Other gauges gave no clear mean sea-level prediction gain beyond local history at four West Coast sites in 1990-2024

We tested whether prior-month levels at three other gauges improve monthly mean sea-level predictions beyond three months of local history. We used a fixed PSMSL Revised Local Reference snapshot for San Diego, San Francisco, Astoria and Seattle. Each test year used coefficients fit to earlier years from 1960 onward. There were 1,601 site-month predictions in 1990-2024. The equal-site mean squared error reduction was 1.79%. Its conditional 95% five-year block interval was -1.73% to 5.01%. Thus the primary test gives no clear mean gain. Seattle had a 4.80% point gain, but its interval included zero under quadratic trend and rolling training checks. When 1997, 1998, 2015 and 2016 were removed from scoring, mean gain was -0.01%. This check does not identify an ENSO cause. The result gives a small, reproducible spatial prediction benchmark. It does not measure flood risk or test a real-time data service. Four selected gauges, revised data, missing months and conditional uncertainty limit inference. The package includes exact source members and hashes, code, predictions, a separate numerical audit, and a figure.

sea level · West Coast · spatial prediction · local persistence · chronological evaluation · null result