An autumn Pacific temperature contrast raised Texas winter rain forecast error in 1981-2025 linear tests

We tested whether an August-October Nino 1+2 minus Nino 4 temperature anomaly improved November-March precipitation forecasts in all ten Texas climate divisions. Each linear fit used earlier winters only. We used fixed NOAA ERSSTv5 and nClimDiv snapshots, with 30 initial training winters and 45 test winters in 1981-2025. Adding the contrast to Nino 3.4 raised mean variance-scaled squared error by 2.06%. The conditional 95% five-year block interval for the increase was 0.50% to 4.04%. Nine division point estimates had more error. Strength adjustment, regional normalization, shorter training, and omission of three strong test winters retained negative skill. A December-February target left the error difference uncertain. Nino 3.4 alone reduced error relative to the earlier-year mean. This result concerns the extra contrast in the tested models. It does not show that ocean patterns lack useful rain information. Revised source data, a short record, fixed scales, and conditional intervals limit inference. The package gives source hashes, used monthly evidence, code, forecasts, a figure, and a separate numerical audit.

Texas · winter precipitation · ENSO · forecast validation · Pacific temperature contrast · negative result