Arctic Ice Thickness Prediction using Artificial Neural Networks

This paper presents an artificial neural network model to predict Arctic ice thickness and that can be used as a decision making tool for Arctic navigation. Ice thickness historical data from Copernicus database are collected and preprocessed and used to train and validate the artificial neural network. A series of experiments were developed using daily ice thickness values as an input of the model versus using monthly ice thickness values, applying the model on four different geographical areas, and increasing the training and testing percentages. The results show that the model developed based on daily input data is good for predicting short-term periods with low mean square error values but failed in predicting ice thickness values for mid-term periods. The model with monthly input data performed better for mid-term prediction and in three out of the four considered areas. Moreover, increasing the percentage of training data resulted in a reduction of MSE values by more than 40% in some areas.