Extreme events are becoming more frequent in a warming world, causing severe damage to the society. For decades, studies related to extreme events remain with statistical methods on a group of similar events with tests on empirically given large-scale background factors, and are limited by the skill and efficiency of climate models. Why a particular extreme event happens and how different parts around the globe impact it still puzzle climate scientists today.

In a study recently published on Geophysical Research Letters, researchers from Institute of Atmospheric Physics of the Chinese Academy of Sciences, UK Met Office Hadley Centre, CSIRO in Australia, Peking University, and Capital Normal University, developed a novel event-based, global-scale diagnosing framework combining hybrid AI weather forecast model with Green’s function experiments.

The researchers take the August 2022 record-breaking heatwave in South China as a case. AI weather model well reproduced the phenomenon and related physical mechanisms, allowing it to be further used for Green’s function experiments. By dividing the Northern Hemisphere into patches and removing initial anomalies one patch at a time to make predictions, changes in forecast skill help elucidate a shift of high-impact regions towards this specific extreme heatwave. The heatwave was initially controlled by adjacent regions, but then dominated by remote regions in East Europe and North America with concurrent heatwaves, which contributed 1.94℃ and 1.61℃, respectively. Dynamical analysis confirmed the eastward wave propagation pathway responsible for the relay-like sequence that sustains the heatwave.

Additional experiments forecast with initial anomalies from those high-impact regions, which covers only 25% of the globe, well reconstructed the heatwave pattern and high-pressure system with an acceptable underestimation. Compared to the forecast initialized from high-impact regions identified via composite and regression analysis of eight similar South China summer heatwaves during 2021-2025, the comparable forecast skill achieved with a more precise range of high-impact regions highlights the advantage of this event-based diagnostic framework over the conventional statistical methods in past extreme event diagnosing works.

Remote source regions shape the South China heatwave during days 6-11. Forecasts driven only by the identified high-impact regions successfully reproduce the event, whereas removing those regions strongly degrades the prediction. （Image by IAP）

This study goes beyond simply diagnosing single extreme events. Powered by the efficiency of AI models, global-scale Green’s function-like regional perturbation experiments are no longer out of reach, thus providing a rapid and causally robust way to identify key source regions, which can be further used to improve early warning skill and help with the adaptive strategies in a warming world.