Is Jakarta still sinking?

1 The claim under test

"Jakarta is sinking" is one of the most repeated facts about any city on earth. It usually arrives with a rate — ten, fifteen, twenty-five centimetres a year — and a date, 2050, by which some large fraction of the city is said to be underwater. The case under review measures the ground motion honestly, from orbit, and then does what everyone does with it: turns it into a map of who is exposed and a countdown.

Three questions follow, and they have to be asked separately. Is the measurement right? Is the exposure product built on it telling you anything the elevation model did not already say? And does the clock support the reading it invites? We find the answers are yes, no and no — an awkward order, because it means the weakest parts of the page rest on its strongest part.

2 What is already known, and over which years

Jakarta's subsidence has been measured for four decades and the record is unusually specific. Abidin et al. (2015) tabulate it by method and period: levelling gives 1–9 cm/yr for 1982–1991 and 1–25 cm/yr for 1991–1997; GPS campaigns 1–28 cm/yr over 1997–2011; InSAR 1–12 cm/yr over 2006–2010. Chaussard et al. (2013), using ALOS PALSAR over 2007–2009, report a Jakarta mean vertical rate of 7.2 cm/yr and a maximum of 21.8, with the coast running 9.5–21.5 cm/yr. Abdullah et al. (2021), on campaign GPS for 2015–2019, report a maximum of 6.2 cm/yr at Muara Baru and a southern average of 1.16.

Read down the figure and the story is not "the case disagrees with the literature". It is that the rate has a date attached, and the dates are falling. The case's page explains its lower numbers as a matter of scale — a ~75 m cell average cannot reproduce a persistent-scatterer point peak. That is true and it is part of the answer. Sections 6 and 7 show it is not the larger part.

3 One city, one period, four different answers

Before trusting any single field it is worth asking what the spread between fields is. Four Sentinel-1-era studies of Jakarta over overlapping years are available, and they do not agree.

Sidiq et al. (2025) report up to 15 cm/yr for Greater Jakarta on Sentinel-1 SBAS over 2017–2023 — the identical years as the field under review. Lapadat et al. (2025), with a full three-dimensional decomposition in a Sunda-plate-fixed frame, find a maximum of 7.85 cm/yr in the Muara Angke bowl over 2014–2025. Widodo et al. (2025), on L-band ALOS-2 validated against 255 GPS points, give an average of 5–6 cm/yr for northern and north-western Jakarta. Ohenhen et al. (2026) — the source of this case's field — put the fastest pixel in any Jakarta municipality at 5.64 cm/yr and the maximum within a 1 km coastal strip at 3.6.

Four independent Sentinel-1-era analyses of one city over overlapping years span a factor of four. This case reports its rates to two decimal places and treats the deposit as truth.

That spread is the honest error bar on any single-source subsidence product, and it dwarfs the formal uncertainties any of the studies publish. It does not invalidate the case; it sets the precision at which any of its numbers should be read.

4 Data and method

The field. Ohenhen et al.'s Java-wide vertical land motion product, Sentinel-1, 2017–2023: 253,568 points inside the Jakarta window, 216,294 unique coordinates after averaging overlapping frames, gridded to 0.001°. Two things about it matter for this review. First, vertical motion was decomposed from ascending and descending line-of-sight geometry by the depositors, so the case inherits a two-dimensional decomposition rather than making a line-of-sight assumption of its own. Second, the depositors tied the result to the IGS14 global frame by an affine transformation using ten GNSS stations — including stations from the Susilo et al. archive that this case then uses to validate it.

Coverage. Every mainland kelurahan has at least 55% of its cells measured, with a median of 86%. The published field images fill isolated gaps for display only; all statistics use the raw grid, and our recomputation of every headline number from the live published table reproduces the page to within rounding.

The exposure layer. WorldPop 2020 constrained population (10,724,798 against a census of 10,562,088), GHSL built-up surface, and Copernicus GLO-30 elevation, all on one 100 m grid, aggregated to 261 mainland kelurahan. Ground height per cell is the lowest of that cell's roughly nine 30 m pixels.

The clock. Ground height in year y is the 2025 height plus the measured 2017–2023 rate times the elapsed years. Cells without radar coverage are held still.

The tests. Eight checks were specified before they were run, all on data already on the case's disk: (A) refit every GNSS station on an early and a late sub-period; (B) compare the radar to GNSS refitted on the radar's own years; (C) recompute the exposure counts on the cell minimum, the cell 25th percentile and the cell mean; (D) apply the elevation-epoch correction; (E) sweep the sea-level datum; (F) re-rank with the radar switched off; (G) re-run the clock with a decaying rate; (H) reproduce the depositors' own published per-municipality table. Every result below is published, including the two that support the case.

5 Finding one — the regrid is faithful, and this is the check the case is missing

Before criticising what the case built on the field, we should establish that it read the field correctly. Ohenhen et al. publish per-municipality statistics for the same data in their supplementary table S2. That is a direct, like-for-like target: if the case's ingest, quality filter, gridding and zonal aggregation are faithful, our numbers should land on theirs.

Every municipality lands on the line. The largest disagreement in the mean is 0.032 cm/yr, and the ordering, the medians, the standard deviations and the share of each municipality subsiding faster than 1 cm/yr all reproduce. An independent pipeline — different resampling, different boundary file, different zonal method — recovers the depositors' published table.

That is the strongest validation available to this case, and it is not on the case page. It also matters for everything that follows: when we say below that the case's product is wrong, we are not saying its measurement is.

6 Finding two — the field is tied to the ground better than the case claims

The case validates its field against three continuous GNSS stations and reports a pass inside its ±5 mm/yr tolerance: at CJKT the radar reads -4.6 mm/yr against a published GNSS rate of -6.4. Its methodology footer used to explain the gap this way: "the InSAR field is referenced to stable ground in the same region, so a small frame offset (≈1–2 mm/yr) is expected."

That explanation is wrong on two counts. The field is not referenced to local stable ground: the depositors transformed it into IGS14, the same family of frame the GNSS solutions are in (IGb14), so no frame offset should be expected at all. And the published GNSS rates are fitted over each station's whole record — at CJKT, 2010 to 2022 — while the radar covers 2017–2023. The two numbers describe different decades.

We refitted each station's daily vertical series on the radar's own window, with annual and semi-annual terms to absorb the hydrological load cycle. Our whole-record refits reproduce the published rates almost exactly (CJKT −6.1 vs -6.4; CTGR −2.9 vs -2.9; CBTU −0.5 vs -0.5 mm/yr), which is the check that the refitting is sound.

Compared like for like, InSAR minus GNSS is −0.03 ± 0.89 mm/yr across the three stations, with a largest single residual of 0.98. Compared as the case compares them, it is +1.78 ± 0.51.

Two things follow. The deposited field is tied to an absolute datum an order of magnitude better than the case gives it credit for. And the +1.78 mm/yr the case writes off as a frame offset is not noise: it is the same sign and roughly the same size at all three stations, and it is measuring something real.

This check is not fully independent, and the case does not say so. Ohenhen et al. aligned their field to IGS14 using ten GNSS stations drawn partly from the Susilo et al. archive — the same archive the case's GNSS gate tests against. A check between a product and the data used to align that product can only fail if something is badly broken. The case presents it as external corroboration. The honest description is a consistency check, and the genuinely external validation is Section 5's replication and the multi-study comparison in Section 3.

7 Finding three — Jakarta's subsidence has roughly halved, and its neighbours' has not

If the radar's window disagrees with the GNSS record's window by a consistent amount, the ground changed speed between them. The Susilo et al. archive on the case's disk holds daily vertical solutions for 20 continuous stations along north Java from 2010 to 2022 — a direct test the case never runs.

At CJKT the rate falls from −8.6 to −3.8 mm/yr — a 56% reduction. The other two stations inside the case's working area do the same: CTGR −3.9 → −1.1, CBTU −1.0 → +0.9 mm/yr. All 3 slowed, by an average of +3.18 mm/yr.

The contrast is the finding. Across the rest of the north Java coast, 12 of 17 stations got faster, by an average of −1.62 mm/yr. At Pekalongan the ground is dropping at nearly 12 cm a year and accelerating — roughly three times the rate of the fastest decile of the fastest kelurahan in Jakarta.

Jakarta is no longer the fastest-sinking city on Java, and on this evidence it has not been for some years. The famous numbers are real, and they are historical.

This is not our discovery. It is the central claim of the paper the case's own field comes from: Ohenhen et al. report that 45% of subsiding Java shows statistically significant deceleration against 17% acceleration, and specifically that Jakarta has shifted "from rapid subsidence during 2007–2010 to near stability or even uplift in 2017–2023 in 50% of the land areas," attributing it to groundwater extraction bans, industrial relocation and piped-water expansion. The case uses the dataset and not the finding. Our contribution is to confirm it from an entirely different instrument — continuous GNSS rather than InSAR — and to show that the residual the case's own validation table writes off is exactly this signal.

Robustness, and what this does not show. The Susilo archive documents equipment changes that are not modelled as offsets. Refitting the late window from 2018, so that any step in late 2017 falls outside it, gives CJKT −5.2 rather than −3.8 mm/yr — a 40% reduction instead of 56%, and a mean InSAR-minus-GNSS residual of +0.40 rather than −0.03. The direction and rough magnitude survive; the second decimal does not. Beyond that: the GNSS record ends in 2022, and three stations is a thin sample for a city of 660 km². Deceleration at three points is not deceleration everywhere, and a rate that has halved can un-halve if extraction resumes.

8 Finding four — the fast ground has moved inland, away from the exposure

The case's hero used to promise "millimetres a year in the centre, centimetres on the north coast." Its own ranking says otherwise. Of the 20 fastest-subsiding kelurahan by median rate, 12 are in West Jakarta, 4 in South Jakarta and only 4 in North Jakarta; only 6 of the 20 sit in the coastal belt at all. Ohenhen et al.'s own table agrees: 45.6% of West Jakarta subsides faster than 1 cm/yr against 28.1% of North Jakarta.

Rank correlation between sinking rate and low-ground share across all 261 measured kelurahan is ρ = 0.149. They are very nearly independent quantities, and the case's exposure index averages their ranks as though they were two readings of one thing.

The two fastest-sinking kelurahan make the point unaided. Meruya Selatan sinks at −2.14 cm/yr at the median — faster than any kelurahan in North Jakarta — with low ground at 9.3 m. Meruya Utara, −2.04 cm/yr, low ground 6.7 m. Between them they hold 159,793 people. Both contribute exactly zero to the 2050 exposure count.

That is not a defect in the measurement; it is the measurement telling you something. Fast ground on a high terrace is a foundation, buried-pipe and drainage-gradient problem, not a flooding problem, and it is where the remaining extraction is. Section 13 shows a published policy analysis reaching the same conclusion from different data.

9 Finding five — the headline exposure number is a choice of estimator

The case's most quoted statistic is that 391,029 people live on ground below +1 m. Ground height per 100 m cell is the lowest of that cell's roughly nine GLO-30 pixels — a defensible attempt to find roads and yards rather than roofs in a surface model. It is also a minimum of nine noisy samples, and it behaves like one.

On the cell minimum, 391,029 people are below +1 m. On the cell's 25th percentile, 144,346. On the cell mean, 35,965 — a factor of 10.87 between the first and the last. The median gap between the two estimators is 2.04 m of ground. Read on the cell mean, the published headline is the population below +2.93 m, not below +1 m.

Below mean sea level the sensitivity is worse — a factor of 23.11 — and there the estimator meets a second artefact. GLO-30 pins water and near-shore cells to exactly 0.00 m. 61,713 people live in 1,981 cells sitting at exactly that value, poised one arbitrary millimetre above the threshold.

The case's own numbers contradict its own headline. If 121,394 people are below mean sea level in 2025 and 165,280 in 2026, the 2025 figure is not a measurement of Jakarta. It is a measurement of where a digital elevation model stops rounding.

The context makes this avoidable rather than unlucky. The depositors of the very field this case uses paired it with DeltaDTM, a bare-earth model corrected with ICESat-2 and GEDI and published with a mean absolute error of 0.43 m. The case took a raw surface model and recovered a ground proxy with a minimum operator. Neither choice is unreasonable on its own; only one of them has a stated error budget.

None of this makes the exposure map worthless. The ranking of kelurahan is far more robust than the absolute counts, because the estimator bias is broadly common across the city. But a count published to the nearest person, when the estimator moves it by a factor of eleven, cannot bear the weight put on it.

10 Finding six — the clock is, to three decimal places, an elevation map

The case's central product ranks kelurahan by residents on ground below +1 m in a chosen year, animated from 2025 to 2050. The obvious question is whether the radar is doing any work in it. We switched the radar off — velocity set to zero in every cell — and re-ran the same ranking.

The 2050 ranking correlates ρ = 0.959 with the identical ranking computed with no subsidence at all, and shares 19 of its top 20 kelurahan with it. It correlates ρ = 0.963 with low-ground share alone and ρ = 0.98 with the trivial product of population and low-ground share. Against the measured sinking rate — the case's entire instrument — it correlates ρ = 0.219.

Twenty-five years of subsidence moves the city total from 391,029 to 468,034 people, a change of 19.7%, against a base that Section 9 showed is itself uncertain by a factor of eleven. The signal the radar contributes is an order of magnitude smaller than the uncertainty in the layer it is added to.

The clock's own internals say the same. Of 261 measured kelurahan, 14 already have low ground below mean sea level, 81 are not subsiding fast enough for the clock to run at all, 163 need more than a century to reach it — and exactly 1 crosses mean sea level inside the 2025–2050 window the page animates.

What this does not mean. It does not mean the radar is useless; it means the radar is being asked the wrong question. A quantity that is 96% elevation should be presented as an elevation product with a subsidence adjustment, and the subsidence measurement should be published where it is the whole signal — Section 8's inland hotspots, Section 7's deceleration — rather than buried inside a composite where it contributes four per cent.

11 Finding seven — three assumptions, none of them priced

The clock states its assumptions plainly, which is to its credit. What it does not state is how much each one is worth. We priced three of them on the same grid.

The elevation is twelve years old. GLO-30 was acquired 2011–2015 and is used as the 2025 surface. Applying the correction the case declines — dropping every cell by its own measured rate over 12 years — moves today's count from 391,029 to 432,584. That is a number the published clock does not reach until 2037: the uncorrected epoch is worth 12 of the clock's 25 years.

The sea does not stay still. "Below mean sea level" is evaluated against a fixed EGM2008 datum in every year to 2050. Ohenhen et al. measure absolute sea-level rise in the Java Sea at 5–7 mm/yr over 2001–2024, 25–75% above the global mean; twenty-five years of that is 12.5–17.5 cm. Raising the datum by 15 cm moves the 2050 below-MSL count from 192,155 to 227,284 and the below-+1 m count to 536,337. Ten centimetres of sea level is worth 40% of everything twenty-five years of subsidence achieves.

The rate is not a constant. Section 7 measured a halving time of about 4.85 years at CJKT. Running the clock with the rate decaying on that schedule rather than held flat gives 6.8 equivalent years of subsidence by 2050 instead of 25, and a 2050 count of 422,303 against the published 468,034 — 59% of the projected rise disappears.

Two of these make the clock optimistic and one makes it pessimistic, and the case's page claims only the first direction: it argues the uncorrected elevation makes the clock "if anything, optimistic" without mentioning that its constant-rate assumption pushes hard the other way. The honest statement is that the 2050 number is bracketed by roughly 422,303 to 508,612 on assumptions alone, before any uncertainty in the measurement and before the factor of eleven in Section 9.

There is precedent for getting this wrong. Chaussard et al. (2013) closed with exactly this kind of extrapolation: "assuming subsidence continues at an average rate of 10 cm/year and an average elevation above relative sea level of 2 m, … the coastal part of Jakarta will be below relative sea level in 20 years" — that is, by about 2033. Thirteen years later, the same coast is measured at under 4 cm/yr by the field this case uses. The projection failed not because the measurement was wrong but because a rate was assumed constant while the policy that drove it changed.

12 Finding eight — exposure is being read as risk, and the best-known number about Jakarta is not evidence

The case is careful to call its output exposure. Its page furniture is not: a countdown, a play button, a coral ramp and a year that ticks forward all invite the reading "this is what will happen." Exposure is one of three factors. Hazard — how often and how deep water actually arrives — and vulnerability — who can absorb it — are both absent.

The case's own failed plausibility check is the evidence. Its exposure index correlates with recorded 2021–24 flood events at ρ = 0.164, against a pre-registered threshold of 0.5. Low ground correlates negatively (ρ = -0.135): Jakarta's frequent floods are riverine, along the Ciliwung and its tributaries, on higher ground in the east. The case publishes this failure in red and diagnoses it correctly, which is more than most such products do — but it then treats it as a limitation rather than as the finding. Where Jakarta floods and where Jakarta sinks are different places, and an agency that plans drainage from the subsidence map will defend the wrong neighbourhoods. The one place the two coincide is the January 2020 extreme, which covered 12% of the 20 most exposed kelurahan against 5% city-wide — a signal for rare compound events, not for the annual flood season.

The same discipline should be applied to the single most-cited fact about this city. "Forty per cent of Jakarta lies below sea level" traces to a World Bank working paper of 2011, whose own figure caption hedges it as what the area "is said to constitute", sourced to a provincial mapping department dataset of 2000 and footnoted to a non-peer-reviewed trade book. No DEM, no vertical datum, no method. On this case's grid the figure is 1.6% of DKI's land area on the generous cell-minimum estimator (10.5 km² of 651), or 4.2% if every cell the DEM pins to exactly 0.00 m is counted as below — and 0.2% on the cell mean.

We are not able to say the 40% figure is wrong, because there is nothing to test it against: no peer-reviewed Jakarta-specific estimate of area or population below mean sea level exists. Hooijer and Vernimmen (2021) put 1.7 million Indonesians below 0 m with a 68% interval of 0 to 5 million, on a grid too coarse to resolve a city. What we can say is that the most repeated quantitative claim about Jakarta rests on a citation chain that does not terminate in a measurement, and that every modern elevation model puts the true figure at least an order of magnitude lower.

Those three numbers are the honest summary of the field, and none of them appears on the case page. 4.1% of residents on fast ground is a serious, actionable finding. It is not the same claim as "Jakarta is sinking", and the difference between them is where the policy lives.

13 What follows for decisions

Evidence is only worth gathering if it changes an action. Read strictly, this instrument supports four uses and forbids a fifth.

1. Redrawing the groundwater-free zone. DKI's current instrument, Pergub 93/2021, prohibits extraction from 1 August 2023 for buildings above 5,000 m² or eight storeys, across an enumerated list of corridors and precincts concentrated in the Central and South Jakarta business districts. Central Jakarta is the one municipality this field measures as not subsiding at all (+0.01 cm/yr mean). The fast ground is in West Jakarta. Widodo et al. (2025) reach the same conclusion from L-band InSAR and rank the redesign priority West, North, South, East, Central. This case's own ranking independently supports that reordering, and does not say so.
2. Verifying that the policy worked. Section 7 is a quantitative statement that Jakarta's subsidence responded to a decade of intervention, measured on an instrument independent of the one that produced the field. That is the highest-value output here — it tells a ministry which lever moved the ground — and it is currently discarded as a calibration residual.
3. Asset-level engineering triage. Subsidence on high inland ground threatens foundations, buried pipe and drainage gradients — PUPR, PAM Jaya, the toll and rail operators — long before it threatens anyone with seawater. The exposure map cannot see this because it filters on elevation.
4. A design-life prior for coastal defence. For the NCICD sea wall and the North Jakarta embankments the relevant quantity is relative sea level: subsidence plus sea-level rise, at the structure. Ohenhen et al. publish exactly that for two points in Jakarta Bay. This case has all three components and combines only one of them.
5. Not: a countdown to 2050. On the case's own grid the 2050 headline moves by a factor of 10.87 on the ground estimator, by 12 years of clock on the elevation epoch, and by 59% on whether the rate keeps falling. Exactly 1 of 261 kelurahan crosses mean sea level inside the window being animated. The clock should be relabelled as a momentum illustration with its assumption bracket shown, or withdrawn.

The gap this case should close is not between Jakarta and data — Jakarta has four decades of subsidence data. It is between the measurement and the licence decision, the pipe replacement, the design life. That gap is crossed by publishing where the ground is still moving and who is allowed to pump there, not by a countdown.

14 What remains open

1. An independent reprocessing. The case's own deferred check — its own LiCSBAS run over LiCSAR frame 098A against the deposited field — remains the right test and remains not done. Section 5 validates the regrid against the depositors' published table, which is a different and weaker thing than an independent interferometric chain. Section 3's factor-of-four spread is the reason it matters.
2. A ground model with a stated error budget. Section 9's factor of eleven is not resolvable by choosing a different percentile. It needs a bare-earth model whose vertical datum and noise are known at building scale — DeltaDTM, Indonesia's DEMNAS, or municipal lidar — which replaces an estimator argument with a measurement.
3. Relative sea level, not ground alone. Combining the subsidence field with a published local sea-level projection is a day's work on data already here, and it is the number every coastal-defence decision actually needs.
4. Extend the GNSS test past 2022. The archive ends where the interesting part begins. Whether the deceleration of Section 7 held through 2023–2026 — across the 1 August 2023 ban coming into force — is answerable from the same network, and would turn a retrospective into a monitor.

15 References and reproducibility

1. Ohenhen, L.O., Shirzaei, M., Kumar, P., Aditiya, A., Tiwari, A., Davis, J.L., Kolawole, F., Chaussard, E., Sadhasivam, N., Dasho, O., Zhong, W., James, R.H., Daramola, S., Nicholls, R.J. & Minderhoud, P.S.J. (2026). Land subsidence on Java Island and its contributions to relative sea level change. Science Advances 12, eaec0172. doi:10.1126/sciadv.aec0172. Dataset: Zenodo doi:10.5281/zenodo.15786356 (CC BY 4.0) — the velocity field under review; per-municipality statistics in table S2.
2. Susilo, S., Salman, R., Hermawan, W., Widyaningrum, R., Wibowo, S.T., Lumban-Gaol, Y.A., Meilano, I. & Yun, S.-H. (2023). GNSS land subsidence observations along the northern coastline of Java, Indonesia. Scientific Data 10, 421. doi:10.1038/s41597-023-02274-0. Dataset: Zenodo doi:10.5281/zenodo.7775016 (CC BY 4.0) — the daily station series refitted in Sections 6 and 7.
3. Abidin, H.Z., Andreas, H., Gumilar, I. & Wibowo, I.R.R. (2015). On correlation between urban development, land subsidence and flooding phenomena in Jakarta. Proc. IAHS 370, 15–20. doi:10.5194/piahs-370-15-2015 (table 1). See also Abidin, H.Z. et al. (2011), Land subsidence of Jakarta (Indonesia) and its relation with urban development, Natural Hazards 59(3), 1753–1771, doi:10.1007/s11069-011-9866-9.
4. Chaussard, E., Amelung, F., Abidin, H. & Hong, S.-H. (2013). Sinking cities in Indonesia: ALOS PALSAR detects rapid subsidence due to groundwater and gas extraction. Remote Sensing of Environment 128, 150–161. doi:10.1016/j.rse.2012.10.015 (table 2).
5. Abdullah, F.M., Andriyanto, H., Nababan, J.R., Abdillah, F. & Sulistyawan, R.I.H. (2021). Results of land subsidence measurement using GPS method in the Jakarta groundwater basin in 2015–2019. IOP Conf. Ser. Earth Environ. Sci. 873, 012034. doi:10.1088/1755-1315/873/1/012034.
6. Sidiq, T.P., Gumilar, I., Abidin, H.Z., Meilano, I., Purwarianti, A. & Lestari, R. (2025). Spatial distribution and monitoring of land subsidence using Sentinel-1 SAR data in Java, Indonesia. Applied Sciences 15(7), 3732. doi:10.3390/app15073732.
7. Lapadat, A.M., Andreas, H., Brouwer, W.S., van Diepen, S., Pradipta, D. & Hanssen, R.F. (2025). Land subsidence in Jakarta in three dimensions (2014–2025) using InSAR–GNSS datum connection and the strapdown decomposition. EarthArXiv preprint, doi:10.31223/X5R45H. Not peer-reviewed at the time of writing.
8. Widodo, J., Trihatmoko, E., Setyaningrum, N., Izumi, Y., Handika, R., Ardha, M., Arief, R., Sobue, S., Nurlinda, N., Pranantya, P.A., Wiranu, J.R. & Khomarudin, M.R. (2025). Technical and policy analysis: time series of land subsidence for the evaluation of the Jakarta groundwater-free zone. Urban Science 9(3), 67. doi:10.3390/urbansci9030067.
9. Pronk, M., Hooijer, A., Eilander, D., Haag, A., de Jong, T., Vousdoukas, M., Vernimmen, R., Ledoux, H. & Eleveld, M. (2024). DeltaDTM: a global coastal digital terrain model. Scientific Data 11, 273. Dataset doi:10.4121/21997565.
10. Hooijer, A. & Vernimmen, R. (2021). Global LiDAR land elevation data reveal greatest sea-level rise vulnerability in the tropics. Nature Communications 12, 3592. doi:10.1038/s41467-021-23810-9 (table 1).
11. World Bank (2011). Jakarta: urban challenges in a changing climate. Mayors' Task Force on Climate Change, Disaster Risk and the Urban Poor, working paper 65018 — the source of the "40% below sea level" figure, there attributed to DKI Jakarta Department of Land and Mapping (2000).
12. Pemerintah Provinsi DKI Jakarta (2021). Peraturan Gubernur No. 93 Tahun 2021 tentang Zona Bebas Air Tanah. Promulgated 26 October 2021, Berita Daerah 2021 No. 63013; prohibition effective 1 August 2023 (article 8) for buildings ≥ 5,000 m² and/or ≥ 8 storeys (article 2).
13. European Space Agency / Airbus (2022). Copernicus DEM GLO-30 global 30 m digital surface model. doi:10.5270/ESA-c5d3d65. WorldPop (2020), Global High Resolution Population Denominators, Indonesia, doi:10.5258/SOTON/WP00660. Pesaresi, M. & Politis, P. (2023), GHS-BUILT-S and GHS-POP R2023A, European Commission JRC.

Corrections applied. This review changed seven claims on the case page; each correction is live there.

Provenance. The only hand-entered numbers are the published literature rates in Figures 1 and 2 and the Ohenhen table S2 column in Section 5, each transcribed from the sources cited above with the period it was measured over. Field vintage 2017–2023; snapshot 2026-08-30.