Akademik Veri Yönetim Sistemi
Earth Science Informatics, cilt.19, sa.6, 2026 (SCI-Expanded, Scopus)
Estimation of sea ice concentration (SIC) is not only important for the Earth’s temperature and water balance, but also for the protection of human activities such as ship navigation, oil and gas exploration, and fishing in polar regions. The decrease in SIC directly affects global climate change. Knowing in advance how sea ice will change over time is important in terms of taking the necessary precautions and preventing undesirable situations. In this study, a convolutional long short-term memory (ConvLSTM) model is designed to predict SIC, surface albedo (SAL), and fractional cloud cover (CF) of the South and North Poles for different seasons and months in 2030 and 2040. The model is trained using 46 years of satellite observation data and generates 16-year predictions of SIC, SAL, and CF using a sliding-window strategy. The results indicate a substantial contraction of sea ice concentration in both hemispheres toward 2030 and 2040, consistent with observed historical trends. The proposed framework utilizes artificial intelligence and perceptual metrics to learn spatiotemporal patterns in SIC, SAL, and CF. Therefore, produced outputs should be interpreted as a data-driven scenario extension of satellite observations rather than a deterministic climate prediction. The best model achieved a Structural Similarity Index Measure (SSIM) of 0.9682, a Deep Image Structure and Texture Similarity (DISTS) score of 0.0662, and a Mean Absolute Error (MAE) of 0.0426 in the validation phase for single-year prediction. When predicting the year 2024 using data from 1986 to 2008, the model achieved an SSIM of 0.9239, a DISTS score of 0.1087, and an MAE of 0.0623.