Smart algorithms make climate predictions faster and more accurate

Climate scientists have a new tool at their disposal: machine learning. This technology, a type of artificial intelligence, allows researchers to better understand and model the complex climate system in ways that were not possible before. In an article published this week in Nature Reviews Physics, the researchers, including physical oceanographer Henk Dijkstra, demonstrate how machine learning is transforming climate research.

Machine learning can help to fill in gaps in climate data, making datasets larger, more complete, and globally consistent. More data leads to better insights into how the climate behaves and how it is changing. The technology also enables scientists to simulate processes in the climate system that traditional models often do not capture, such as cloud formation or turbulence. By including these “invisible puzzle pieces”, climate models become more realistic and accurate.

Perhaps most impressively, machine learning breaks through barriers in forecasting climate phenomena. Take El Niño, a weather pattern that can cause extreme droughts in Australia and flooding in Ecuador. Where traditional models often fail to predict beyond six months, machine learning can help make reliable forecasts up to 18 months in advance. This gives countries more time to prepare and mitigate potential damage.

According to Henk Dijkstra, Professor of Dynamical Oceanography at Utrecht University and one of the authors, the use of machine learning in traditional models marks a new step in climate research. In his ongoing ERC Advanced Grant project (2022-2027), focused on the tipping point of the Atlantic Meridional Overturning Circulation, he uses machine learning to reconstruct historical ocean circulation patterns and predict future changes.