Methodology

From datasets to pages

The starting point is a set of open geographic files: populated places, rivers, lakes, peaks, landforms, glaciated areas, dams, ports and ancient sites. A build step reads each file and produces a record per named feature, with a name, coordinates and whatever measurements the source holds, such as population, elevation, dam height or harbour depth. Polygon features are given a centre point at the middle of their bounding box, which is what the page map is centred on.

Records that carry nothing to describe are dropped at this stage. Unnamed ice polygons are not turned into pages, and port records with a placeholder name or no size class and no depth figures are excluded. Common map abbreviations are expanded, so a source label such as "Mt." is written out as "Mount".

How Wikipedia summaries are matched

Many place names are shared by several places. To avoid printing the summary of a different place with the same name, a Wikipedia article is accepted only when the coordinates recorded for that article lie within a set distance of the coordinates the atlas already holds. The distance depends on the kind of feature: it is small for a point such as a summit and large for a feature that extends across a region.

Distance alone is not enough, because a river and the province named after it lie close together. An article must therefore also be the right kind of thing: its own short description has to say it is a river for a river page, a mountain for a peak page, a city or town for a city page, and so on. Where an article's coordinates are held on Wikidata instead of in the article, those are used for the distance check. Point features such as cities, peaks, dams, lakes and ports are matched only when coordinates are available from one of the two; for larger features an article with no coordinates is accepted only when its title is exactly the name of the feature. Disambiguation pages, list articles and very short stubs are rejected. A port is matched only to an article about the harbour itself, not to the article about the town beside it, and a river is not matched to an article that describes itself as a town or city of the same name. Country summaries use the article that Wikidata links to that country. Most city records carry a Wikidata identifier in the source data, and for those the article that Wikidata links to the identifier is used, subject to the same distance check.

Some entities therefore have no summary. Either English Wikipedia has no article under any of the names tried, or the article is about a different kind of thing, or its coordinates fall outside the permitted distance or cannot be found. In each case the page shows its data without a summary. An absent summary is preferred to a wrong one.

The summary is the opening of the article, shortened at a sentence boundary, and each one links to the full article. Lead images are requested only where the file is freely licensed, and each image links to its Wikimedia Commons page, which records the author and licence.

How countries and states are assigned

Where a dataset states a country for a feature, that statement is used. Cities, ports, peaks, dams and lakes mostly carry one. The different spellings and codes used by the sources are reconciled to ISO 3166-1 codes. A peak or lake that its source places in two countries is listed under both.

Lakes, islands, glaciers, mountain ranges, deserts, plateaus and other landforms are listed under every country they lie in, not only the one at their centre. Each mapped outline is sampled at about 600 points and compared with the Natural Earth country boundaries, which gives the share of the feature in each country. A country is included when it holds at least a twentieth of the feature, at least 25,000 km² of it, or a part covering a tenth of the country itself. Natural Earth draws the largest ranges and deserts as rough outlines, so the countries Wikidata records for each feature with an encyclopedia article are added as well. The Sahara is therefore listed under each of the countries it crosses, and the Himalayas under Nepal and Bhutan as well as India and China. A feature that spans several countries is written up in full on each of their pages, and that write-up starts from the part of the feature in that country.

Rivers, and any feature that none of these sources places, are given the country of the nearest mapped city, provided that city is within 200 km of the feature's centre point. A mountain, lake, island or landform with no city in that range is given the country whose Natural Earth boundary contains its centre point, or the nearest boundary within 300 km, so that a remote Antarctic plateau is still described on a country page. Other features with no city in range are left unassigned and appear in no country table.

The nearest-city rule can misplace rivers near a border. A river reach whose midpoint is closer to a city across the frontier is listed under the neighbouring country. A long river is listed under the countries where the midpoints of its mapped reaches lie, which may not be every country it flows through.

States and provinces are built from the first-level division named for each city. Peaks, lakes, dams and ports take the state of the nearest city in the same country, within 150 km. Rivers and islands are not assigned to states.

Which entities get their own page

An entity gets a page when there is enough to say about it. The rest remain in the data and are listed as rows in the tables for their country and state, without a page of their own.

Cities

None. Every city is a row in the city table of its country's page, with its population, region and role, and the cities of a state or province are listed on its page too. The harbour of a port city is described on the page of its sea. The cities page lists the national capitals, the largest cities and the great port cities.

Peaks, lakes, islands, ranges, deserts, plateaus, glaciers and other landforms

None. Each one with a matched Wikipedia article is described on its country's page for that kind of feature: highest mountains, largest lakes, largest islands, or mountain ranges, deserts and landforms. Where the country has fewer than four mountains or landforms, or fewer than five lakes or islands, the description is on the country page itself. A feature on a border is described under the first country listed for it. A reservoir whose record names a dam is described with that dam.

Ports

One page per sea or coast, not per harbour. Every harbour, from the World Port Index and from the harbour data the source carries on port cities, is listed on the page of the sea it lies on: the IHO sea area that contains it or, for a harbour up an estuary or on a lake, the nearest one. Harbours on the open oceans are grouped by coast and country, and those of the Great Lakes and the St. Lawrence above Montreal have a page of their own. A harbour with a matched Wikipedia article is described in full there.

Rivers

A matched Wikipedia article. Where the source divides one river into several named reaches that resolve to the same article, one page is built and the other reaches point to it.

Dams

None. Each dam is described on the page of the river it stands on, or, when that river has no page, on its country's dams page.

Archaeological sites

None. A site that is itself a World Heritage Site (the two resolve to the same article or stand within 1 km of each other) is described on the heritage page; every other site is described on the archaeology page.

Wonders

None. Each wonder is described in full on its country's page.

States and provinces

At least four mapped cities in the division. States of the United States always have a page.

Countries

An ISO 3166-1 country code with at least one mapped city.

Where a source holds the same feature twice under two names, and both records resolve to one article, one page is built. The current build stands as follows.

Areas, ranks and counts

The areas of lakes, islands, deserts, ranges, plateaus and glaciers are computed from the outline in the map data, treating the Earth as a sphere. The outlines are generalised for display at small scales, and islands within a lake are not subtracted, so a computed area can differ from a surveyed figure by several per cent. Glacier and lake outlines also record the extent at the time of mapping, not the present one.

Country population and area are taken from Wikidata. Each country's figure is the one recorded there, so the reference year differs from country to country. City populations are those given in the populated-places dataset, which describe the wider urban area and are usually larger than the population inside the city limits. The population shown for a state or province is the sum of its mapped cities, not a census total.

The rankings sort the entities that are in the atlas by one stated measure and show the first fifty. A rank is a position among indexed entities, not among everything on Earth: a dam or peak missing from the sources is missing from the ranking. Counts given on country pages, such as the number of ports or lakes, are counts of mapped records and depend on the assignment rules above.

Historical polities

The empires and states catalogue combines records from a comparative databank of historical polities with summaries from the corresponding Wikipedia articles. Start and end years are those of the source and often mark a convention, such as a founding or a conquest, on which historians differ. Mapped territories are approximate. Pre-modern states seldom had surveyed borders, and an outline on the map indicates a zone of control at a given date, not a line that existed on the ground.

Licences

Text summarised from Wikipedia is used under the Creative Commons Attribution-ShareAlike 4.0 licence, with a link to the source article on each page. Wikidata is released under CC0. Natural Earth map data is in the public domain. Images are served from Wikimedia Commons under the licence stated on each file's page.

The other sources, each with its licence and a link to the publisher, are set out on the data sources page, which is the authoritative list.

How the data is refreshed

The pages are rebuilt from the sources; they are not edited individually. A change at a source appears here after the data is next fetched and the site rebuilt. Each source has its own reference date. A map dataset reflects the date of its release, a Wikidata figure the date of the estimate recorded there, and a Wikipedia summary the state of the article when it was fetched. A page can therefore combine facts of different ages.

Known limitations

• Coverage follows the sources. The atlas does not contain every dam, peak, port or town, and coverage is uneven between regions.
• Features near a border, and features assigned by the nearest city, can be listed under the wrong country or state.
• The geographic check reduces wrong Wikipedia matches but cannot exclude them, particularly for large features where the permitted distance is wide.
• Names follow the source datasets, which mostly use English or romanised forms. Other spellings may not be found by the site search.
• Disputed territories and borders are shown as the sources show them. The atlas takes no position on them.
• Wikipedia summaries are in English only and reflect the article at the time of fetching.
• The pages are generated automatically and are not individually checked by an editor or a subject specialist. Figures used for any consequential purpose should be confirmed against the original source.