Predicted SSH trend
AVISO observed SSH trend (same window)
About this demo
E6 is a graph-attention sea-level trend forecaster trained on the CESM2 large ensemble (90 members × 200 anchors). It maps the past 10 years of yearly summary statistics of five climate variables (SHF, SSH, SST, TAUX, TAUY) on a 1° ocean grid to the OLS slope of forced-response SSH at five horizons (5/10/15/20/30 years). At inference it is run zero-shot on AVISO/HadISST/ERA5 observations with per-timestep SSH demean to match the CESM2 Boussinesq convention. The model is published as part of the doctoral thesis on emulation-style climate forecasting; see the Sea-level trends chapter for the full architecture, training protocol, and skill results.
Output is the ±vmax-clipped trend pattern in mm/yr with the area-weighted global mean removed (the model predicts pattern, not GMSL). vmax is set to 1.5× the area-weighted standard deviation of the predicted field, so the colormap fills with the bulk of the signal and the tails saturate.