What this gives you: an answer for one neighbourhood — how fast its ground is moving, whether its low ground is already under the reference line, and how often it has actually flooded.
What it does not give you: a flood forecast. Sinking and flooding turn out to be only loosely related here, and the check that tested it failed. Switch the map to its third view and you can see that for yourself.
What the whole city looks like, counted
Of 261 neighbourhoods on the mainland — the 6 islands of Kepulauan Seribu sit outside the radar's view and are left off the map rather than guessed at — this is how they divide:
| Ground here is… | Neighbourhoods | People |
|---|---|---|
| already below the line | 14 | 803,014 |
| sinking fast | 3 | 247,910 |
| sinking | 14 | 1,090,013 |
| sinking slowly | 149 | 5,580,026 |
| stable | 19 | 793,867 |
| rising | 62 | 2,204,672 |
So 31 neighbourhoods are either already under the line or losing a centimetre or more of ground a year, and 81 are stable or rising outright. The city's middle neighbourhood is falling 0.21 centimetres a year — about the thickness of two coins. 803,014 people live in the fourteen where the low ground has already gone under, and that is a measurement, not a projection.
"Already below" needs saying precisely. It means the lowest tenth of that neighbourhood's ground sits at or under the reference line — not that the neighbourhood is underwater, and not that anyone's house is. Land below the line stays dry as long as something keeps the water out. What it loses is the margin.
But isn't Jakarta sinking much faster than that?
It was. The figures everyone knows come from the 1990s and 2000s: levelling surveys found up to 25 cm a year in 1991-1997, GPS campaigns up to 28 over 1997-2011, radar up to 21.8 in 2007-2009. By 2015-2019, ground surveys were reporting a maximum of 6.2. The fastest thing in this record, over 2017–2023, is Pluit at about 3.99 cm a year. A city that has closed wells and piped in water does not keep sinking at its historical rate, and quoting the old number as the current one is the most common mistake made about this place.
The harder problem is that studies of the same years still disagree, and all of them are right. Published Sentinel-1-era figures for Jakarta run from 3.6 to 15 cm a year:
- 15 cm/yr — the single fastest point in Greater Jakarta, 2017-2023 (Sidiq et al. 2025)
- 7.85 cm/yr — the fastest point in one bowl at Muara Angke, 2014-2025 (Lapadat et al. 2025)
- 5.64 cm/yr — the fastest pixel in North Jakarta, 2017-2023 (Ohenhen et al. 2026)
- 5.5 cm/yr — the average across north and north-west, 2017-2022 (Widodo et al. 2025)
- 3.6 cm/yr — the maximum inside a 1 km coastal strip, 2017-2023 (Ohenhen et al. 2026)
None of those is wrong. One reports the single fastest point in the whole metropolitan area; another the fastest pixel in one district; another an average across a region. This page reports the fastest tenth of a neighbourhood, because a neighbourhood is the thing a reader can find themselves in. Ask "how fast is Jakarta sinking" and the honest reply is another question — fastest where, and averaged over what — and the four-fold spread above is what that question is worth.
Why there is no year on this page
The obvious thing to build here is a countdown: pick a neighbourhood, get the year it goes under. This case can compute that, and the reason it is not the headline is that the honest version is almost empty.
At the rate the radar measured, of the 261 neighbourhoods with a reading, 14 are already below, 1 crosses the line within the next 25 years, 2 cross it later this century, and 163 would not cross it for over a century. The remaining 81 have no countdown at all, because their ground is steady or rising. A countdown clock would show a blank for almost everyone who typed their name into it.
The number that moves elevenfold before anyone forecasts anything
Here is the part a government or an NGO should read before quoting any figure from work like this — including this work.
"How many people live on land below one metre" sounds like a count. It is a count plus a decision, and the decision does more work than the data. A neighbourhood is not one height; it is thousands of small patches at different heights. If you take each patch's lowest point you get 391,029 people below the one-metre line today. If you take each patch's average you get 35,965.
| Counting each patch by its… | Below 1 m, 2025 | Below 1 m, 2050 |
|---|---|---|
| lowest point — the cautious reading | 391,029 | 468,034 |
| lower quarter | 144,346 | 188,719 |
| average — the flattering reading | 35,965 | 51,703 |
Same radar, same year, same city: the count differs by a factor of 10.87 on one methodological choice, with no forecasting involved. This page and its map use the cautious reading throughout and say so. Any headline number of this kind that does not tell you which choice it made is not a number you can regulate on.
The projection to 2050 then sits on top of that. It assumes the ground keeps falling at the rate it fell during 2017–2023, in a straight line. Ground does not do that. Assume instead that subsidence halves every 4.85 years — what happens when groundwater extraction is curbed — and the 2050 figure below one metre is 422,303 rather than 468,034. Neither is a prediction. Both are arithmetic on an assumption someone picked.
And it is not a flood map
Before this work was built, one of the checks written down in advance was whether the sinking ranking matches where Jakarta has actually flooded. It does not. Compared against recorded flood events across the same 261 neighbourhoods, the two rankings line up at 0.16, where 1.0 would be perfect agreement and 0 would be none. The check required 0.50. It failed, and it is published as failed.
Counted in neighbourhoods rather than in correlations, the mismatch is starker. Of the 14 whose low ground is already under the line, only 6 have any recorded flood at all. And of the 14 neighbourhoods that have flooded most often, not one is among them.
You can see it rather than take it on trust: put the map on how fast the ground is falling, then on where it has actually flooded. The bright areas move to different places. Show me that on the map.
That result is not a defect in the measurement — the ground movement itself agrees with the published literature to 0.03 centimetres a year, and with independent ground stations. It is a fact about flooding: what floods in Jakarta is decided by drainage, rivers, rainfall and sea walls, and subsidence is one slow input among them. Sinking ground raises the long-run risk. It does not tell you which street floods next season.
What a government or an NGO can actually do with this
Three things follow, and the third is uncomfortable.
- Target, do not blanket. The problem is concentrated: 17 neighbourhoods carry 1,050,924 people, and 62 neighbourhoods are rising. Money spent evenly across the city is mostly spent where the ground is fine.
- The lever is groundwater, and the effect size is knowable. Curbing extraction is what turns the straight line into a curve, and the difference between those two arithmetics is 45,731 people by 2050 — a benefit worth costing against a regulation.
- There is no evidence yet that it is working city-wide. Ground stations were split before and after 2017 and compared: of 20 stations, only 3 slowed down. West Jakarta's 3 stations all slowed; the 17 in the east got faster on average. Whatever is being done is either not reaching the east or has not reached it yet. That is the finding a regulator most needs and least wants.
What you must not do with this map. Do not use it to value a house, to tell a family their home will be underwater, or to argue a neighbourhood should be abandoned. It reports how fast ground moved over a 2017–2023 window and how low that ground is — not what will happen to a building, a street, or a person. It is also not a flood map, as its own failed check shows, and using it as one would put attention on the wrong streets.
Could this run in daily operations?
DEPLOYABLE WITH LIMITS — the map of which neighbourhoods sit below the line — not flood prediction
Every pipeline here is a batch backfill: it fetches a season or an archive, not a live feed. Anything operational means rebuilding the ingest for near-real-time arrival, whatever the verdict below says.
What stops it. The registered check that failed is the one that would matter most for planning: ranking neighbourhoods by how fast they are sinking does not predict where floods have actually been recorded. One further check is still unresolved.
What it would need. Nothing for the descriptive map, which is a measurement rather than a forecast. Ranking flood risk with it needs the failed check to pass first.
If you want the detail
The methods, the 5 checks set in advance and how each came out, the comparison against ground stations and against the published literature, and everything still uncertain are on the technical article, and the technical dashboard carries the full velocity field and the statistics behind it. Every figure here is read from the same record as those, so they cannot drift apart.
Words on this page, in plain language 3 terms
- Sentinel-1
- A European radar satellite. Because radar makes its own signal it sees through cloud and at night, which matters in a country that is cloudy for half the year.
- radar
- Instead of photographing reflected sunlight, the satellite sends its own microwave pulse and listens for the echo. It works through cloud and at night; it does not see colour.
- fold
- One slice of the data in a repeated test. How the slices are drawn can matter more to the score than the model does.
Ground movement from Ohenhen et al. 2026 (Zenodo 10.5281/zenodo.15786356), CC BY 4.0. Data vintage 2026-08-30. Outlines simplified to 25 m for the web from the same geometry the analysis used. Written to be read without a background in radar or statistics; nothing was rounded to make a point.
Found something wrong on this page? Report a correction — it opens a pre-filled issue — or email taufik.adi@openstudy.id. Corrections are credited by name in the errata, and one that changes a finding says so on the page.