Temperature gave no clear added Atlantic one-year steric forecast gain at 2013-2024 origins

We tested whether temperature-only steric sea level improves one-year forecasts of total steric sea level beyond total-level history. We used NOAA upper-700-m basin means for 2005-2025 and 48 seasonal forecasts at 2013-2024 origins. Fits used earlier completed target years only. In the Atlantic with the Arctic, adding temperature increased mean squared error by 17.65%. The conditional 95% three-year block interval for error reduction was -23.08% to +2.64%. Thus, the test gave no clear added gain. Pacific error fell by 23.30% against the fitted history model, but both fitted models had more point error than simple persistence in all three basins. The Pacific gain also depended on the regression penalty. One revised product, short training and test records, model choice and conditional uncertainty limit inference. No useful forecast skill or physical cause is established. Code, exact source hashes, aggregate losses, all score draws, a separate numerical audit and a figure are included. Raw source files remain external.

steric sea level · temperature · salinity · chronological forecast · persistence · Atlantic · null result