Previous-month NAO gave no clear added winter sea-level forecast gain at 60 European gauges in 2005-2024

We tested whether previous-month North Atlantic Oscillation (NAO) improves winter monthly mean sea-level forecasts after recent local sea level is known. We used fixed PSMSL and NOAA snapshots at 60 gauges in 45-65 N, 15 W-30 E. Models used earlier years only and were tested in 2005-2024. Added lag NAO reduced mean squared error by 4.58%. The conditional 95% three-year block interval was -4.20% to +11.61%, so the added gain was uncertain. Root mean squared error fell from 146.03 to 142.65 mm. Cell weights and a 20-year training window also gave uncertain gains. A diagnostic model with the target month's NAO reduced error by 24.67% (10.54% to 35.40%). That model cannot be a forecast because it uses future atmospheric data. The evidence thus leaves the added lag forecast value uncertain while retaining a concurrent link under the tested models. This is a retrospective benchmark, not a test of physical cause or flood extremes. The short test, selected long records, revised data, incomplete daily sampling, and conditional intervals limit inference. The package gives source rows, hashes, station notes, code, forecasts, block draws, a figure, and a separate numerical audit.

NAO · sea level · PSMSL · monthly forecast · local persistence · conditional uncertainty · European coasts · null result