A six-month cycle cut seasonal sea-level test error by 9.49% at 284 retained gauges in 2010-2024

We tested whether a six-month cycle learned from earlier sea-level records reduces later seasonal-shape error beyond a one-year cycle. We used a fixed PSMSL RLR monthly snapshot. Completeness rules retained 284 gauges, with training in 1980-2009 and testing in 2010-2024. Each year had its observed annual mean removed, so this is a seasonal-shape test, not an absolute forecast. Adding the six-month cycle reduced mean squared error by 9.49%. The conditional 95% common-calendar three-year block interval was 5.72% to 12.69%. Error fell at 253 of 284 stations. Equal weighting of 62 occupied geographic cells gave a 9.71% reduction. Twelve monthly means gave only a further 0.27% reduction beyond the two cycles, with limits of -1.32% to 2.00%. An independent Fourier audit checked all station errors and showed why positive phase alignment alone need not improve error. One revised product, selected complete records, uneven geography, short testing, and conditional uncertainty limit the result. No physical cause or flood effect is shown. Code, input rows, source hashes, all errors, bootstrap draws, station notes and a figure are included.

sea level · semiannual cycle · seasonal shape · PSMSL · chronological evaluation · harmonic analysis