Fig. 3: Comparing IceNet with dynamical and statistical prediction benchmarks.
From: Seasonal Arctic sea ice forecasting with probabilistic deep learning
a Mean binary accuracy versus lead time over the validation and test years (2012–2020) shown for IceNet, SEAS5, and the linear trend model. b IceNet’s binary accuracy averaged across the validation and test years, shown for each forecast calendar month and lead time, with the heatmap values shown within each grid cell. c, d Heatmaps of the difference between b and the equivalent heatmaps of SEAS5 and the linear trend model respectively, illustrating IceNet’s improvement over those models.
