Fig. 5: Quantifying the benefit of CMIP6 pre-training and ensembling in IceNet.
From: Seasonal Arctic sea ice forecasting with probabilistic deep learning
a IceNet’s ensemble-mean binary accuracy relative to that of another 25-member ensemble without CMIP6 pre-training (i.e., training only on observational data). b IceNet’s ensemble-mean binary accuracy relative to the mean binary accuracy of the 25 individual CMIP6 pre-trained ensemble members. c IceNet’s ensemble-mean binary accuracy relative to the mean binary accuracy of the 25 individual ensemble members without CMIP6 pre-training. Each value is averaged over the validation and test years, 2012–2020.
