The Arctic's inhabitants have relied on the seasonal sea ice for generations. It grows and recedes throughout the year. It is a key habitat for marine mammals and polar bears. According to an Arctic Council report, the Arctic water and air have warmed three times faster since 1971 than the rest. This is making the ice expand and contract in unpredictable ways.
Artificial intelligence is being used by some researchers and scientists to forecast when and where the Arctic Ocean will become ice-covered. AI algorithms are used in conjunction with existing models that use Physics to analyze what is happening at the oceans' surface. This dynamic zone is where cold underwater currents meet strong winds to form floating rafts. This information is increasingly valuable to tribal members in Arctic, commercial fishermen in Alaska and global shipping companies that are interested in finding shortcuts through open areas of water.
Leslie Canavera is the CEO of Polarctic. A Lorton, Virginia-based scientific consulting firm, Polarctic has developed AI-based forecast modeling models. He says that current models of sea ice are becoming less accurate due to climate change's unpredictable pace. Because they are based upon environmental processes that are rapidly changing, this is why they are becoming less accurate.
Canavera, a Yupik tribal member who grew up in Alaska, said that we don't have a good understanding of climate change or what's going on in the Arctic. You can use statistical modeling to look at the averages, but that's not enough. Artificial intelligence is a form of artificial intelligence that can see trends and learn from them.
Existing physics-based models can capture hundreds of years worth of scientific data about ice conditions, current meteorological circumstances, the speed and position of the polar jetstream, cloud cover and ocean temperature. These data are used to predict future ice cover. It takes a lot of computing power to crunch these numbers and many hours or days to create a forecast with conventional programs.
Although AI requires complex data and lots of initial computing power, once the algorithm has been trained with the right data, it is able to detect patterns in climate conditions faster than physics-based models. This was according to Thomas Anderson (a British Antarctic Survey data scientist who created an AI ice forecast called IceNet). Anderson claims that AI methods can run thousands of times faster than traditional models, such as the one we used in IceNet. They also learn automatically. AI isn't smarter. AI is not going to replace physics-based models. My view is that the future will be based on both information sources.
Anderson and his collaborators published their sea ice forecast model in Nature Communications in August. IceNet employs deep learning, an AI form that automates detection of credit card fraud and operates self-driving vehicles. It also runs personal digital assistants. The model is trained to give a six-month forecast for each 25-kilometer square in the region. This forecast is based on actual observational data from 1979 to 2011 and simulations of the Arctic climate. IceNet beat the leading physics-based model when it was trained, and given current ocean and meteorological conditions. This is especially important for summer, when sea ice retreats annually, according to the Nature Study.
