A Study of Deep Learning Algorithms for Long-term Prediction and Correlation Identification of Arctic Ice

• Main Authors: Liu, Jiahao, Feng, Yuan, Song, Shengyu, Xu, Yuanxiang
• Format: Conference Proceeding
• Language: English
• Subjects: Arctic; Data models; Deep learning; demarcation of the sea ice edge; Livelihood; long-term prediction; multi-factor ocean data; Oceans; Prediction algorithms; Predictive models; sea ice concentration
• ISSN: 2768-1904
• DOI: 10.1109/CSCWD61410.2024.10580757

Over the past few decades, global warming has accelerated the rate of polar sea ice melt, taking a toll on local ecosystems and the livelihoods of indigenous peoples. Early prediction of sea ice anomalies can help reduce negative impacts and prevent potential disasters. Therefore, this paper uses a prediction model based on the Temporal Convolutional Network (TCN) to provide assistance for the development of marine meteorology. We used the TCN-based prediction model to analyze sea ice concentration data in the region from 60° to 90°N and 180°W to 180°E. Firstly, 64 years (from January 1959 to September 2022) of ERAS reanalysis data was used to predict Arctic sea ice as a single variable. Then, based on the research conclusions of relevant literature, multi-source ocean data related to Arctic sea ice changes were selected and added to the model to achieve multi-variable prediction. This paper not only predicts Arctic sea ice concentrations, but also identifies and verifies factors associated with Arctic sea ice changes from deep learning remote sensing data by observing changes in RMSE, MAE and Binary accuracy.