Spatio-temporal prediction model for sea ice density based on ConvLSTM network

Aiming at the problem that the traditional deep learning prediction model has low accuracy in predicting sea ice density and is difficult to capture the spatiotemporal variation of sea ice density, this paper constructed an improved spatiotemporal prediction model ConvLSTM to predict the Arctic sea ice density. A deep learning network was used to capture temporal and spatial changes in Arctic sea ice density over a 45-year period. Finally, the monthly average data of 2023 is output and displayed as the result. The results show that compared with the traditional prediction model, the proposed model is superior to the LSTM algorithm in prediction accuracy. LSTM network can overestimate the extent of sea ice density data to different degrees, especially in the melting season of sea ice, and ConvLSTM model has a better performance. The RMSE, MSE, MAE, R2 and SSIM of the model in this paper are 0.0644, 0.0043, 0.0145, 0.9318 and 0.9427, respectively.