Pre-registration: gauge trend, land motion, splice to AR6 (written 13 Nov 2026 before any trend number was computed)

I have looked at the raw annual values (they sit between 6.80 and 7.12 m on the PSMSL datum) but have fitted nothing.

Quantity

Relative sea level at Constanța harbour: the sea surface measured against the gauge's own benchmark. It includes any sinking or rising of the quay. AR6 total is also relative sea level, so no land-motion correction is applied to either; the two are compared as they are.

Primary trend

OLS of annual RLR values on year, over the 59 years present, 1933–1996. Reported in mm/year with a 95% interval from the OLS standard error inflated for AR(1) residuals (effective sample size n_eff = n (1−r1)/(1+r1), r1 = lag-1 residual autocorrelation measured on the present consecutive pairs, t distribution with n_eff−2 degrees of freedom). The rise is called distinguishable from zero only if that interval excludes 0.

Sensitivities (all reported, none can replace the primary)

Pre-registered test (can fail, and is shown either way)

Prediction: the Constanța relative rise 1933–1996, after removing the AR6 background land motion at location 379, matches the global-mean rise. Global comparison, quoted from AR6 WG1 Chapter 2 (checked in the search result of 13 Nov 2026): 1.3 [0.6–2.1] mm/year for 1901–1971 and 1.9 [0.8–2.9] mm/year for 1971–2006; the comparison interval for 1933–1996 is therefore 0.6–2.9 mm/year (central 1.3–1.9). Land motion is the AR6 verticallandmotion median at location 379 read as the 2020–2100 slope from the ssp245 file (sign convention: see Correction 1 below). Pass = (gauge trend minus sinking rate) lies within 0.6–2.9 mm/year. Fail = outside; shown on the page either way. AR6's background land motion at tide-gauge sites is partly estimated from tide-gauge records, so the check is not independent of this record; the page says so.

Splice to AR6 (gauge stops 1996, AR6 baseline is the 1995–2014 mean)

The page's zero is "the 1995–2014 mean sea level", as in AR6. The gauge cannot measure it (no data after 1996). Rule: zero is the primary OLS line evaluated at 2004.5, the middle of the baseline period. The uncertainty of that anchor (95% prediction interval of the line at 2004.5, AR(1)-corrected) is drawn as a band and added in quadrature to every "cm higher than in your birth year" statement as the stated margin. Years 1997–2019 are drawn as a gap (dashed, no data), never as a measured line. The gauge year-to-year scatter (river flow, wind) stays visible as dots.

Reader's number

Black Sea caveat

AR6 projections are built from global models; the Black Sea is enclosed (Bosphorus) and its regional dynamics (Danube inflow, basin-scale response) are poorly represented. The caveat sits beside the fan on the page.

Terrain

Copernicus GLO-30 is a surface model with a stated vertical accuracy of a few metres; heights are relative to the EGM2008 geoid, not to the Black Sea mean level. The shore view shows land lower than a stated height above today's sea, not a flood map: storm surge, lagoon and river levels, sea walls, and DEM error are not modelled, and the page says so.

Corrections and deviations after the first results (13 Nov 2026)

Back