Nov 3, 2021 9:00 AM

As the Arctic Warms, AI Forecasts Scope Out Shifting Sea Ice

Global warming is making it harder to predict the movement and location of the ice cover, crucial information for fishing and global shipping.

The AI Database →

Application

Prediction

Technology

Machine learning

Neural Network

For generations, the inhabitants of the Arctic have counted on seasonal sea ice, which grows and retreats during the year. Polar bears and marine mammals rely on it as a hunting spot and a place to rest; Indigenous people fish from openings in the ice known as polynyas, and use well-known routes across the ice to travel from place to place. But the Arctic air and water has warmed three times faster than the rest of the planet since 1971, according to a May 2021 report by the Arctic Council, and this warming is causing the ice to expand and contract in unpredictable ways.

Some scientists and research firms are now deploying tools powered by artificial intelligence to provide more accurate and timely forecasts of what parts of the Arctic Ocean will be covered with ice, and when. AI algorithms complement existing models that use physics to understand what’s happening at the ocean’s surface, a dynamic zone where cold underwater currents meet harsh winds to create floating rafts of ice. This information is becoming increasingly valuable for tribal members in the Arctic, commercial fishers in places like Alaska, and global shipping companies interested in taking shortcuts through open patches of water.

Leslie Canavera, CEO of Polarctic, a Lorton, Virginia–based scientific consulting firm that has developed AI-based forecast models, says the uncertain pace of climate change means that existing models of sea ice are becoming less accurate. That’s because they are based on environmental processes that are quickly shifting.

“We don't have a great understanding of climate change and what's happening in the [Arctic] system,” says Canavera, who is a Yup’ik tribal member and grew up in Alaska. “We have statistical modeling, but then you're looking at more of the averages. Then you have artificial intelligence, where it's able to see the trends in the system and learn.”

Existing physics-based models capture hundreds of years of scientific records about ice conditions, current meteorological conditions, the speed and location of the polar jet stream, the amount of cloud cover, and ocean temperature. The models use that data to estimate future ice coverage. But it takes large amounts of computing power to crunch the numbers, and several hours or days to produce a forecast using conventional programs.

While AI also requires complex data and a lot of initial computing power, once an algorithm is trained on the right amount and kind of data, it can detect patterns in climate conditions more quickly than physics-based models, according to Thomas Anderson, a data scientist at the British Antarctic Survey who developed an AI ice forecast called IceNet. “AI methods can just run thousands of times faster, as we found in our model, IceNet,” Anderson says. “And they also learn automatically. AI is not smarter. It's not replacing physics-based models. I think the future is leveraging both sources of information.”

Anderson and his colleagues published their new sea ice forecast model in August in the journal Nature Communications. IceNet uses a form of AI called deep learning (also used to automate detection of credit card fraud, operate self-driving cars, and run personal digital assistants) to train itself to provide a six-month forecast in each 25-kilometer square grid across the region, based on simulations of the Arctic climate between the years 1850 to 2100 and actual observational data recorded from 1979 to 2011. Once the model was trained and given current meteorological and ocean conditions, IceNet beat a leading physics-based model in making seasonal forecasts about the presence or absence of sea ice in each grid square, particularly for the summer season, when the ice goes through an annual retreat, according to the Nature study.

A separate team of scientists at the Johns Hopkins University Applied Physics Laboratory has developed a forecast model that uses a form of AI called convolutional neural networks to examine satellite images of the ocean surface and make predictions of how quickly ice will form in the coming week. Neural networks are able to sort digital pixels more quickly than humans, and are used in facial recognition algorithms, for example. The JHUAPL model uses digital satellite images and combines them with meteorological data that is collected on the ground at the same time , according to Christine Piatko, a senior staff scientist at the laboratory and principal investigator on the project.

Right now, forecasters at the US National Ice Center in Colorado compile weekly Arctic ice forecasts by hand, analyzing images taken by orbiting satellites and comparing them with historic data. But that method might not be good enough now that the Arctic Sea is rapidly losing its ice cover. In fact, it may be completely ice-free in the summer months by 2050, according to estimates by a group of 21 research institutions published in 2020.

The opening of the Arctic Sea means more ship traffic, and more ships means better forecasts are needed, Piatko says. Until now, physics-based models and manual estimates of ice coverage have been adequate. Forecasters “only had to make forecasts for a few ships or for special missions,” Piatko says. “But as there's increased activity, you can imagine different scenarios where they might need information in a more timely manner. We are trying to anticipate that need.”

Polarctic’s Canavera is collaborating with Canadian officials to develop ice forecasts for residents of the territory of Nunavut, and develop better understanding of critical food resource areas that are changing thanks to climate change. A separate project called SmartIce uses data from small battery-powered sensors embedded in the sea ice that record both temperature and ice thickness, and that information can be used to aid navigation and keep Indigenous people safe.

Her firm is also developing localized sea ice forecasts for Alaskan commercial harvesters who want to fish at the edge of the ice, which is a productive area for cod and crabs. “You need a strong forecast of where the ice edge is going to be and how it's going to change. Because if the ice comes in too fast, or the forecast is wrong, then the fishermen can lose their fishing gear—it becomes ghost plastic in the ocean, and they're out of profits,” Canavera says. “We’re hoping to solve that problem and develop a solution.”

Forecasting sea ice is just one application of artificial intelligence as scientists try to understand the changing climate. AI algorithms can also be deployed to forecast electric power supply, demand, and carbon dioxide emissions; automate detection of methane leaks; and even predict improvements in energy efficiency of office buildings and homes, according to a 2019 paper by a group of 22 renowned computer scientists presented at the world’s leading AI conference, known as NeurIPS.

Anderson and his Icenet colleagues are aiming to boost the IceNet model’s accuracy to forecast ice conditions at greater resolution down to grids that are only several hundred meters across rather than 25 kilometers. But still, he says that AI models are no substitute for the on-the-ground knowledge of Arctic coastal residents. “Nothing beats being on the shoreline and saying, ‘You know, I cannot go out onto the sea ice today, because the sea ice is too thin or there's no platform,’” he says. “But what these predictions can do is give people a general sense of, okay, what is the trajectory of the ice in the surrounding area? That's filling a really big and important gap, where advances in AI forecasting could make a huge difference.”

Update 11.3.2021 at 1:32 PM: This story was updated to correct the reference to SmartIce.