Monthly Arctic Sea Ice Concentration Prediction Based on the ConvLSTM Model

Sea ice concentration (SIC) reanalysis data is important inputs to the coupled ocean-atmosphere system (COAS). Using simple and fast methods to forecast arctic SIC in the mid-long term will be conducive to the construction of the COAS. In this paper, the convolution LSTM model (ConvLSTM) was used to iteratively predict the monthly series of arctic SIC reanalysis data for 6 to 12 months. B y comparing the root mean square error (RMSE) of SIC distribution in the process of monthly prediction and iterative prediction, this paper preliminarily demonstrates the influence of different seasons on the iterative prediction accuracy of the model, and evaluates the feasibility of ConvLSTM model in the prediction of periodic tensor data such as sea ice. On the compressed dataset, the RMSE of the SIC distribution varies from 0.06 to 0.15, with an average of 0.1. The experimental results verify the prediction ability of ConvLSTM model for the periodic change of monthly tensor data of SIC, that is, the fluctuation of iterative prediction results affected by seasons is small, and the prediction deviation of freezing period is slightly higher than that of melting period and is within an acceptable range. In addition, in order to obtain a high-resolution prediction matrix, it is necessary to add a reasonable block mechanism to the model and evaluate the requirements of ConvLSTM model for computing resources and environment.