Researchers from the UK The deep learning model was able to predict the spread of sea ice in the Arctic for several months with an accuracy of 95 percent. The article of scientists has been published in the journal Nature Communications.
predict what the spread of sea ice in the Arctic is difficult. In this case, due to the sensitivity to growing due to global warming temperatures, its area in the summer over the past four decades has fallen half.
Researchers from the Antarctic Great Britain and Institute of Alan Tyurring developed a model of deep learning Icenet consisting of several convolutional neural networks. It was trained – both on climate models and on the satellite monitoring of the ice cover of the Arctic – to create ice distribution maps for the next six months.
Artificial intelligence (AI) with an accuracy of 95 percent predicted the spread of sea ice in the next two months – which is better than traditional dynamic models building their forecasts only in physical laws. What is important, Icenet is much faster: in comparative tests, she worked on a laptop two thousand times faster than the dynamic model on a supercomputer.