ENSO Forecasts
Every ensemble member from seven seasonal prediction systems, as one interactive outlook — percentile fans, individual models, ONI against RONI, and the whole thing measured against every ENSO event since 1970. Each model is anomalized against its own 1993–2016 hindcast, so drift and bias are removed before anything is drawn.
The plume — all models, every member
The shaded fan is the percentile spread across all members of all models (10–90% and 25–75%); coloured lines are each model's ensemble mean; the bold white line is the multi-model mean (each model weighted equally, so the big NCEP and BoM ensembles don't dominate). Toggle RONI to subtract the tropical background — the honest measure of how strong this event would feel — and members to see every individual trajectory. The observed index leads in from the left. Dotted lines mark ±0.5 / 1.0 / 1.5 / 2.0 °C. See also: NOAA's new SFS model, fully visualized — an independent 31-member outlook to weigh against this multi-model plume.
How the forecast has evolved
One line per monthly C3S issue — the multi-model mean from that month's outlook, colour-ramped from indigo (oldest) through cyan to gold (newest), each with its member P10–P90 envelope in the matching tint. Watching the lines stack up shows how the multi-model system converged on the event — and the envelopes show whether the ensemble narrowed as it shifted, or just slid its whole distribution. Click an issue in the legend to toggle its line and envelope together. The white line is what actually happened. Issues are archived as they publish, so this chart deepens every month.
Against every ENSO year since 1970
Each grey line is one calendar year of the observed index; the shaded band is the record envelope (the highest and lowest value ever observed in each calendar month since 1970). Each named El Niño gets its own colour, with heavier lines for stronger events — strong and very-strong events start visible, the rest wait in the legend. The current year is drawn bold white, and the gold dashes extend it with this issue's multi-model mean (monthly values, anchored at the last observed month). Every year here is computed against the same fixed 1991–2020 base — CPC's historical tables use sliding 30-year bases that shift by era, which would make decades incomparable; the RONI series is likewise rebuilt from ERSSTv5 with one base and one σ-scale throughout, so 1972 and 2026 sit on the same ruler.