6) This approach opens the door to operational attribution systems that can assess climate change impacts even before an event unfolds, thereby improving preparedness and response.

More details in the paper: agupubs.onlinelibrary.wiley.com/doi/10.1029/...

5) In every case, human-made climate change made the heatwaves more intense.
We also found that sea surface temperature forcing can modulate the signal in some cases — but atmospheric initial conditions used to perturb the counterfactual forecasts are the main driver.

4) We applied this to 4 major heatwaves worldwide 🌍 using:
• FourCastNet-v2
• Pangu-Weather
• NeuralGCM (AI + physics hybrid)
All three captured key dynamics like persistent high-pressure systems and can predict heawaves accurately 5 days in advance.

3) Our method uses a pseudo–global warming (PGW) framework:
• Run forecasts for today’s climate 🌡️
• Run forecasts for a preindustrial climate 🕰️
Only the thermodynamic initial conditions change — atmospheric circulation at the initial state stays the same.

2) Traditional attribution methods are powerful, but they can take days, weeks and need heavy computing.
Here, we tested a faster route using three state-of-the-art AI models that run in minutes, without supercomputers.