Accurately estimating Arctic sea-ice thickness has long been a challenge due to the region’s remoteness and the limitations of historical data. Recent research by Léo Edel and his colleagues has introduced a novel approach combining machine learning and data assimilation, establishing a breakthrough in this critical area of climate science. Typically, machine learning and data assimilation are used to look at the past, but with this new method we can now make predictions for the past where satellites could not measure sea-ice thickness.
The Arctic acts a little like a “canary in a coal mine”, like an early warning system for climate change. Changes in sea ice provide insights into widespread environmental impacts. Sea-ice thickness itself is a critical metric, influencing the interplay between the ocean, atmosphere, and ecosystems. Reliable, long-term records of sea-ice thickness enable better initialization of models, paving the way for more accurate predictions of Arctic conditions and their far-reaching effects on global climate systems.
Improving Data from Satellite Observations
Since the late 1970s, satellite missions from NASA and others have been providing invaluable datasets for understanding Arctic sea ice. However, reliable sea-ice thickness datasets have only been consistently available since 2011 with the introduction of the satellite product CryoSat-2 – SMOS (CS2SMOS, see info box). Satellite data prior to 2011 lacked the quality required for accurate assessments, leaving decades of historical records in need of significant correction.
Edel and his team addressed this gap by reconstructing Arctic sea-ice thickness data from 1992 to 2010, effectively extending the period of reliable observations from To achieve this, they utilized a hybrid model that combines machine learning algorithms with data assimilation techniques developed at the Nansen Center. This methodology represents the first application of such an advanced hybrid approach to a complex model with real-world observations.
A High-Resolution Dataset with Global Implications
The resulting dataset offers corrected Arctic sea-ice thickness information from 1992 to 2010, with a resolution of 10×10 km. This enhanced data not only reveals patterns of distribution and spatial variation in sea-ice thickness but also tracks changes over time. Such insights are crucial for improving long-term predictions of Arctic ice conditions, which carry significant global implications – ranging from sea-level rise to changes in weather and ocean systems.
This advancement also sets the stage for further applications, as the method holds potential to improve other key variables related to sea ice and ocean systems. By refining understanding of the Arctic’s past and present, this research creates a foundation for better-informed climate action on a global scale.
This work is part of the Nansen Center’s TARDIS project. Through collaborative efforts, the project leverages cutting-edge technologies to address critical gaps in climate research.
Key researchers: Léo Edel, Jiping Xie, Anton Korosov, Julien Brajard, Laurent Bertino
