Funafuti is restless.

Pacific islands are usually described as vanishing under water. The reality is more complex in places like Funafuti. Here we look at its shape-shifting. The sea level in this part of Tuvalu swings millimetres between one year and the next, with a floor level that has lifted millimetres in two decades. Funafuti's shoreline and population are adapting.

click to isolate

+ mm higher floor in two decades

Funafuti's own tide gauge rises on average ~ mm a year, similar to the world average. However, movements in Funafuti are not smooth: the median year-to-year was mm over the same period.

If we look at levels rather than rates, measured against the same gauge gauge's – average, Funafuti sat at + mm before , while since it's at + mm. That is a floor mm higher, in two decades.

Sources UHSLC tide gauge, station 25 · NASA global mean sea level · NOAA

Methodology & sources

Every figure on this page comes from the datasets below. Note that each figure is recomputed in notebooks/figure-audit.ipynb in the GitHub repository, which fails if the page and the data disagree.

1 · Datasets

DataIn this repoSource & licence
Daily tide-gauge readings, station 25 (Funafuti) sources/raw/uhslc_daily_TV_25.json University of Hawaii Sea Level Center
Population by elevation band, DF_POP_LECZ sources/raw/POP_LECZ.csv SPC Pacific Data Hub .Stat
Greenhouse gas per person, GHG_EMI_CAPITA sources/raw/GHG_EMI_CAPITA.csv SPC Pacific Data Hub .Stat (Tuvalu row only)
Annual coastlines and rates_of_change, v0.7.0 (Landsat, 30 m) data/funafuti_2023.json, data/funafuti_points.json Digital Earth Pacific / SPC GEM. CC BY-NC 4.0 (this project is non-commercial).
2017 Population & Housing Mini-Census, usual residents sources/raw/Mini-Census-2017-Preliminary-Report.pdf and its text extract Tuvalu Central Statistics Division, preliminary report

2 · Cited figures that are not computed

  • Global mean sea level, quoted as a comparison (not a figure): Neither NASA Sea Level Change nor NOAA publishes an average over the satellite record; both publish the total, and the two totals do not agree. NOAA puts 2023 at 101.4 mm above 1993 levels, i.e. 3.4 mm/yr over 30 years; NASA/JPL says "10 centimetres since 1993", i.e. 3.2 mm/yr over 31. Funafuti's mm/yr sits inside that, so the page says the two are similar and quotes no number for the world. Today's global rate is higher, NASA gives 0.44 cm/yr, about mm/yr, against 0.20 cm/yr in 1993.
  • World emissions, t CO₂e per person: EDGAR v9, as used by the UNEP Emissions Gap Report 2024. All gases, excluding land use, matching the Tuvalu figure.
  • Island growth, + ha (%), –: Kench, P.S., Ford, M.R. & Owen, S.D. (2018), Patterns of island change and persistence offer alternate adaptation pathways for atoll nations, Nature Communications 9, 605; Table 1, Funafuti row (the figures are not in the article text). 19 of Funafuti's 29 islands grew, 10 eroded.
  • Reclaimed land, ha: Tuvalu Coastal Adaptation Project, two phases with a source each. ha completed late under TCAP-I (UNDP / Green Climate Fund), stated in UNDP's release on the second phase. ha completed 14 October under TCAP-II (Australia, New Zealand and the United States), reported at the handover by the UN in the Pacific. UNDP's own project pages describe the second phase only in the future tense, so they are not the source for its completion, but media sources evidence it was completed.
  • Falepili ballot, registrations for places a year: Australian Department of Foreign Affairs and Trade, and the Development Policy Centre's preliminary assessment of the first ballot.
  • ICJ advisory opinion, 23 July 2025: Obligations of States in respect of Climate Change, case 187, given for context.

3 · What was done, chapter by chapter

1. Sea level. Daily readings → monthly means (≥ 15 days) → annual means (≥ 10 months), as anomalies against the gauge's own – mean. The quoted trend is Theil–Sen, mm/yr, i.e. the median of all pairwise slopes. Least squares gives mm/yr, but 1998 sits at mm (the 1997–98 El Niño) near the start of the record and levers the line: drop it and the same fit gives . A claim that Funafuti is rising markedly faster than the world would rest on that one year, so the page does not make it. The mm swing is the median change between consecutive years; the + mm floor is median-to-median between – and –, for the same reason.

2. People. The % and % are read straight from DF_POP_LECZ. The census figures are read out of the report's own text, and are usual resident counts throughout, of , which reproduces the report's own stated 60.2% share. Where the report's prose and its table disagree we follow the table, which is arithmetically self-consistent.

3. Imbalance. Tuvalu's t against the world's t, all gases as CO₂e excluding land use on both sides, divided is circa ×. The source publishes to the nearest 0.1 t, so the page says "about". The chart is on a linear scale, with the world's bar cut into units of Tuvalu's own figure so the ratio can be counted rather than asserted.

4. Shoreline. Of points in the Funafuti bounding box, carry good certainty by the source's own flags; of those, have a trend significant at p ≤ 0.05 and do not. The are counted rather than discarded, because a shore that does not move is also a the finding. Symbols are nudged a few pixels off the shore, sitting a median m apart. The years before 2012 are marked low-confidence throughout, as up to 45% of points are flagged as outliers in 1999, 2000, 2007, 2008 and 2010; 2002 and 2005 are missing entirely.

5. Built. The figures come from various reports. The only arithmetic is putting the two cited figures on one axis: a rate each ( against ha/yr, ×) and a share of the atoll (% of ~ ha). The paper's Table 1 prints the atoll at 261.2 ha in 1971, which our re-calculation of 10.06 ÷ 3.85% reproduces to 0.1 ha; the end figure adds the gain to it, so it is good to about ±1 ha.

6. Leaving. One division: ÷ = years. Both count the same kind of thing, entries and places for entries, which is why it is the only arithmetic done here.

Limits

  • A shoreline is a waterline and not necessarily land. It says where the shore sits, not how much habitable land exists, and it cannot tell engineered land from natural accretion. This data shows where the shore moved, never why.
  • The gauge measures sea level relative to the land; note that the fast-delivery data used receive less quality control than the research-quality series.
  • The atoll size vs pixel size may create artifacts. I'm not 100% sure about this, but note that the atoll is × km, so on my monitor for example one screen pixel covers – m against a median movement of m across the whole record, which is too close to visualize properly (or so I suspect).
  • People are never mapped. The census counts residents by island, not by address, so the map carries one marker only in Fogafale.
  • The land comparison mixes two scopes, on purpose. Kench et al.'s gain is the whole atoll; the reclamation is on Fogafale alone.
  • A registration is an entry, not a person. No official source splits into applicants and dependants, and families are reported to have entered more than once, so it is never divided by a population. Also, the widely quoted "more than half of Tuvalu's citizens" does not appear here.

One warning, and further reading

This page's shoreline finding, i.e. that most of Funafuti's coast did not measurably move, and that more of what moved grew than shrank, resembles a claim climate denial outlets have made use of since Kench et al. published. That reading requires ignoring what the measurement is and its context. As said, a shoreline is a waterline, not habitable land; the atoll gaining sediment says nothing about whether the runway floods, whether the freshwater stays fresh, or whether the sea is rising (which we know it is, on account of Funafuti's own gauge). This page tries to chart all these dimensions objectively, and give a unifying narrative. In short, growth and danger are not alternative stories: a landform that moves, under infrastructure that cannot.

Credits

Built for the Pacific Dataviz Challenge 2026. Data © their providers: SPC / Pacific Data Hub, University of Hawaii Sea Level Center, and Digital Earth Pacific / SPC GEM (CC BY-NC 4.0). Code MIT; typefaces Fraunces and Inter, SIL Open Font License 1.1.

No framework and no build step: hand-written HTML, CSS and ES modules, D3 vendored locally. Data prepared with Python in scripts/ (see its README for the run order), the 2 GB coastlines GeoPackage is never downloaded, GDAL /vsicurl reads only the Funafuti byte ranges. Coded and tested using a Claude Code assisted workflow.

Everything behind this page — the data pipeline, the notebook that recomputes every figure, the raw sources and the page itself — is at github.com/puntofisso/pacific-dataviz-challenge.