Tech

How AI can help forecast how much Arctic sea ice will shrink

In the next week or so, the sea ice floating atop the Arctic Ocean will shrink to its smallest size this year, as summer-warmed waters eat away at the ice’s submerged edges.

Record lows for sea ice levels will probably not be broken this year, scientists say. In 2020, the ice covered 3.74 million square kilometers of the Arctic at its lowest point, coming nail-bitingly close to an all-time record low. Currently, sea ice is present in just under 5 million square kilometers of Arctic waters, putting it on track to become the 10th-lowest extent of sea ice in the area since satellite record keeping began in 1979. It’s an unexpected finish considering that in early summer, sea ice hit a record low for that time of year.

The surprise comes in part because the best current statistical- and physics-based forecasting tools can closely predict sea ice extent only a few weeks in advance, but the accuracy of long-range forecasts falters. Now, a new tool that uses artificial intelligence to create sea ice forecasts promises to boost their accuracy — and can do the analysis relatively quickly, researchers report August 26 in Nature Communications.

IceNet, a sea ice forecasting system developed by the British Antarctic Survey, or BAS, is “95 percent accurate in forecasting sea ice two months ahead — higher than the leading physics-based model SEAS5 — while running 2,000 times faster,” says Tom Andersson, a data scientist with BAS’s Artificial Intelligence lab. Whereas SEAS5 takes about six hours on a supercomputer to produce a forecast, IceNet can do the same in less than 10 seconds on a laptop. The system also shows a surprising ability to predict anomalous ice events — unusual highs or lows — up to four months in advance, Andersson and his colleagues found.

Tracking sea ice is crucial to keeping tabs on the impacts of climate change. While that’s more of a long game, the advanced notice provided by IceNet could have more immediate benefits, too. For instance, it could give scientists the lead time needed to assess, and plan for, the risks of Arctic fires or wildlife-human conflicts, and it could provide data that Indigenous communities need to make economic and environmental decisions.

Arctic sea ice extent has steadily declined in all seasons since satellite records began in 1979 (SN: 9/25/19). Scientists have been trying to improve sea ice forecasts for decades, but success has proved elusive. “Forecasting sea ice is really hard because sea ice interacts in complex ways with the atmosphere above and ocean below,” Andersson says.

In 2020, the sea ice in the Arctic shrank to its second lowest extent since satellite monitoring began in 1979. This animation uses those observations to show the change in sea ice coverage from March 5, when the ice was at its maximum, through September 15, when the ice reached its lowest point. The yellow line represents the average minimum extent from 1981 to 2010. Current forecasting tools can accurately predict these changes weeks in advance. A new AI-based tool can predict these changes with nearly 95 percent accuracy several months in advance.

Existing forecast tools put the laws of physics into computer code to predict how sea ice will change in the future. But partly due to uncertainties in the physical systems governing sea ice, these models struggle to produce accurate long-range forecasts.

Using a process called deep learning, Andersson and his colleagues loaded observational sea ice data from 1979 to 2011 and climate simulations covering 1850 to 2100 to train IceNet how to predict the state of future sea ice by processing the data from the past.

To determine the accuracy of its forecasts, the team compared IceNet’s outputs to the observed sea ice extent from 2012 to 2017, and to the forecasts made by SEAS5, the widely cited tool used by the European Centre for Medium-Range Weather Forecasts. IceNet was as much as 2.9 percent more accurate than SEAS5, corresponding to a further 360,000 square kilometers of ocean being correctly labelled as “ice” or “no ice.”

What’s more, in 2012, a sudden crash in summer sea ice extent heralded a new record low extent in September of that year. In running through past data, IceNet saw the dip coming months in advance. SEAS5 had inklings too but its projections that far out were off by a few hundred thousand square kilometers.

“This is a significant step forward in sea ice forecasting, boosting our ability to produce accurate forecasts that were typically not thought possible and run them thousands of times faster,” says Andersson. He believes it’s possible that IceNet has better learned the physical processes that determine the evolution of sea ice from the training data while physics-based models still struggle to understand this information.

“These machine learning techniques have only begun contributing to [forecasting] in the last couple years, and they’ve been doing amazingly well,” says Uma Bhatt, an atmospheric scientist at the University of Alaska Fairbanks Geophysical Institute who was not involved in the new study. She also leads the Sea Ice Prediction Network, a group of multidisciplinary scientists working to improve forecasting.

Bhatt says that good seasonal ice forecasts are important for assessing the risk of Arctic wildfires, which are tied strongly to the presence of sea ice (SN: 6/23/20). “Knowing where the sea ice is going to be in the spring could potentially help you figure out where you’re likely to have fires — in Siberia, for example, as soon as the sea ice moves away from the shore, the land can warm up very quickly and help set the stage for a bad fire season.”

Any improvement in sea ice forecasting can also help economic, safety and environmental planning in northern and Indigenous communities. For example, tens of thousands of walruses haul out on land to rest when the sea ice disappears (SN: 10/2/14). Human disturbances can trigger deadly stampedes and lead to high walrus mortality. With seasonal ice forecasts, biologists can anticipate rapid ice loss and manage haul-out sites in advance by limiting human access to those locations.

Still, limitations remain. At four months of lead time, the system was about 91 percent accurate in predicting the location of September’s ice edge.IceNet, like other forecasting systems, struggles to produce accurate long-range forecasts for late summer due, in part, to what scientists call the “spring predictability barrier.” It’s crucial to know the condition of the sea ice at the start of the spring melting season to be able to forecast end-of-summer conditions.

Another limit is “the fact that the weather is so variable,” says Mark Serreze, director of the National Snow and Ice Data Center in Boulder, Colo. Though sea ice seemed primed to set a new annual record low at the start of July, the speed of ice loss ultimately slowed due to cool atmospheric temperatures. “We know that sea ice responds very strongly to summer weather patterns, but we can’t get good weather predictions. Weather predictability is about 10 days in advance.”

Most Related Links :
newsbinding Governmental News Finance News

Source link

Back to top button