing climate, as MJO strongly modulates local convective systems in Southeast Asia. We develop an AI-empowered emulator framework based on a conditional diffusion model to generate the precipitation field in a counterfactual world, where the enhanced convective phases of MJO are more (less) frequent than the current climate. We then estimate the intensity-frequency curves of extreme precipitation (drought) events and quantify the uncertainty using the generated large ensemble of samples. This counterfactual emulator allows us to isolate the influence of MJO phases and frequencies on extreme event probabilities, making it feasible to simulate a wide array of circulation states and examine their impacts under various climate change scenarios. By overcoming computational barriers, the study offers a clearer understanding of climate extremes in response to changing circulations for policymakers and stakeholders, enabling climate-informed resilience planning and evidence-based governance policy.