Research to improve heat wave prediction using machine learning

Main findings

• Enhanced predictive capabilities: Machine learning-based systems can compete with, and in some cases exceed, traditional predictive systems. Improvements have been seen in particular in Northern Europe, where traditional methods have been difficult.

• Key timing insights: The best forecast skill comes from atmospheric forecasters four to seven weeks in advance in summer (mid-March), providing the best lead time for preparedness actions.

• Lives and livelihoods at risk: Early warning capabilities could help agricultural industries and health services implement effective strategies against heatwaves that cause increased mortality and economic losses across Europe

• Cost-effective innovation: The new approach requires much less computational power compared to traditional systems, making seasonal forecasting much more accessible.

• Successful model-to-reality translation: The system accurately predicts real-world heatwaves from 1993 to 2016, including extreme events like 2003 and 2015.

The study, “Seasonal prediction of European heat waves using a feature selection framework,” published in Nature Communications Earth & Environmental, demonstrates CMCC’s leadership in integrating cutting-edge artificial intelligence and climate science to address one of Europe’s most pressing climate challenges: heat waves.

It shows how machine learning (ML) and artificial intelligence (AI) techniques are revolutionizing climate science by enabling more accurate and cost-effective predictions than traditional approaches. Furthermore, in the Nordic region, where traditional dynamic forecasting systems require large amounts of computational resources and have reliability issues, this data-driven approach provides an alternative.

“ML will become a fundamental part of how we study climate change,” McAdam says. “While this work has demonstrated the utility of ML in predicting extreme events, it is only the first step in defining how to do so in order to obtain interpretable and physically meaningful results.”

Heatwaves cause devastating effects across Europe, including agricultural losses, spikes in energy use, health crises and increased mortality. Recent fatalities in 2003, 2010 and 2022 highlight the urgent need for early warning systems to mitigate the effects of heatwaves. This is especially important because climate projections suggest that heat waves will become more intense in the coming decades, making accurate seasonal predictions critical to saving lives.

“Early warning of a hot summer could help society be better prepared to reduce economic losses and reduce risks to life,” McAdam explains. “Seasonal forecasts made in the spring can, in principle, indicate whether the summer will be warmer than average.”

innovative methodology

The system employs an optimization-based feature selection framework that identifies the optimal combination of atmospheric, oceanic, and terrestrial variables to predict the likelihood of heatwaves across Europe. This approach uses ML techniques to analyze approximately 2,000 potential predictors and select the most predictive combinations for each geographic location.

Not only does this method perform as well as, and in some cases better than, traditional prediction systems, it also provides information about which predictor variables were used in the process, a valuable scientific resource. The ability to pinpoint which atmospheric and oceanic predictors contribute most to forecast skill at different times and locations across Europe could inform future research into the physical mechanisms behind extreme heat events, for example.

For example, the study found that European soil moisture, temperature patterns, and atmospheric circulation emerged as the most important regional predictors, while distant signals from the tropical Pacific and Atlantic Oceans also contributed to forecast skill.

A persistent challenge in seasonal forecasting is the poor performance in Scandinavia and north-central Europe. In contrast, the new data-driven approach developed in this paper improves skills in these previously problematic areas.

One of the most innovative aspects of this study was training the ML system on paleoclimate simulations spanning years 0 to 1850, providing far more training data than is available in the observational record. Despite this unusual approach, the system successfully transferred its learning to accurately predict real-world heat waves from 1993 to 2016.

“We didn’t yet have enough real-world data to adequately train the predictions, so the ML model actually learned about the heat wave factors in the model world, but we were able to apply that training to the real world,” McAdam said.

efficiency issues

In addition to increased efficiency, the computational requirements are significantly reduced, making seasonal forecasting using this technique available to a wider range of researchers and institutions. While running traditional dynamic systems requires enormous supercomputing resources, this approach specifically focuses on heat wave prediction with minimal computational overhead.

“In our work, we successfully extended data-driven ML-based forecasting to seasonal timescales using a fraction of the computational resources of traditional approaches,” said McAdam.

The system provides reliable seasonal predictions of heatwaves months in advance, allowing proactive measures to be taken to reduce the impact of heatwaves on society and the economy. This opens new possibilities for climate services across sectors such as agriculture, public health, energy management, and emergency planning, as well as opportunities to combine ML approaches with dynamic systems generated by CMCC to leverage the strengths of both approaches.

The framework has the potential to be adapted to other extreme events, start dates, and seasons of interest, and represents an important milestone in CMCC’s mission to advance climate science through innovative methodologies and establish new standards for seasonal forecasting and climate risk assessment.

Read the paper:

McAdam R. et al., Seasonal prediction of European heat waves using a feature selection framework, Nature Communications Earth & Environmental, DOI: 10.1038/s43247-025-02863-4

The European-Mediterranean Climate Change Center (CMCC) is a leading research institute specializing in climate science, providing cutting-edge insights and innovative solutions for climate adaptation and mitigation strategies. CMCC plays a vital role in global climate research, working closely with international partners to advance climate modeling, forecasting, and policy advocacy. www.cmcc.it