A spring level-change interaction gave no clear next-winter ENSO forecast gain in 1981-2024 tests

We tested whether the effect of spring ENSO change depends on spring state in a simple forecast model. We used NOAA CPC ERSSTv5 RONI and ONI and 44 target winters from 1981 to 2024. Fits used earlier issue years only. Adding the product of MAM level and MAM minus DJF change to a linear history model reduced RONI mean squared error by 8.77%. Conditional 95% five-year block limits were -4.56% to 21.08%. The ONI reduction was 7.89%, with limits of -13.34% to 27.80%. Last-30-year fits and full quadratic models also gave uncertain gains. This test adds a simple state-dependent term to a published linear benchmark. It gives no clear added forecast gain for the tested form. It does not show equal warm and cold dynamics. Revised data, future-based reference periods, 44 winters, and conditional intervals limit inference. Code, fixed inputs, every forecast, source hashes, all checks and a figure are included.

ENSO · RONI · ONI · nonlinear forecast · spring change · null result