Simulation And Prediction Of Global Sea Ice Data Based On Deep Learning

The sea ice data is one of the inputs of the earth numerical prediction model,which can directly affect the accuracy of the model system’s prediction results of sea surface wind field and pressure field in middle and high latitudes.With the intensification of global warming and Arctic amplification in the past two decades,the global total sea ice has shown an accelerated downward trend,and the changes of relevant sea ice indexes mainly based on sea ice concentration are abnormal.The current sea ice prediction scheme based on coupled model can accurately simulate the physical change process of arctic sea ice,but the physical parameterization construction scheme of dynamic and thermal processes of sea ice model is complicated,and the quality of multi-source input data in this scheme lacks independent control means and redundant design.In this paper,based on deep learning technology,the numerical distribution prediction of monthly sea ice concentration was studied.By constructing the sea ice area prediction model(LSTM-SIA)and the sea ice concentration prediction model(CONVLSTM-SIC)respectively,and using the mathematical relationship between the prediction results,the feasibility of deep learning algorithm model in sea ice data simulation and prediction scheme is verified.Aiming at the problem of sea ice area prediction,this paper firstly calculates the global monthly sea ice area in recent 40 years by using the preprocessed sea ice concentration reanalysis data,and constructs the time series data set of arctic/antarctic sea ice area in groups.The sea ice area prediction model LSTM-SIA is constructed based on LSTM algorithm.The model is used to train the sea ice area data set,and the global monthly sea ice area value from January to June 2021 is output by iterative prediction.The evaluation indexes of the predicted results show that the LSTM-SIA model can accurately predict the monthly trend of sea ice area in the Arctic and Antarctic,and the prediction sequence of sea ice area can be used as an important reference for the prediction results of sea ice concentration.For the prediction of sea ice concentration,the compressed data set and the block data set of the arctic/antarctic sea ice concentration are constructed in groups.In this paper,the Conv LSTM-SIC model was built based on Conv LSTM algorithm to train the constructed sea ice concentration data set,and the prediction results of compressed sea ice concentration data in different starting months from 6 to 12 months and the global sea ice concentration prediction results from January to June 2021 were obtained by iterative prediction.The evaluation index of sea ice concentration prediction results shows that Conv LSTM-SIC model can well iteratively predict the global monthly sea ice concentration distribution,and the prediction results in the Arctic are better than those in the Antarctic.At the same time,the sea ice area series obtained by calculating the predicted results of global sea ice concentration is consistent with the predicted series of LSTM-SIA model in trend,which indicates that the two models have good applicability in processing the time series data with obvious periodic characteristics such as sea ice.