Stepwise Regression Predicts Arctic Sea-Ice Extent with High Accuracy
The Arctic is undergoing a dramatic transformation due to global warming, shifting from multi-year thick ice to a "New Arctic" dominated by first-year thin ice. This younger ice is more susceptible to melting, posing challenges for sea-ice prediction and the stability of the ice cover.
Accurate sea-ice forecasting is crucial for understanding the climate system and ensuring safe Arctic navigation. However, due to the complex interplay of atmospheric, oceanic, and other factors, precise prediction remains a significant international research focus.
A recent study by Associate Professor Baoqiang Tian from the Institute of Atmospheric Physics, Chinese Academy of Sciences, and Professor Ke Fan from Sun Yat-sen University introduces a novel real-time prediction method for September Arctic sea-ice extent. This method combines initial sea-ice conditions with thermodynamic and dynamic processes, utilizing stepwise regression to select effective predictors and incorporating the interannual increment approach.
The research, published in Atmospheric and Oceanic Science Letters, demonstrates the method's high predictive performance for September Pan-Arctic sea-ice extent. When compared to LSTM (long short-term memory) neural networks, the new approach exhibits smaller prediction errors and greater stability in independent tests from 2014 to 2022. Its prediction accuracy even surpasses that of the forecasts released by the Sea Ice Outlook.
Despite LSTM's strong performance during training, its real-world prediction robustness is inferior to the new method. This limitation may be attributed to the limited availability of sea-ice data, which can lead to overfitting in complex machine learning models. Professor Ke Fan explains that their prediction method not only considers the independence of predictors to avoid overfitting but also amplifies the prediction signal through the interannual increment approach, enhancing the model's predictive capability.
The study's findings highlight the potential of stepwise regression in predicting Arctic sea-ice extent, offering valuable insights for climate research and navigation safety in the rapidly changing Arctic environment.
