Linear autumn ENSO forecasts gave no clear gain for a California winter precipitation contrast in 1981-2026

We tested whether September-November Nino 3.4 improved forecasts of the next January-March south-minus-north precipitation index in California. We used two fixed NOAA sources and division scales from 1951-1980. North combined divisions 1 and 2. South combined divisions 6 and 7. Each forecast used earlier target years only. In 46 evaluation years, 1981-2026, linear-model mean squared error skill was -2.57% relative to the earlier-year mean. Its conditional 95% five-year block interval was -12.71% to 5.70%. Skill for the common component was 7.88%, with an interval of -20.76% to 26.30%. Their skill difference was uncertain. Both regional fitted linear associations were positive, so subtraction removed much of the shared association. A quadratic model gave contrast skill of 7.98%, with an interval of -0.10% to 15.32%. Thus, the result depends on model form and does not exclude useful ENSO information. Fixed regions, revised data, a short record, and conditional intervals limit inference. The package preserves source rows, provenance, code, forecasts, resamples, figures, a scalar audit, and offline repeat evidence.

ENSO · California · precipitation · regional contrast · chronological forecast · null result · Nino 3.4