- Constanța from the sea. Mamaia beach, Lake Siutghiol and the port, drawn from real elevation data (Copernicus). Heights are stretched three times, so that the low strip of beach stands out.
- A gauge, from 1933. Each white dot is one year's average sea level at the Constanța harbour gauge (PSMSL archive). The wall is only a drawing; the dots are measurements. Zero is the 1995–2014 average, placed with the trend line because the public series stops in 1996.
- The sea moves more than the trend. Single years sit about 6.8 cm above or below the line (one standard deviation). The archive notes that sea level on the Romanian coast is likely heavily influenced by river flow, and Sulina, one of the Danube's mouths, is about 135 km to the north.
- A rise you cannot prove. The line through the 59 dots rises 1.2 mm a year, about 7.5 cm over the record. Its 95% margin is ±1.2 mm a year, which only just includes zero. On its own, this gauge does not establish a rise at the 95% level.
- Then the public record stops. The public series ends in 1996. The harbour gauge kept recording: the Romanian marine institute's own documentation describes monthly data for 1998–2018 tied to the 1933 zero, available only on request, so this page cannot use it. The dashed line to 2020 is not a measurement: it follows the trend line to 2004, then joins the AR6 value for 2020.
- From here, a projection. From 2020 the bands come from the IPCC's sixth assessment report (AR6), computed for this very gauge. On the middle road: +22 cm above the 1995–2014 average in 2060 and +43 cm in 2100, likely between +16 and +75 cm.
- Three futures. Blue: emissions fall fast. Amber: the middle road. Red: they keep rising. Bands are the likely range (17th to 83rd percentile). Use the buttons to move the sea. AR6's models represent an enclosed sea like the Black Sea less well than an open coast.
- What 43 cm does to this shore. In the middle scenario the sea in 2100 is about 43 cm higher. Amber marks ground within 1.25 m of the sea's level that the sea can reach. At +43 cm, 0.08 km² (8 hectares) lies lower than the sea in this view; along this shore the waterline moves inland by about 1 m on average. The model's height errors are larger than these heights: it shows where the coast is low, not a flood forecast.
Can the gauge see the rise?
The Constanța series in the PSMSL archive (station 379) has 59 yearly values between 1933 and 1996; five years are missing (1935, 1944, 1980, 1986, 1993), 7.8% of the span. That meets the bar set before looking (at least 60 years, under 15% missing). It is also where the record ends: nothing after 1996 is in this series.
The method was written down before any trend was computed: a straight line through the yearly values, with the margin widened for the way neighbouring years resemble each other. The rise counts as shown only if the 95% margin excludes zero. It does not.
- All 59 years: rise per year
- +1.19 ± 1.24 mm
- Without the two years PSMSL flags (1985, 1992)
- +1.34 ± 1.29 mm
- 1933–1964 alone
- +2.2 ± 3.0 mm
- 1965–1996 alone
- −1.5 ± 3.6 mm
- Leaving out any ten consecutive years (54 ways)
- +0.5 … +1.6 mm
- Speeding up or slowing down (per year²)
- −0.09 ± 0.14 mm
- Land rising at the gauge, AR6 estimate
- 0.2 mm
- Sea-level rise once land motion is allowed for
- +1.4 mm
- Global rates 1901–1971 and 1971–2006 (AR6 Chapter 2)
- 0.6 … 2.9 mm
Pre-registered check: after allowing for land motion (AR6 puts the land at this gauge rising by about 0.2 mm a year, which hides a little of the sea's rise), the Constanța rise should fall within the range spanned by AR6's global rates. It does (1.4 mm inside 0.6–2.9), a weak test, because both intervals are wide. The AR6 land-motion estimate is partly built from tide gauges, so it is not independent of this record. The first and second halves disagree in sign, which is what a noisy record looks like.
How the test was fixed beforehand
Before any trend was computed, the method was written to a file: ordinary least squares on the yearly values, margin corrected for neighbouring-year similarity, the rise called shown only if the 95% margin excludes zero.
The chart's zero is the 1995–2014 average as AR6 defines it. Because the public series ends in 1996, zero is the trend line at 2004.5, the middle of that period.
One choice changed after seeing the numbers: the first draft used the prediction interval of a single year (±14.8 cm) for the zero's uncertainty. That is the wrong quantity for a 20-year average, so the page uses the interval of the line itself plus the noise of a 20-year mean (±6.7 cm in total). All three are in the file.
How much higher is the sea than the year you were born?
Type your birth year. Up to 1996 the answer uses the gauge's trend line; after that it uses AR6, because the public series stops in 1996. The margin includes the uncertainty of the gauge trend.
![]()
A picture with your numbers and the chart. If Save does nothing in your app, press and hold the picture.
What this does not show
- One gauge, 64 years, ending 30 years ago. The line from 1996 to 2020 is a straight guess between the record and AR6, drawn dashed. The zero of the chart is a straight-line stand-in for the 1995–2014 average; its own uncertainty is ±6.7 cm.
- Yearly values scatter around the trend line by about 6.8 cm (one standard deviation). PSMSL's note says Romanian sea levels are likely heavily influenced by river flow; I did not test that. The gauge measures the sea against its own quay, so any sinking or rising of the quay is inside the numbers.
- AR6's background land motion at this gauge is about 0.2 mm a year of rise of the land, which slightly lowers the relative sea level. It is a background estimate, partly built from tide gauges, so it is not independent of this record.
- AR6 gives a likely range of 17th to 83rd percentile. The dataset's median is the midpoint of the medians of two probability ranges that AR6 merged, and the lower edge of the middle-scenario band eases down a little between 2090 and 2100; both are kept as published. The Black Sea is semi-enclosed, joined to the ocean by the Bosphorus, and coarse global models resolve such seas poorly (Royston et al. 2022).
- The terrain is Copernicus GLO-30, a surface model that includes buildings and trees, with heights above a geoid and errors of a metre or more. The sea and Lake Siutghiol are flat in it (the lake sits at 1 m). The relief is stretched three times. Storm surges, waves, sea walls, river and lake levels are not modelled. The amber glow marks ground within 1.25 m of the sea's level that the sea can reach; it is a map of low ground, not a flood forecast. Of the ground lower than +43 cm, 74% lies north of the port, along Mamaia and beyond, and the rest in the port.
- Today's gauges in the harbour report live, but no published link joins their zero to the 1933 datum, so they are not joined to this record.
Sources
- Sea level: PSMSL, Permanent Service for Mean Sea Level, station 379 CONSTANTZA, RLR annual and metric monthly data (psmsl.org), downloaded 13 Nov 2026. Holgate et al. 2013, Journal of Coastal Research 29, 493–504.
- Projections: IPCC AR6 sea-level projections, medium confidence, Garner et al. 2021 (doi:10.5281/zenodo.5914709), location 379; Fox-Kemper et al. 2021, AR6 WG1 Chapter 9. Global rates: AR6 WG1 Chapter 2.
- Projections method: Kopp et al. 2023, Geoscientific Model Development 16, 7461–7489 (doi:10.5194/gmd-16-7461-2023). We thank the projection authors for developing and making the sea-level rise projections available, multiple funding agencies for supporting the development of the projections, and the NASA Sea Level Change Team for developing and hosting the IPCC AR6 Sea Level Projection Tool. Licence CC BY 4.0.
- Data after 1996: NIMRD (Romanian marine research institute), EO4SIBS Dataset User Manual, December 2020, p. 29: Constanța sea level, monthly 1998–2018, 'restricted (download upon request)'. Reported to me from that document; its server was unreachable when I tried to open it myself.
- Black Sea in global models: Royston et al. 2022, Ocean Science 18, 1093 (doi:10.5194/os-18-1093-2022).
- Terrain: Copernicus DEM GLO-30, © DLR e.V. 2010–2014 and © Airbus Defence and Space GmbH 2014–2018, provided under COPERNICUS by the European Union and ESA.
- Method: the gate check and the pre-registration (English), written before the trend was computed.