Two ENSO slopes gave no clear winter sea-level error gain at Kwajalein in withheld 1951-2025 winters

We tested whether separate warm and cold Nino 3.4 slopes improve concurrent winter mean sea-level estimates at Kwajalein. We used fixed PSMSL and NOAA CPC sources for 72 complete December-February winters in 1951-2025. Five-year test folds had a one-year training gap on each side. The two-slope model reduced mean squared error by 3.82% relative to a single-slope model. The conditional 95% five-year block interval was -6.92% to 13.82%. RMSE fell from 37.48 to 36.76 mm. Full-record warm and cold slopes were -69.80 and -38.07 mm per degree C. These descriptive slopes suggest asymmetry but do not establish useful error reduction. Removing winters 1983, 1998, and 2016 from all fits and tests changed the gain to -1.32%. All planned error-gain intervals included zero. This is a concurrent model test at one gauge, not a forecast or a causal test. Source revisions, missing winters, few large events, trend choice, and conditional uncertainty limit the result. The package includes exact source extracts, hashes, code, all test estimates, block draws, a figure, and a numerical audit.

ENSO · Kwajalein · sea level · response asymmetry · blocked validation · null result