A sign term did not reduce mean monthly sea-level prediction error at six US gauges in 1990-2024

We tested whether different responses to positive and negative monthly sea-level anomalies improve predictions at six selected US gauges. We used the PSMSL RLR snapshot of 19 October 2026. Each test year in 1990-2024 used only earlier years from 1960 to fit the models. The baseline used seasonal persistence and three recent local levels. One added absolute-anomaly term raised the equal-site mean squared error by 0.56% across 2,454 predictions. The conditional 95% five-year block interval for error reduction was -1.07% to -0.02%. One-year and three-year block intervals included zero, as did the linear-trend check. All four model designs had negative mean point gains. Thus the test gives no evidence of useful added prediction accuracy from this sign term. This extends a prior descriptive campaign test with chronological prediction and seasonal slopes. Six selected gauges, revised observations, missing months, a simple model, and conditional uncertainty limit the result. This is not a test of flood duration or a physical cause. The package includes code, exact source hashes, annual error summaries, an independent score audit, and a figure. External source records remain outside the package.

sea level · seasonal persistence · chronological prediction · nonlinear response · null result