The future of extreme events predictions
CMCC Foundation - Centro Euro Mediterraneo sui Cambiamenti Climatici
The Foundation Euro-Mediterranean Centre on Climate Change (CMCC) is a research centre on climate science and policy
Global leading experts explore past and future predictions for hurricane activities and medium-range weather. Gabriel Vecchi, from Princeton University, covered his latest research on past and future changes in tropical cyclone formation during a recent CMCC Lecture. Vecchi explored the complex interplay between climate change and future cyclone frequency. His work is essential to the ongoing scientific debate aimed at refining cyclone predictions to improve preparedness and prevent future impacts.
Vecchi also addressed how AI and Machine Learning can contribute to advancements in the fields of climate predictions, a central topic to a previous CMCC Lecture by Rémi Lam, recently included in Nature’s top 10 Scientists of 2024, with the pioneering GraphCast model by Google DeepMind.
During the CMCC lecture titled “Linking past and future hurricane activity changes,” Gabriel Vecchi—who is Knox Taylor Professor of Geosciences, Director of the High Meadows Environmental Institute (HMEI), and Deputy Director of the Cooperative Institute for Modeling the Earth System (CIMES) at Princeton University—addressed the contrasting results provided by existing studies regarding future cyclone frequency in response to global warming. Climate scientists, including Vecchi’s research group, are investigating the reasons behind such variability in model projections and exploring alternative methods, such as moving away from traditional indices and studying multi-stage processes of cyclone genesis, to reduce uncertainty and improve our ability to plan for the future.
Improving predictions, improving preparedness
Models have shown consistency in past predictions in response to warming but diverging outcomes for future trends. A study from Vecchi’s group confirmed that historical tropical cyclone frequency does not depend on mean warming, but rather on specific patterns of ocean temperature change. As noted by Vecchi, “Historical records do not serve as an adequate test of what the response of hurricane frequency to warming is.” To determine the impact of climate change on future cyclone frequency, a breakthrough could be achieved through indirect methods, such as analyzing how the El Niño-Southern Oscillation (ENSO) phenomenon (El Niño and La Niña events) influences hurricanes, to then identify what model behaves more like observations.
Another key aspect of cyclone risk management is understanding whether the frequency of the most extreme events will increase as the planet warms. Hurricanes with a speed exceeding 50 meters per second, classified as ‘major hurricanes,’ cause the vast majority of severe impacts. As mentioned by Vecchi, there is a broad scientific agreement concerning a predicted increase in major hurricanes due to climate change, particularly in certain coastal areas of the United States and the Caribbean, with possible unprecedented impacts on human societies.
New frontiers
To enhance accuracy, the use of AI and Machine Learning in extreme event predictions has expanded significantly over the past few years. Rémi Lam, a researcher at Google DeepMind and one of Nature’s Top 10 scientists of 2024, presented his innovative approach to medium-range weather prediction in the CMCC Lecture “GraphCast: Learning skillful medium-range global weather forecasting.” Lam developed a model based on a global grid neural network trained on atmospheric data, called GraphCast, which obtained remarkably accurate results.
When asked about the impact of AI and machine learning on the study of cyclones, Vecchi focused on how these new tools can contribute to science advancements by integrating with dynamical modeling: “They are an important element of our tools going forward,” Vecchi said in his CMCC Lecture. “They need to be part of a larger universe where dynamical understanding underpins how we interpret things, where we invest in improving our observations and our dynamical modeling techniques, where we work into the hybrid dynamical statistical systems, where the statistical system is AI machine learning.”
CMCC work is actively contributing to this newly developing field by participating in international research initiatives such as the CLINT project, which aims to develop an artificial intelligence framework to analyze large datasets with a combination of machine learning techniques and algorithms. Within the CLINT project, researchers are enhancing the detection, causation, and attribution of these extreme events and improving genesis potential indices through machine learning.
Precipitation trends
Even though there is still uncertainty about hurricane frequency in a changing climate, their intensification is now commonly agreed upon. As reported by Vecchi, rainfall is expected to increase by 7% for each degree of warming. The amount of rainfall associated with future tropical cyclone genesis is vital information for the stakeholders involved in risk prevention and management worldwide, as well as for governments and administrations aiming to plan a long-term adaptation strategy. CMCC researchers assessed the reliability of observational and reanalysis datasets to represent precipitation trends, with a focus on Hurricane Freddy, one of the most intense recent tropical cyclones, in terms of energy discharged. The study suggests that observational data should be used with caution in trend analysis, especially when incorporating new satellite data over time.
More information: