Prediction of Arctic Sea Ice Concentration Using Multi-Model Ensemble with Deep Neural Network

Sea ice maintains the Earth's temperature by reflecting solar energy and keeping the polar regions cool. However, due to the recent accelerating global warming, the area of sea ice has been declining, so observing and forecasting sea ice is very necessary to understand global climate change. In this study, we predicted sea ice concentration (SIC) using regional climate model (RCM) data with high spatial resolution and deep neural network (DNN), which can deal with non-linearity problem as a kind of machine learning, for more accurate prediction. In order to improve the accuracy, we used nine RCM data provided in the North Pole as ensemble members to produce ensembles of input variables and use it as input data. We built two types of models using multiple linear regression (MLR) and DNN using RCM and SIC data between January 2006 and December 2016. The accuracy of the model constructed using ensembles Bayesian Model Averaging (BMA), which employs a weighting scheme for each member using posterior probability, is much higher than that of the model constructed using only one RCM data. Also, the DNN model showed best performance among them with the correlation coefficient 0.884 and the root mean square error (RMSE) of 0.199. Finally, we tried to examine the climate change in the near future by predicting the SIC from 2017 to 2030 using RCM data and DNN model which showed high accuracy.