Researchers at the Mubadala Arabian Centre for Climate and Environmental Sciences (ACCESS) at NYU Abu Dhabi have developed a new algorithm that can forecast Arctic sea ice extent up to nine months in advance, offering a new way to anticipate changes in the Arctic that can affect the global climate system.

Called the Random Analogue Predictor (RAP), the algorithm uses historical sea ice data to identify past patterns that resemble current conditions and uses what happened next to generate possible future forecasts. RAP also provides an estimate of uncertainty for each forecast.

Arctic sea ice plays an important role in the global climate system because it reflects solar energy back into space, while the darker ocean absorbs it. Changes in Arctic sea ice can also influence atmospheric and oceanic patterns far beyond the region, making the ability to forecast these changes in advance valuable for better understanding wider climate impacts.

“Forecasting Arctic sea ice several months ahead is a difficult problem, and an increasingly important one as the Arctic continues to change,” said Francesco Paparella, inaugural director of Mubadala ACCESS at NYU Abu Dhabi and senior author of the study. “Our approach is deliberately simple, but it performs competitively with much more complex forecasting models. Importantly, it also provides an estimate of its own uncertainty, making it a useful and transparent benchmark for evaluating future forecasting methods.”

The researchers found that RAP produced forecasts with a level of skill comparable to models used by the Sea Ice Prediction Network (SIPN). For September sea ice extent, its forecast error was comparable to that of 34 models used for seasonal forecasting.

Unlike physics-based models, which simulate the atmosphere, ocean and sea ice, RAP uses only the historical record of Arctic sea ice extent. By identifying similar past conditions, it generates an ensemble of possible forecasts. The spread of these forecasts provides an estimate of uncertainty, allowing users to assess how much confidence to place in a prediction.

The researchers propose RAP as a benchmark for both physics-based and AI-driven models. Its simplicity and interpretability provide a clear reference point for assessing more complex approaches.