AI sheds light on glacier retreat in the high Arctic

It’s no surprise that glaciers are shrinking as the climate warms—especially in Svalbard, Norway, which is experiencing some of the fastest warming on the planet. Glaciers that flow into the ocean, known as marine-terminating glaciers, have long been difficult to study because of their remote location and complicated behaviours. But as TIAN LI writes, artificial intelligence (AI) is helping scientists better understand how and why these glaciers are changing.

Glaciers are slow-moving rivers of ice formed from years of accumulated snowfall. They store large amounts of freshwater and serve as sensitive indicators of climate change. When temperatures rise, glaciers lose mass by melting and calving (the term for when large chunks of ice break off into the ocean).

These changes can have significant and far-reaching consequences for marine ecosystems and coastal communities. For example, glaciers help supply freshwater to people and wildlife, and the meltwater they produce can deliver important nutrients to downstream environments. But glacier meltwater also contributes to sea level rise, increasing the risk of coastal erosion and flooding and endangering low-lying communities.

The Arctic is especially vulnerable to these effects. Its glaciers, particularly those that are marine-terminating, are experiencing some of the most dramatic and rapid changes anywhere on the planet.

Watching glaciers from space using AI

Studying glacier calving fronts in the high Arctic is no easy task. Fieldwork is challenging, expensive and often limited by the harsh and remote environmental conditions. Fortunately, satellite remote sensing is revolutionizing this work. Earth observation satellites can capture high-resolution images of the Earth’s surface, providing both optical and radar images that reveal detailed information at glacier fronts.

However, the sheer volume of satellite imagery now available presents a new challenge: How do we efficiently extract meaningful information from millions of images?

To tackle this, we trained a deep learning model to detect calving fronts automatically under a wide range of environmental conditions and for a variety of glacier types. By teaching the model to interpret both optical and radar images, we enabled it to identify calving fronts across diverse settings with high accuracy. We then [1] applied this model to analyze more than a million satellite images of 149 marine-terminating glaciers in Svalbard using open-access data from Google Earth Engine. This allowed us to track changes in glacier fronts from 1985 to today, offering unprecedented detail about how glaciers in the high Arctic have evolved.

What the model is telling us

We found that 62 per cent of marine-terminating glaciers in Svalbard experience seasonal cycles—that is, the calving fronts retreat in summer and advance in winter. Ocean temperatures have a large impact on peak seasonal retreat rates, with glacier retreats on the west coast occurring before those on the east coast, likely because warm ocean waters carried by the West Spitsbergen current arrive at different times in different areas.

These seasonal changes show us just how sensitive and important interactions between the ocean and ice are. Over longer time periods, we observed that across Svalbard, 91 per cent of glaciers have been shrinking significantly since 1985. The peak retreat rate occurred in 2016, when the weather was unusually warm, possibly because of an extreme weather pattern called atmospheric blocking, which traps warm air. Over the past 40 years, atmospheric blocking events have been occurring more frequently and with greater intensity. This is mainly in response to amplified Arctic warming. We expect that glacier retreat rates will accelerate in the future, causing even more ice loss.

The behaviours of marine-terminating glaciers in Svalbard offer valuable insights into what may lie ahead for glaciers across the Arctic. What we are seeing now is just the start—and our AI model can be applied to glacier systems in other locations. Thanks to tools like this, we are gaining insights into glacier dynamics in unprecedented detail. What we discover could ultimately inform efforts to reduce the risks that glacier retreat poses to vulnerable coastal communities.