Ground motion, measured day by day
The ground under Romania's towns: 26 millimetres a year north-east, measured by GPS stations
Romania is a south-east European country of Carpathian mountains, the Danube plain and a Black Sea coast, with deep earthquakes under the Vrancea bend of the Carpathians. For a child born in 2000, the estimated displacement of the ground under its towns from 2000 to 2026 is about 71 centimetres north-east (global coordinate system IGS20), riding the Eurasian plate. How can a whole country travel that far while the distances between its towns barely change? Because the plate carries them all together. Relative to the plate, the is small — but not the same everywhere: south of the Carpathians, 12 of 13 stations point coherently south. Below: the map of measurements in both frames, then your town's number.
The whole map
Each thread is an antenna drawing its own journey
The map holds the real weekly positions of Romania's permanent GPS stations, from 1999 to 2026. Displacements are enlarged 120,000 times, the same enlargement on every picture on this page; the ruler below shows what 10 centimetres of travel looks like.
Every thread sets off in the same direction. Scroll on: first, one antenna alone, day by day.
One antenna, day by day
Bucharest: 9,851 measured days, 73 centimetres of travel
The BUCU station in Bucharest has recorded its position almost every day since 1999: 9,851 daily positions. On one day the point jumps by a few millimetres; over 27.6 years the points settle onto a nearly straight line, at 26.4 mm a year to the north-east.
A permanent station is an antenna fixed to a building or a pillar, listening to the satellites continuously. Every station velocity in the MIDAS product, which estimates long-term velocities from daily positions, is a line fitted like this one by the Nevada Geodetic Laboratory.
One country, one direction
26 millimetres a year north-east, everywhere
The arrows show ten years projected from each station's fitted velocity, at the same enlargement; the threads in the first picture span different durations from antenna to antenna, so their lengths are not comparable with the arrows. The median of the 93 stations in Romania with records long and precise enough for this comparison is 23.0 mm a year east and 12.9 mm a year north: together 26.4 mm a year, at an azimuth of about 61°.
The motion is nearly the same in the west, in Moldavia and in the Danube Delta, because the shared motion belongs to the whole Eurasian plate, carrying them all together.
Relative to what?
Subtract the plate and the arrows nearly vanish
Every velocity depends on the frame it is stated in. In the global IGS20 frame, fixed by stations worldwide, Romania rides with the plate. Take the plate's motion away (the NGL product with the Eurasian plate held still) and very little remains: the Romanian stations' median is 0.55 mm a year, and 90% of stations stay under 1.15 mm a year. What little remains is not random, though: south of the Carpathians the stations point coherently south (median southward component 0.56 mm a year, median magnitude 0.71 mm a year), while in Transylvania and Dobruja the medians are 0.38–0.29 mm a year with no common direction. The same southward sign, 0.5 to 2 mm a year at individual stations in the South Carpathians and Moesia (the region south of the Carpathians), is reported from a separate processing by the peer-reviewed study of Muntean et al. (Solid Earth, 2026); at the 73 stations the two solutions share, they typically differ by 0.13–0.15 mm a year per component.
The pictures can be viewed in either frame, at the same scale.
What remains, magnified
The same remainder, at fifty times the magnification
The comparison above keeps one scale, and should be read that way first: remove the plate and the arrows nearly vanish. This picture changes the magnification — fiftyfold — and shows only the remaining motion measured at each station, with no field interpolated between stations. Hollow circles mark velocities below about 0.25 mm a year, comparable to the typical differences between velocities estimated from neighbouring stations and measured velocities. No direction is shown for these stations; a hollow circle does not mean that the antenna measures no motion.
At this scale you can see what the national median hid: in Transylvania and Dobruja the arrows are small and point every which way, but south of the Carpathians they point almost all south — 12 of 13 stations, with a median component of 0.56 mm a year. The country rides north-east as one piece with the plate; at the same time, parts of it move slowly relative to each other. The second motion is real at regional scale, but for a single town it stays below this network's resolving power.
Vrancea, up close
Small remaining velocities at the stations around Vrancea
At the Carpathian bend, the great earthquakes are born deep down. At the surface, the 6 stations in the area with long enough records have a median remaining motion of 0.55 mm a year relative to the plate, below 1 mm a year, too small for this network to distinguish a town's motion confidently from noise; typical per-component prediction discrepancies at stations left out in turn are 0.241–0.251 mm a year. At all six stations, the measured motion relative to the plate remains below 1 mm a year.
The peer-reviewed 2026 study, on 99 stations with a separate processing, finds the same six stations still under one millimetre a year (largest: 0.70 mm a year there, 0.99 mm a year here). The verdict concerns the slow surface motion measured between earthquakes; it says nothing about seismic hazard, which belongs to the deep slab.
Rising and sinking
The vertical estimates do not support a national conclusion
The circles are antennas: red marks an estimated downward velocity, green upward, size growing with its magnitude; hollow circles have estimates consistent with zero. Stations with at least fourteen years of data have a median of -0.40 mm a year (17 stations); stations with series from 2012–2020 have -1.22 mm a year (35 stations). The two medians differ by 0.83 mm a year. The processing in the 2026 study explicitly accounts for abrupt changes in recorded position. For antennas with long records, it and the MIDAS product give medians near zero and similar results. At antennas where MIDAS does not separately model such changes, the 2026 processing gives a median vertical velocity 1.34 mm a year higher. The association is strong, but the cause at each antenna is not established here — at Bicaz, for instance, the two products have opposite signs. The same study finds local uplift of 0.5–2 mm a year at stations in the Carpathians and Dobruja; the stations are sparse and their records cover different periods; these data do not allow a reliable estimate of ground uplift or subsidence for Romania as a whole or for an individual town.
Delta antennas show the most downward values in one product only (SULN -6.3 mm a year), with no second-product station to confirm them. The grey cross at Cluj marks station CLUJ: its 11.5 mm a year northward velocity relative to the plate is not confirmed by neighbouring stations and is excluded from the town estimates. And one sinking street is invisible to stations tens of kilometres apart; that scale belongs to Europe's satellite-radar ground-motion service, EGMS.
Your town and your year
Your town and your year
Type a place (there are 7,149, from GeoNames) and a year. You get how many centimetres the ground there has travelled since that year, at the rate measured by the surrounding stations, together with the network's typical estimation discrepancies and the stations that support the answer. Clicking the map works too.
And in Britain: a north–south pattern in the antennas
The same products, with the same quality filter, give 151 stations in Great Britain. Their north-east ride with the plate is the same story; vertically, the station estimates form a north–south pattern: the median vertical velocity south of 52°N is -0.65 mm a year and north of 56°N it is 0.63 mm a year, in the same IGS20 frame. The pattern is consistent with the glacial isostatic adjustment the Environment Agency describes for Britain — land still rising in the north and sinking in the south since the last ice age — without these estimates from satellite positioning (GNSS) alone establishing the cause. At 20 common stations in the wider search region (including one in France and one in Northern Ireland) the second product's vertical velocities differ by a median of 0.26 mm a year, and leaving the 225 regional stations out one by one predicts the vertical with a median discrepancy of 0.31 mm a year. The drowned land between Britain and the continent has its own page: Doggerland.
Method, limits, sources
Where the numbers come from
Station velocities are the Nevada Geodetic Laboratory's MIDAS product, in the global IGS20 frame and in the Eurasian-plate-fixed variant: robust velocities from daily positions without separately detecting abrupt changes in recorded position (Blewitt et al. 2016, doi:10.1002/2015JB012552). The threads on the map are weekly medians of the same daily series, observed dates only, gaps broken. Stations kept: at least six years of data, at least a thousand good days, uncertainties at most 0.5 mm a year horizontal and 1.5 mm a year vertical; co-located duplicates reduced to the longest series. In Romania, 93 stations remain.
The second-product test
The EUREF Permanent Network's multi-year solution C2415 (EPN_IGS20_C2415, released 4 September 2026; the same GPS observations, processed separately — the independence is of computation, not of observations) shares 28 stations in the region. The median absolute station-by-station difference between the two products is 0.09 mm a year east, 0.11 north and 0.39 up (horizontal 90th percentiles: 0.32 and 0.32). The comparison covers the 28 common stations in the region, not all 93 Romanian stations.
Prediction discrepancies and the detection cutoff
Each of the 161 regional stations in turn was predicted from its neighbours with the same rule used for towns (nearby stations count more, with their contribution decreasing smoothly with distance on a 120 km scale; the estimate is also pulled toward the regional median, or toward zero for motion relative to the plate). The median discrepancy between prediction and the left-out station's value is 0.241 mm a year (east), 0.251 (north) and 0.632 (up); the horizontal 90th percentiles reach 0.90–0.76 mm a year. These are observed prediction discrepancies, not local confidence intervals. At town scale, this network cannot distinguish motion relative to the plate below about 1 mm a year confidently from noise. The estimate does not establish a town's exact motion, because the network's precision is not known separately at that point. To see how well the town is covered, check the stations used and the distance to the nearest one. When a whole region is left out in turn (seven regions), the median north-component discrepancies rise to 0.57 mm a year, with 90th percentiles of 1.25–1.58 mm a year: interpolation between neighbours is better than across regional gaps. The regional sign survives even that test: with the country's south left out entirely, the southward sign is predicted correctly at 12 of the 13 southern stations, though the predicted magnitude drops to about half. The 13 southern stations form two spatial groups, both with southward medians (−0.50 and −0.59 mm a year), so the sign does not hang on one station or one group. The southward drift south of the Carpathians also appears when estimates use distance scales of 60, 120 or 240 km, when the added weight of the reference value is 0, 0.15 or 0.5, and when excluded stations are included again. Across these variants, the median north–south components estimated for towns remain between −0.63 and −0.24 mm a year: the negative sign means southward motion.
Stations excluded from the estimates
The estimates exclude stations whose motion relative to the plate differs from the prediction from neighbouring stations by more than 2 mm a year on a component and more than six times the median error. In Romania, these are CLUJ, ORAD, RESI, CAM7. The big case is Cluj (CLUJ), with an 11.5 mm a year northward remainder no neighbour confirms; it is drawn with a grey cross. With no exclusions at all, the national medians change little: 26.4 mm a year of speed and a 0.57 mm a year remainder (97 stations). The stability of the national median does not establish the precision of estimates near excluded stations; for those towns, check which stations were used and how far away they are. The cross marks a horizontal outlier unconfirmed by neighbours; its cause (the antenna, or local motion) is not established here.
What this page cannot say
A velocity without its frame means nothing. An antenna moves with its building, and a short series or an antenna change can lie by millimetres a year, so an antenna's vertical velocity does not establish how much the town is rising or sinking. A town's number is an estimate from surrounding stations at the measured rate, run to 2026 and assumed constant before measurements began; years without stations are not observed years. A sinking street needs small-scale satellite radar (Europe's EGMS); these GPS stations are tens of kilometres apart.
Sources and licences
Nevada Geodetic Laboratory, MIDAS IGS20 velocities and daily station-position series (downloaded 12 October 2026). References: Blewitt, Hammond & Kreemer (2018) and Blewitt et al. (2016). EUREF Permanent Network, multi-year solution C2415 — CC BY 4.0 under the EPN Data Access and Use Policy of 2 July 2026. Places and counties: GeoNames, CC BY. Relief: NOAA ETOPO 2022, public domain, via the basemap of the Historical Atlas of Romania. Published comparison: Muntean et al. (2026), Solid Earth 17, 747–762, doi:10.5194/se-17-747-2026 (CC BY 4.0, with supplementary velocity tables; its 2024 conference version is abstract EGU24-17907). British land movement: Environment Agency, Flood risk assessments: climate change allowances (GOV.UK, updated 27 May 2022).