Fig. 1: The IceNet model.
From: Seasonal Arctic sea ice forecasting with probabilistic deep learning
IceNet receives 50 monthly averaged climate variables as input (Supplementary Table 2), centred on the North Pole. IceNet, a deep learning U-Net model, receives these inputs and processes them through a series of convolutional blocks with batch normalisation (Supplementary Table 1). The number to the left of the convolutional blocks denotes the number of feature maps in each convolutional layer, while the number beneath denotes the feature map resolution. IceNet’s outputs are forecasts of three sea ice concentration (SIC) classes (SIC ≤ 15%, 15% < SIC < 80%, and SIC ≥ 80%) for the following 6 months in the form of discrete probability distributions at each grid cell. The latter two ice class probabilities are summed to obtain the sea ice probability, p = P(SIC > 15%).
