GTAmap Global multi-year administrative divisions
About

About GTAmap

GTAmap’s goal is not another “global boundary” file but an administrative-division dataset that can answer “which year, and on what evidence”. This page explains where the data comes from, where it still falls short, and which conclusions it cannot support.

Introduction

Why administrative divisions need a time axis

Administrative divisions are not constants. Between 2000 and 2025 Brazil created and merged thousands of municipalities; France has pushed commune mergers since 2017, with 3,782 in total; Japan’s Heisei consolidation involved 1,819 adjustments; Mali expanded its second-level units from 53 to 160 in 2023. Back-filling any “current boundary” into an analysis of ten years ago silently introduces errors — and usually you cannot see them.

Mainstream global boundary datasets (GADM and others) provide a single time slice: one “present-day” set of boundaries. Count 2000 data with 2020 boundaries and the number of units, the areas and the population denominators are all misaligned, and the data will not tell you.

GTAmap’s approach is to bind three things to every feature:

  • Time: snapshot_year + date_precision (day / year / start-of-year / census year / version year / interval), so “which moment does this boundary represent” has a definite answer;
  • Source: src_name / src_url / src_vintage / src_license point feature by feature to a downloadable origin, instead of a vague “multi-source integration”;
  • Confidence: conf_grade (A–D) + conf_note state the nature of the source and its known issues, so you know whether it can serve as a legal boundary.

The third asset is the event library: 20,006 change events recording splits and mergers (change type, effective date, legal reference, confidence). It explains why boundaries moved and can also be used to cross-check whether the time series itself is complete.

Scale & coverage

Scale and coverage

The data is “full coverage at graded precision”: all 243 entities are present, and precision is stated honestly entity by entity.

Global unit counts and real-geometry coverage in the 2025 anchor year
Level Units of which real geometry Coverage Nature of main sources
ADM0 country/territory 24323295.5% Direct official collection + aggregators + a few derived (composite ADM0)
ADM1 first-level 3,4923,42097.9% Mostly direct official collection, with GADM / OSM substitutes
ADM2 second-level 55,07554,15198.3% Mostly direct official collection; uncovered units derived from the union of their parents
244
Country/territory time-series packages
243 entities + EU NUTS
511,244
Total time-series rows
One row = one unit’s boundary in one snapshot year
20,006
Change events
177 event datasets in v1.1
43.01 GB
Release package size
Per-file sizes and SHA-256 checksums are listed in manifest.csv.

Temporal coverage is very uneven — the most important known characteristic

GTAmap covers “the globe”, but not “the globe year by year”. The number of years varies enormously: a few countries have 27–34 real annual snapshots, most have only 1–5. The table below groups entities by number of years — use it to judge whether a country supports year-by-year analysis.

These detailed counts describe the v1.0 baseline. Current v1.1 coverage is available in the download catalog.
Years Entities Share Examples and notes
1 snapshot 9038.6% Mostly small dependencies outside the GADM dual anchors; no time-series analysis possible
2–5 snapshots 11649.8% GADM dual anchors (≈2018 / ≈2022), census years, election cycles
6–15 snapshots 156.4% United States 16, France 10, EU NUTS 6 editions, etc.
16–25 snapshots 73.0% Brazil 18, Japan 22, Netherlands 23, Germany 19, etc.
26+ snapshots 52.1% China 27 (2000–2026), Italy 27, Iceland 27, Estonia 30, Norway 34

These detailed counts describe the v1.0 baseline. Current v1.1 coverage is available in the download catalog.

Provenance

Composition of data sources

Each letter describes source provenance, not an overall quality score. One country may have different grades at different administrative levels. Use the feature-level conf_grade and geom_origin fields when filtering or citing.

Number of countries per grade with typical institutions
Grade Meaning Entities Typical sources
A Official, collected directly Native boundaries from national statistical or mapping agencies 52 Real official geometry for that year
B Official via aggregator / GADM fallback Official data passed through an aggregator, or replaced by a GADM anchor 169 Derived from the union of official lower-level units
C · OSM / third party Community or third-party sources 20 Replaced with a GADM anchor
D · Derived / inferred Derived from the union of other levels or from research methods 2 Back-filled with the 2026 baseline geometry (to be fixed)

The table assigns each entity its most conservative grade across ADM0–ADM2 (243 total). The 2025 ADM2 feature counts are A 31,452 · B 20,183 · C 2,815 · D 625; use each feature’s conf_grade for analysis.

Source channel distribution (src_type; a country may have several)
Source channelEntitiesNotes
official_via_unNative boundaries from national statistical or mapping agenciesReal official geometry for that year
official_via_gadmOfficial data passed through an aggregator, or replaced by a GADM anchorDerived from the union of official lower-level units
officialCommunity or third-party sourcesReplaced with a GADM anchor
research_derivedDerived from the union of other levels or from research methodsBack-filled with the 2026 baseline geometry (to be fixed)
official_via_aggregator16Official data via a third-party aggregation platform
derived3Derived from the union of parent units
aggregator_osm2OSM history pipeline (confidence C)
Other (mirrors / vendors / government partners / wikis)4Occasional sources; per-feature src_name tells you which

Geometry provenance field geom_origin

ValueMeaning
official_actualReal official geometry for that year
dissolved_from_officialDerived from the union of official lower-level units
official_via_gadm_anchorReplaced with a GADM anchor
current_carried_backBack-filled with the 2026 baseline geometry (to be fixed)
Want reliable geometry only? Filter on geom_origin == "official_actual"; a code example is in the home page quick start.
Methods & limitations

Methods and limitations

Each of the following affects how the data should be used. We would rather spell everything out than have you hit these traps in a paper.

Historical geometry coverage varies by entity

Some historical rows use geometry carried back from the 2026 baseline and are marked geom_origin=current_carried_back. Reusing one geometry across years does not show that a boundary remained unchanged.

Where historical official sources are unavailable, some features use two version anchors rather than annual observations. They are marked date_precision=version_year; inspect the date and source fields before time-series analysis.
For OSM sources (North Korea, Iran, Kazakhstan, Cuba and others, confidence C) the event date records when the change appeared in OSM data, not the legal effective date. Treat it as an upper bound rather than an exact value when counting the year a change happened.

An empty event table ≠ no boundary change

An empty event table for a single-snapshot area means there is no second comparable official basemap in the record; it does not mean that no boundary change occurred.

Some sources include lagoons, inland water or near-shore water in the boundary and others do not, which can make the same unit jump in area between comparable years. This is flagged in each feature’s conf_note, but run an outlier check before any area time-series analysis.
Temporal coverage varies widely by country. Check the actual years and levels in the download catalogue before designing an analysis; global year-by-year summaries may be unbalanced.
Change events are unevenly distributed because source availability and collection effort vary. Event counts reflect documentation coverage, not necessarily the rate of boundary change; account for this when comparing places.
Change events split or merge units, and unit_id changes accordingly (which does not contradict “stable IDs”: what is stable is “the same unit in the same year”, not “the same patch of land”). To track units across years, build the mapping from the lineage tables in the event library (*_lineage.csv).
Disputed areas: disputed areas currently use a single view (the gbOpen convention), and the multi-view mechanism has only been demonstrated for the Japan–Russia Northern Territories; it is not yet in place everywhere. If your research touches disputed boundaries, check src_url and conf_note first and prepare an alternative view yourself.
Map standard: The global base map follows the Chinese national cartographic standard for the China–India border areas and the Diaoyu Islands. Country packages distributed for download contain each country’s own official boundary representation. Other disputed areas follow international de facto conventions. See the data documentation for the complete boundary policy.
Known gaps

Known gaps (what comes next)

The gaps below are known and logged; from v1.1 the reachable official sources are being prioritised.

Back-fill to fix (17 entities)

Official sources are unreachable or were interrupted, so these currently use the 2026 baseline geometry flagged current_carried_back:

ALB · ARM · AZE · BIH · COD · CYP · DZA · GEO · GHA · IDN · KHM · MKD · PSE · SRB · TUN · UKR · XKX

Restricted sources (7 countries)

The upstream source requires login, payment or is blocked, so only substitutes are available short term:

KOR · HUN · RUS · BLR · KAZ · UZB · CUB

Currently substituted with GADM dual anchors or the OSM history pipeline.

Geometry grade to replace (27 entities)

Geometry quality is grade D and needs replacing whether or not a historical series exists:

AGO AZE BGR BIH BLZ BRN CMR CRI CYP CZE DNK DZA ETH GAB GEO ISR JOR KGZ LBN NPL PAK PRT SUR SWE UGA UKR URY

Five substitute routes (verified working, for entities whose official source is unreachable)
Route Applies to Grade produced Limitations
① HDX / UN OCHA COD-AB Conflict and crisis countries, most African and Pacific small states B Often includes several yearly versions, which can form a real historical series; watch the difference between humanitarian and official statistical definitions
② GADM multiple versions Countries with restricted official sources and all small dependencies B Only two time anchors (≈2018 / ≈2022); licence restricts non-commercial use
③ OSM history pipeline Countries with official blocks C Real history only from 2007 onward; dates are mapping times; ODbL applies
④ Parent-country sources for dependencies 52 dependencies / overseas territories A/B Requires mapping the administrative hierarchy between dependency and parent country
⑤ User assistance Sources needing an API key or login, such as Taiwan and South Korea A/B Depends on external cooperation and timing is uncertain; an official account would speed this up considerably
Found a wrong boundary or event? Submit it through the feedback page with iso3 + snapshot_year + your source link — feedback with sources is handled first.
Changelog

Version history

Three stages from “one current boundary” to “global time series + event library”.

  1. v1.0 — global time series released2026-09-26

    • 244 country/territory time-series packages (243 entities + EU NUTS), three levels, 511,244 rows in total
    • Event library expanded to 20,006 events (177 datasets) with change type, effective date, legal basis and confidence
    • 4 global anchor years (2000 / 2010 / 2020 / 2025) × ADM0/1/2 as uniform snapshots
    • GADM fallback pipeline completed: 60 entities with dual anchors, zero failures (52 small dependencies + 12 restricted + 4 blocked countries)
    • China integrated from CTAmap v2.1: 27 years × three levels = 88,209 rows for 2000–2026, 408 events with Ministry of Civil Affairs approval numbers, plus a separate ten-dash-line layer
    • Release package 48.1 GB / 706 files with full-file SHA-256 in manifest.csv and two inventories (by country, by event)
    • Per-feature provenance upgraded to 8 fields (src_id / src_name / src_url / src_license / src_vintage / src_type / conf_grade / conf_note)

  2. v0.3 — official replacement and historical series ingest2026-09

    • Replaced simplified geometry with official sources for Tier-1 countries: US TIGER, IBGE, IGN, MLIT/GSI, BIG Indonesia and more
    • First real historical series ingested: Brazil 18 years, six EU NUTS editions (37 countries), three US vintages, Italy 2 years
    • Event library expanded from 5,336 events, adding three collection channels (legal references / official version diffing / OSM history)
    • Established the conf_grade A–D criteria and the four-value geom_origin specification

  3. v0.2 — precision from national official sources2026-09

    • On top of the v0.1 global baseline, replaced geoBoundaries simplified versions for Tier-1 countries with official sources (better geometry precision)
    • Back-filled lineage and stable shapeIDs so units correspond before and after replacement
    • Built the official source register: 232 entities, 152 national agencies registered

  4. v0.1 — global uniform baseline2026-09

    • Built a 232-entity unified baseline on geoBoundaries: ADM0 232 / ADM1 3,235 / ADM2 49,305
    • It was a “currently available” snapshot, with different vintage years per country (2017–2025), not a uniform year-by-year baseline
    • Geometry for 106 of those entities came straight from OpenStreetMap — exactly the countries where later official substitution paid off most
    • First event library: 5,336 events + 2,040 lineage records
Version numbering: a major version (v1.0) means a breaking change in coverage or field specification; a minor version (v1.1) adds countries/years or fixes geometry without changing the schema. Always cite the version number.
Credits & contact

Credits and contact

Data source credits

  • National statistical and mapping agencies: IBGE, US Census TIGER, MLIT, CTAmap, IGN, BKG, ISTAT, SSB, SCB, StatCan, ABS, INEGI, Eurostat and others
  • UN OCHA humanitarian data (HDX COD-AB)
  • GADM (Global Administrative Areas)
  • OpenStreetMap contributors and Geofabrik
  • geoBoundaries (v0.1 global baseline)

Licence and attribution

  • Data: CC BY 4.0 (placeholder, final terms pending release)
  • Code: Apache-2.0
  • Upstream restrictions still apply (GADM non-commercial, OSM ODbL); see the licence note

Contact

Figures on this page: release version and file counts follow the GTAmap v1.1 manifest and final acceptance gate (2026-09-29). Coverage levels and source details are derived from country package manifests and GeoParquet metadata.

If a figure here disagrees with the release package, the package wins — please tell us via the feedback page.

Cartographic convention

Cartographic convention fields and guidance

The boundary policy is stated in the “Map standard” section above. This section explains how to identify, filter and reproduce that convention in the data. See the full boundary policy for the case-by-case treatment of Kashmir, Western Sahara, Crimea, Israel/Palestine and other areas.

The cartographic convention used for this unit’s geometry: CN_standard follows China’s national cartographic standard (China, Taiwan, and Indian/Japanese units related to the disputed China–India/China–Japan areas); international_default uses the prevailing international convention for other units.
Boolean field indicating whether the unit’s geometry includes internationally disputed territory (including the China–India border areas, Diaoyu Islands, Taiwan and units related to the South China Sea). Filter disputed = false to obtain a non-disputed subset.
Text note describing the cartographic convention. Under the unified statement, disputed units describe the Chinese national cartographic standard; this field is empty when disputed = false.

Country packages and the global basemap use different conventions. data/countries/*.zip preserves each country’s official representation (for example, CTAmap for China, official Indian sources for India, and official Japanese sources for Japan). The global GeoJSON and PMTiles apply the convention described above. State clearly which representation you use.

Research use and disclaimer. Boundaries are cartographic representations, not determinations of sovereignty. This dataset is for research and education, implies no project position on disputes, and does not constitute legal recognition or denial of any boundary.

Privacy policy

Feedback data privacy

This site is a purely static website with no backend user database. Here is how your privacy is protected when you use the feedback form and browse the site:

The feedback type, the country/level/year involved, the problem description, the evidence link and the optional email you enter in the feedback form.
GitHub Issue submissions: the content is publicly visible on the GitHub Issue page and may be indexed by search engines.
Online form submissions: forwarded by the form service to the maintainers’ mailbox and used only for data corrections.
Local storage: your submissions are kept in your browser’s localStorage; we cannot read or delete them remotely, and you can export or clear them on the feedback page at any time.
Please do not send sensitive personal data: avoid national ID numbers, home addresses, phone numbers or trade secrets in the description or attachments.