A spring RONI change term cut next-winter forecast error by 30% in a 1981-2024 retrospective test

We tested whether a known spring change predictor also improves forecasts of the relative ENSO index, RONI. We used NOAA CPC ERSSTv5 seasonal tables and 44 test winters from 1981 to 2024. Each fit used earlier issue years only. Adding MAM minus DJF RONI to MAM RONI reduced next-DJF mean squared error by 30.43%. Conditional 95% five-year block limits were 13.50% to 42.07%. Adding ONI levels to that RONI model raised error by 3.92%; limits for error reduction were -8.65% to 0.59%. A last-30-year fit retained the RONI change gain but had a larger error increase from ONI. A JJA change term gave no clear gain. This extends prior campaign ONI tests to a relative index and a common-target test of added ONI information. It does not show a new physical cause or real-time skill. Revised data, reference periods that use future observations, 44 winters, and conditional intervals limit the result. Code, fixed inputs, every prediction, source hashes, bootstrap draws, and a figure are included.

ENSO · RONI · ONI · seasonal prediction · relative temperature · retrospective validation