A place often has several official numbers. What separates them is almost never measurement error — it is that they measure different things: the island or the whole prefecture, with or without disputed territory, registered residents or usual residents. This page states how this site chooses, and what choosing wrongly costs.
Six rules
1. The figure, the outline and the population must describe the same land
If any one of the three covers a different extent, the resulting density means nothing. Hokkaido is the clearest case: Japan publishes 83,422.27 km², which includes four disputed island groups, but the outline data stops at Cape Nosappu and no census is conducted there. This site uses 78,419.22 km² without them, because Japan’s own Statistics Bureau makes the same deduction before computing density.
The cost: Hokkaido reads as 20.75% of Japan here, against 22.07% in the scope-consistent official calculation. That is why no Hokkaido-versus-Japan page exists on this site.
2. If no primary source isolates the island, file the province and say so
Jeju and Bali are both handled this way. Circulating figures for "the area of Jeju Island" run 1,833 / 1,845.9 / 1,846 / 1,847 / 1,849.1 — that spread is the evidence against using any of them, not noise around a true value.
A provincial figure never goes under an island heading. The page is titled Jeju Province, not Jeju Island, because of this rule.
3. An administrative boundary is frequently not a land outline
Jeju’s provincial relation encloses about 11,100 km²; Bali’s about 21,169 km², 3.8× its land; Macao’s SAR boundary includes 85 km² of sea; Singapore’s national relation is the treaty maritime boundary at roughly 1,646 km². Before adopting any administrative relation, measure what it encloses against the official land area.
All four switched to island polygons, with the gap between outline and statistical scope disclosed on the page.
4. Area is never derived from the outline
Outlines draw; official figures measure. The OSM boundary for Tokyo’s 23 wards extends over Tokyo Bay and yields about 724 km², 15% above the official 627.51. Taiwan’s OSM polygon closes river mouths at the bay line and runs about 0.9% high.
Rendered size always comes from the official area, never from the polygon — so an imprecise outline costs shape fidelity, not scale.
5. The population concept travels with the number
Singapore publishes a total population of 6.11m, a resident population of 4.20m and a citizen population of 3.66m. Choosing the wrong one moves density by a third or more. South Korea’s register-based census gives Seoul and the country to the same precision, which is why the national figure here is 51,805,547 rather than the rounded 51,806,000.
Every place states its population definition in its own scope section, not in a footnote.
6. When a source states no measurement date, the date shown is the date it was read
Hainan Island’s 33,900 km² comes from the provincial government, but no survey date appears anywhere on the official page. Rather than invent one, the site labels it as an access date.
Penghu and Green Island are handled the same way.
What was rejected
The following were investigated and then not done. They are listed because why something was not done usually says more about a dataset than what was done.
Built-up share is kept off comparison pages
Five places publish an official built-up figure, but publishers draw the line so differently that the spread exceeds what is being measured: Singapore’s 74.1% is the IPCC "settlements" class, Taiwan’s 5.4% counts building footprints only. Side by side they read as a 14× difference that is almost entirely definitional. Expressing it as a multiple of the territory average inherits the same distortion. The figure therefore appears on place pages and on no comparison page.
Macao built-up land: verified absent, not omitted
After searching the official statistics, Macao publishes no comparable built-up figure. Three workarounds were tried and each amounted to manufacturing a number, so the field is left empty.
Islands do not use Wikidata for sameAs
Wikidata returns an airport, a township and a lighthouse for Green Island, and Miyakojima City for Miyako. That is exactly the class of scope error this site exists to avoid, so islands use individually checked OpenStreetMap relations; only countries and administrative areas use Wikidata.
No travel-scale measures
Distances such as "driving the length of the island" have no consistent official source across places. Producing them would mean estimating, and an estimate would undo everything else on this page.
Orchid Island is excluded from the Taiwan reference ladder (but has its own place page)
The reason previously given here was that the row carries an unexplained "※" marker. That marker has since been identified: it flags a mountain indigenous township, thirty rows carry it, and it says nothing about the figures — so the old reason is withdrawn. Orchid Island still stays off the ladder, now for a different reason: the run from Green Island 15.1 to Matsu 28.8 to Magong 34.0 to Penghu 128 is already dense enough, and another rung there would not make any place easier to picture. Orchid Island has a place page of its own, from the same table.
Kinmen has figures but no place page
Kinmen County’s area and population come from the same table as the rest of this site’s Taiwan data (151.656 km² and 139,313 residents at year-end 2025), and the county total reconciles exactly with its six townships. What is missing is an outline. The OpenStreetMap county boundary encloses 998 km², 6.6 times the land, because it traces maritime jurisdiction. Natural Earth’s "Kinmen" actually draws only the main island — it leaves out Lieyu and Wuqiu — and overstates even that island by about 15%. The union of the six township boundaries runs 14% over. The rule on this site is that a failed outline means no page, and no substitute is drawn by hand, so Kinmen remains a yardstick on the reference ladder and nothing more.