A five-year seasonal correction raised monthly sea-level forecast error by 8.4% at 217 gauges in 2010-2024

We tested whether a recent seasonal correction improved next-calendar-year monthly mean sea-level forecasts. We used a fixed PSMSL RLR archive and 217 gauges. The baseline fit a trend and calendar-month effects to the previous 20 years. The alternative added calendar-month residuals from the previous five years after removal of each residual annual mean. Mean squared error changed by 8.37%, where positive means higher error. MSE reduction was -8.37%, with conditional 95% three-year block limits of -11.70% to -4.82%. Root mean squared error changed from 92.76 to 96.57 mm. The test used 15 years, with years resampled together at all gauges. We also tested a ten-year correction and a half-strength five-year correction. This extends earlier seasonal update and drift tests on overlapping data. It does not establish no seasonal change, a physical cause, or flood forecast performance. Final source revisions, future availability selection, uneven gauge locations and short conditional uncertainty limit inference. The package includes inputs, source hashes, forecasts, code, checks and a figure.

sea level · seasonal correction · chronological forecast · PSMSL · adaptation · forecast benchmark