부산대학교 도서관

This paper presents a novel deep learning model for short-term sea ice forecasting that accounts for the effects of teleconnection and delay phenomena on sea ice changes. The proposed model is a convolutional neural network with an embedded attention mechanism that takes into account the past 21 days of Arctic environmental element data as input. It is capable of providing daily forecasts of sea ice concentration in the Arctic core area for the next 49 days. Notably, the model outperforms both the deep learning sea ice forecast model Icenet and the numerical sea ice forecasting models ArcIOPS and SEAS5, demonstrating its superior short-term forecasting capability for sea ice.