CMCC Machine Learning Model Predicts Heatwaves Up to Seven Weeks in Advance 

Researchers at the Euro-Mediterranean Centre on Climate Change (CMCC) have achieved a breakthrough in climate resilience by developing a machine-learning (ML) model capable of predicting severe European heatwaves up to seven weeks (4–7 weeks) in advance. The new AI system is designed to provide crucial lead time for preparation, addressing heatwaves which are among Europe’s deadliest climate hazards and are becoming increasingly longer, more intense, and more frequent due to climate change. The system’s findings, published in Nature Communications Earth & Environment, highlight a paradigm shift in seasonal climate prediction.

The ML model achieves its predictive skill by analyzing a massive dataset of approximately 2,000 atmospheric, oceanic, and land variables. It successfully identifies key patterns that signal when and where heatwaves are likely to form. Researchers achieved this breakthrough by training the AI system not only on modern observational data but also on computer reconstructions of ancient climates spanning from the year 0 to 1850. This unconventional training method provided the model with hundreds of extra “virtual years” of weather data, enabling it to accurately predict real-world heatwaves that occurred between 1993 and 2016.

Crucially, the CMCC team argues that their AI system can make these highly accurate forecasts while using only a tiny fraction of the computational resources demanded by older, traditional dynamical forecasting models. This computational efficiency significantly reduces costs and democratizes access to seasonal prediction capabilities, making the technology available to a broader range of government agencies and research groups. Early warnings of extreme heat could be used by sectors like public health, energy, and agriculture to save lives, protect crops, and ease pressure on power grids.