A new age of Arctic science discovery – the AI way

Thursday 23 Apr 2020

Filed under

When we see news reports on climate change on our TV, they are often accompanied by footage of a polar bear walking over the icy Arctic landscape. But the Arctic is more than just the iconic polar bear, it consists of a complex ecosystem that has developed, and thrived, in what we would consider some of the harshest conditions on Earth. However, the Arctic is under threat like never before, and today, bringing the power of AI to our rich environmental datasets is opening up new scientific insights into our natural world.

Polar expedition boat
Alfred-Wegener-Institut / Michael Gutsche (CC-BY 4.0)

Ships, submarines and satellites - A brief history of sea ice measurements

The force behind these changes is human-induced climate change, and the Arctic is at the frontline of climate change. Due to complex feed-back mechanisms between the atmosphere, ocean and land the Arctic is warming twice as fast as the rest of the planet (this is known as Arctic Amplification). Even if we are able to limit our emissions to 2 C below pre-industrial levels the Arctic could warm by 4-5 C. This could have serious consequences for the region and for the planet.

Perhaps the biggest change to the region over recent decades has been to the frozen Arctic Ocean which has seen a loss of almost half the summer sea ice since the early 1980s. The most substantial change however is loss of the thick sea ice that usually circulates within the Arctic Ocean for many years (known as multiyear ice). About 40 years ago this accounted for around 70% of the ice within the Arctic, now it is only 30%, and by the middle of the century we could see ice free summers.

Over the years our scientific observations and measurements have increased in number and sophistication. First, we had occasional measurements from ice-strengthened ships in summer which allowed scientists to make measurements at one particular time and region. The sea ice data sources increased with the advent of nuclear submarines and their under sea ice Arctic operations within the cold war and beyond. In fact, it was the declassification of their sea ice datasets and subsequent interpretation by scientists that first notified the world that the Arctic sea ice was melting. However, the game changer for Arctic observations was scientific sensors mounted on satellites.

The robotic ‘Arctic’ revolution

Following the first satellite photographs of Earth in the late 1950s, the number and variety of space-borne sensors continued to increase and have provided an unprecedented level of information regarding the different properties of sea ice. Just as the satellite revolution was well advanced, a robotic ‘Arctic’ revolution was starting out. To understand the Arctic, scientists needed a long-term monitoring presence in the region. However, the environment of the Arctic Ocean is too harsh, too logistically challenging, and too expensive for a continued year-round human presence.

We needed robotic platforms that could take year-round measurements at a higher temporal frequency that was not possible with satellites. We now have the possibility to deploy autonomous robotic platforms that collect data from the atmosphere, ocean and sea ice right across the Arctic Ocean. At present however, there are not enough of these systems within the Arctic Ocean. Furthermore, there are measurements that these robotic platforms cannot make and so at the end of 2019 an international partnership launched the MOSAiC campaign - the largest ever Arctic expedition - which has taken a German research icebreaker and allowed it to become an international floating research laboratory by freezing it into the sea ice pack for over a year.

In short, the observational data available to us is growing substantially and is rich and varied, helping us to understand physical processes on scales ranging from centimetres to kilometres. The challenge is how we can be smarter in order to get the best knowledge from these diverse datasets. Artificial intelligence (AI) provides the opportunity to unify these datasets in order to provide a holistic picture of many different Arctic processes, including sea ice. With the recent acceleration of data acquisition, we enter into a new dawn of polar (data) exploration – albeit from the comfort of our desks.

Application of AI algorithms

Since the middle of 2019, Researchers at the Turing and British Antarctic Survey have been harnessing powerful AI algorithms to uncover hidden relationships within the sea ice data that could not be revealed by traditional data analysis methods. With tens of millions of data points from our satellite-derived datasets, we are training our AI algorithms to predict future sea ice, with the ability to learn physical relationships between climate variables over both space and time. Furthermore, using AI explainability methods - which provide us with the ability to interpret the predictions - allows us to ‘open up the black box’ of the AI and draw conclusions on what it has learned from the data, potentially providing new scientific insights.

AI has the potential to make giant leaps in our physical understanding of our changing planet. If we are to tackle the climate emergency head-on the time has clearly come to integrate AI approaches into our ongoing environmental research efforts.

Read our researcher spotlight on blog author Scott Hosking and watch him host the virtual Earth Day Turing Lecture: Plan AI, because there is no Planet B.

Cover photo credit: Stefan Hendricks