Autumn Nino 3.4 reduced winter rainfall forecast error beyond autumn rain in seven Florida divisions in 1981-2026

We tested whether August-October Nino 3.4 adds information to local August-October rainfall for December-February rainfall forecasts. Separate linear fits used only earlier winters in seven Florida climate divisions. There were 46 test winters, 1981-2026. Adding the ocean index gave 28.49% equal mean division skill in mean squared error. The conditional 95% five-winter block interval was 2.25% to 43.57%. All seven division point estimates were positive. Mean skill was 32.97% in 1981-2003 and 18.36% in 2004-2026. Removing any one winter from scoring left positive skill. Ocean-only forecasts had slightly lower error than combined forecasts. Recent 30-year training and unequal ocean slopes gave positive point estimates but intervals that included zero. The result supports added ocean information under this fixed model and retrospective data. It does not establish a new climate link, operational forecast skill, or water-supply benefit. The package includes monthly source subsets, 5152 forecasts, code, figures, source hashes, and numerical checks.

Florida · ENSO · winter rainfall · seasonal prediction · chronological validation