A new study led by CMCC researchers demonstrates the potential to predict variations in the frequency of occurrence of temperature extremes – both heatwaves and cold spells – up to two years in advance, offering unprecedented long-term forecasting capabilities that bridge the gap between seasonal and decadal climate predictions. The research presents significant advances in multi-annual climate forecasting that could transform the way societies prepare for extreme weather events, offering essential information for multi-annual planning in a variety of sectors, such as agriculture, energy, infrastructure and public health.

Temperature extremes pose escalating threats to human health, energy systems, agriculture, and infrastructure, with their impacts extending across multiple sectors and requiring long-term planning and risk management strategies. While seasonal and decadal climate predictions have been operationally available for years, the intermediate multi-annual timescale – covering forecasts from one to several years ahead – has remained largely unexplored despite its high relevance for decision-makers and end users across various socioeconomic sectors.

New research, led by a team of CMCC scientists and published in the journal Environmental Research Letters, addresses this gap by conducting a comprehensive global assessment of temperature extremes predictability on multi-annual timescales using the Community Earth System Model’s Seasonal-to-Multiyear Large Ensemble prediction system.

This study evaluates forecast skill for both warm and cold temperature extremes across all calendar seasons and regions worldwide, extending predictions up to two years ahead.

Unprecedented forecasting capabilities

Understanding and improving the predictability of temperature extremes is of great importance since their effects extend across multiple sectors – threatening human health, straining energy systems, damaging ecosystems and agriculture, and challenging infrastructure and economies. The research demonstrates that changes in the statistics of temperature extremes can be predicted beyond seasonal timescales, with skill persisting up to two years, particularly in the tropics.

“This study helps fill a gap in the literature by examining the predictability of temperature extremes beyond the usual seasonal timescales, considering both warm and cold events across all seasons and regions globally,” explains CMCC researcher and lead author Eirini Tsartsali. “This is particularly important because forecasts on these multi-year timescales are highly relevant for many end users, providing information between seasonal and decadal predictions that can support decision-making and planning in various socioeconomic sectors.”

The study reveals that predictive skill decreases over forecast time, yet notable skill persists in some regions even up to 20 months ahead. For example, Central and northern South America, parts of northern and southern Africa, the Arabian Peninsula, southeast Asia and Indonesia, as well as parts of Australia are the regions with the highest skill, extending through the 2 year period.

Multiple sources driving prediction skill

The research demonstrates that while external climate forcings dominate temperature extreme predictability, significant skill exists even after removing the externally forced signal, indicating that internal climate variability contributes to long-lead predictability of temperature extremes.

When removing external forcing, central and northern South America, Southeast Asia, and parts of Africa showed consistent skill up to forecast season four, especially during boreal spring and winter, while in the extratropics, North America retains skill in boreal spring.

The El Niño-Southern Oscillation plays a major role in this remaining predictability, though the research indicates that other factors also contribute. This finding is particularly significant as it suggests multiple pathways for achieving skillful multi-annual predictions beyond the well-established influence of ENSO.

“This research addresses the importance of providing actionable climate information about extremes on the multi-annual forecast horizon, thus bridging the gap between seasonal and decadal predictions,” notes CMCC researcher and co-author Panos Athanasiadis. “This is an emerging need identified by the growing community of end-users.”

The study represents a significant advancement in climate forecasting capabilities, particularly valuable for sectors requiring multi-annual planning horizons. More accurate long-term predictions of temperature extremes enable authorities and decision-makers to better prepare for heat waves, protect agricultural crops, manage energy demand, and reinforce critical infrastructure.

Broader implications for climate adaptation

Beyond improving scientific understanding of climate predictability, this research offers practical benefits for climate adaptation and risk management. The demonstrated skill in predicting temperature extremes up to two years ahead provides essential lead time for implementing adaptation measures and emergency preparedness strategies.

The framework’s ability to distinguish between different sources of predictability – external forcing versus internal variability – also contributes to improving the reliability and understanding of long-term climate forecasts.

As climate change continues to alter temperature patterns globally, innovative forecasting approaches become increasingly crucial for protecting communities and managing climate risks effectively.

CMCC played a central role in coordinating the study topic in collaboration with the National Center for Atmospheric Research and providing vital scientific guidance throughout the research process. The methodology and analytical software were developed by researchers working at CMCC in collaboration with Steve Yeager from NCAR.

For more information:

Eirini E. Tsartsali, Stephen G. Yeager, Panos J. Athanasiadis, Stefano Tibaldi, Silvio Gualdi, Predictability of temperature extremes in multi-annual forecasts, Environmental Research Letters, Volume 20, 2025, 104004, https://doi.org/10.1088/1748-9326/adfa3c