Multivariate deep learning approach to forecast of an ice-free Arctic summer under the SSP2-4.5 and SSP5-8.5 scenarios

BİLGİLİ M.

Polar Science, 2025 (SCI-Expanded, Scopus) identifier

  • Yayın Türü: Makale / Tam Makale
  • Basım Tarihi: 2025
  • Doi Numarası: 10.1016/j.polar.2025.101308
  • Dergi Adı: Polar Science
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Artic & Antarctic Regions, BIOSIS, Geobase, INSPEC, Linguistic Bibliography
  • Anahtar Kelimeler: Arctic, Ice-free Arctic, LSTM neural network, SARIMA model, Sea ice extent
  • Çukurova Üniversitesi Adresli: Evet

Özet

Sea ice is a basic element of the polar weather and climate system. Arctic sea-ice extent (SIE) has significantly decreased in recent times due to global warming and enhanced ice-albedo feedback. Accurate prediction of Arctic SIE is therefore critical for understanding climate dynamics, assessing environmental impacts, and informing sustainable resource management. Conventional approaches, reliant on physical models, have exhibited deficiencies in replicating the intricate patterns of ice behavior. On the other hand, deep learning approaches have emerged as a potent solution for complex pattern recognition and prediction tasks. This paper presents a thorough review and analysis of using univariate and multivariate long short-term memory (LSTM) neural networks and a seasonal autoregressive integrated moving average (SARIMA) model for Arctic SIE prediction. The proposed methods utilize historical observations of monthly SIE data from 1989 to 2024 derived from the National Snow and Ice Data Center (NSIDC) database. These models are then used to produce projections for the period 2025 to 2100. The results show that under the multivariate LSTM model/SSP2-4.5 and SSP5-8.5 scenarios, the Arctic will be ice-free in September 2072 and 2065, respectively.