World Migration Flows

An open-source, interactive map of global migration — an homage to (and rebuild of) Max Galka's Metrocosm global immigration map, whose original site is no longer online.

Live map: https://code-4you.github.io/world-migration-flows/

Animated particles show estimated migration flows between countries. An Events dropdown replays named migration episodes — the Syrian refugee crisis, the Gastarbeiter era, the fall of the Soviet Union and more — each auto-playing its years with the affected countries highlighted, and each shareable via ?e=<event> links (e.g. ?e=ukraine). Its play button tours every event in sequence. Circles show each country's immigration (blue) and emigration (red) as nested discs — the larger flow outside, the smaller on top, half-and-half when balanced. Hover for numbers, click any country (or use the dropdown) to isolate its flows, and play through decades or single years, 1960–2024.

Data

• 1990–2020: Abel & Cohen bilateral migration flow estimates (da_pb_closed method) — true gross flow estimates from demographic accounting, so both directions of every corridor are visible (e.g. Mexico→US and US→Mexico return migration).

• Country names and positions: UN DESA, International Migrant Stock 2024 defines the country set for countries.json; centroids from world-countries-centroids with manual fixes for bent shapes (Norway, Croatia, Vietnam). The 2020–2024 period file comes from the Gaskin & Abel yearly estimates (2020–2023 summed).

• 1960–1990 decades and single years 1960–1989: UNU-CRIS imputed bilateral migration dataset (Standaert & Rayp 2022), stock differences over the decade or year.

• Single years 1990–2023 and the 2020–2024 period: Gaskin & Abel deep-learning estimates ("Deep learning four decades of human migration", GPL-3.0) — annual bilateral flows by previous residence for ~230 countries, from a neural network integrating UN census stocks, national statistics, QuantMig, and Meta's Facebook-based flow measurements (Chi et al., PNAS 2025). Unlike the raw Meta data, China and Iran are covered. The "Year…" dropdown and its play button (2s per year) step through every year from 1960→1961 to 2023→2024.

• 2024 (selected countries only): officially reported statistics — Eurostat immigration/emigration by partner country (calendar 2024, ~19 reporting European countries) and ABS Overseas Migration 2024–25 arrivals/departures by country of birth (Australia). Because coverage is partial, 2024 appears in the Year dropdown only while a covered country is selected; global views stay capped at the newest complete year. Only both-direction sources are used (inflow-only sources like green-card counts are excluded). Regenerate with scripts/process_reported.py.

• All other numbers are model-based estimates, not counts; the sources use different pipelines, so expect methodological seams at 1990 and 2020. For the same reason the decade files (Abel & Cohen) and the single-year files (Gaskin & Abel) are independent series: a decade's total won't exactly match the sum of its ten years. The most visible one: emigration from rich countries is under-counted before 1990. The pre-1990 stock-difference method only registers an outflow when a country's expatriate stock grows, and deaths of older emigrants usually cancel new departures — e.g. US emigration reads as ~30k/year through the 1980s, then jumps to ~500k/year from 1990 when the flow-based estimates take over. Emigration didn't start in 1990; it just becomes visible then. (1990 itself, the first Gaskin & Abel year, runs hot — treat that single year's magnitudes with extra caution.)

  The seam is also a definition change: before 1990 migrants are tracked by country of birth; from 1990 by country of previous residence. An Egyptian guest worker leaving Saudi Arabia is "an Egyptian" to the old method — his departure never counts as Saudi emigration — but counts as Saudi Arabia→Egypt under the new one. That's why guest-worker states (Saudi Arabia, the Gulf) show almost no emigration before 1990 and large outflows to Egypt, India and Indonesia from 1990: the same worker rotations existed all along, only the newer method can see them. It's also what makes return migration (e.g. US→Mexico) visible from 1990 on.

Regenerate the data files with:

python scripts/process.py          # countries.json only (names + centroids)
python scripts/process_abel.py     # 1990-2020 decades (Abel & Cohen flows)
python scripts/process_early.py    # 1960-1990 decades (UNU-CRIS stocks)
python scripts/process_yearly.py   # single years 1960-1989 (UNU-CRIS stocks)
python scripts/process_gaskin.py   # single years 1990-2023 + 2020-2024 period
python scripts/process_reported.py # reported 2024 (Eurostat + ABS live APIs)
python scripts/process_total.py    # 1960-2024 all-decades view (run last)

(after downloading the source files into raw/ — see the script docstrings). Every flows_*.json has exactly one owning script, so rerunning any single script can't overwrite another source's files.

How migration is measured (and how much to trust it)

Nobody keeps a worldwide ledger of moves. Migration numbers are built from imperfect ingredients, and knowing them explains everything odd about the map:

• Censuses are photos, not videos. A census shows who lives where on one day ("born in Nigeria, living in the UK"). Comparing two photos ten years apart hints at migration — but two photos cannot tell "nothing happened" apart from "one person arrived and one person died." Deaths of emigrants abroad silently cancel real migration in the count.
• That's the pre-1990 weakness. Our 1960–1990 numbers come from such photo-comparisons: the direction and relative size of every flow is right, but magnitudes run low (deaths eat part of them), and emigration from countries with old diasporas is understated most.
• Post-1990 data corrects for deaths. The modern estimates (Abel & Cohen; Gaskin & Abel) first compute how many people should have died in each migrant group using mortality tables, and only call the unexplained remainder "migration". Same photos, right answer.
• Poor-quality censuses are patched by mirror data. A Nigerian migrant doesn't need Nigeria's census to be counted — the UK, US or South African census records them as Nigerian-born. Every migrant only has to be counted somewhere. Refugee flows are covered by UNHCR registration, which works even in poor regions.
• The resulting quality gradient: flows touching rich countries are the most reliable (mirror-counted), refugee crises are well covered (registered), and migration between two weak-data countries (e.g. intra-African labor movement) leans most on modeling.
• The counting is asymmetric, and it skews the picture. Rich countries count arrivals well, so Africa→Europe is among the best-measured data on Earth — while flows into Africa and between African countries depend on weak local statistics and run understated. In reality, most African migrants move to other African countries; the popular image of African migration as "everyone heading to Europe" is partly a measurement artifact — the well-lit corridor looks like the whole story. The modelers (UN, Gaskin & Abel) know this and impute the dark corridors — many intra-African numbers on this map are model estimates rather than counts — but imputation narrows the gap rather than closing it. When a corridor connects two weak-statistics countries, treat its number as a sketch, not a figure.
• Tooltips show flows, consistently. "Net" always equals the displayed In − Out. (Earlier versions showed a raw stock-based net in which deaths abroad masqueraded as reverse migration — a bug caught by a Reddit commenter and fixed.)

Running locally

Any static file server works:

python -m http.server 8000

then open http://localhost:8000/. (Opening index.html directly won't work — the app fetches its JSON data over HTTP.)

Hosting

The app is fully static (HTML + CSS + JS + JSON), so it runs as-is on GitHub Pages, Netlify, or any static host. Note: Tableau Public cannot host custom web apps like this one — it only hosts workbooks built in Tableau's own tools.

Roadmap

• 2020–2024 flows (latest UN estimates)
• 2010–2020, 2000–2010, 1990–2000 options
• 1960–1990 periods via the "Earlier…" dropdown (UNU-CRIS data)
• Click a country on the map to focus it; all-countries dropdown
• Play button cycling through all periods continuously
• Pre-1960 periods: no comparable global bilateral dataset exists back to 1920 — extending further would be limited to a handful of countries with long census records
• Per-country play mode narration / camera movement

Who is abroad right now?

Worldwide, migrants are roughly 90% of everyone who is "somewhere else" right now — about 304 million people live outside their birth country (UN, 2024), versus a few tens of millions abroad on a temporary trip at any instant. (Visits outnumber migrations ~125:1 per year, but a trip lasts days while a migration lasts years, so the migrants accumulate.)

But not everywhere. In these places the foreigners actually present are mostly tourists, so migration isn't the full picture there (2019 snapshot; tourists-present estimated as inbound visits × ~5-day stay):

Country	Tourist share of foreigners present
Cuba	~95%
Andorra	~85%
Czechia	~84%
Tunisia	~82%
Slovakia	~82%
Montenegro	~79%
Vietnam	~77%
Austria	~75%
Jamaica	~72%
China	~62%

The pattern: states closed to immigration but open to tourism (Cuba, Vietnam, China), post-communist Central Europe, and tourism microstates.

The temporary side of the story has its own maps:

• World Travel Flows — every visit on Earth, 1995–2022, including the 2020 collapse
• Visits vs Migration — every corridor colored by how often visits become migration

Contributing

Contributions are welcome — the easiest way to start is opening an issue for a bug, a data problem, or an idea.

Some good first contributions:

• Add a migration event. Events are one-line entries in the EVENTS array at the top of app.js — an id, a display name, a year range (start years, within 1960–2023), and the ISO2 codes of the countries to focus. Please check the flows are actually visible in the data for those years before submitting.
• Improve data. If you know a better public bilateral migration source (especially anything past 2023), open an issue with a link. Each source has its own script under scripts/ that writes plain JSON into data/, so new sources slot in without touching the app.
• Fix or polish the app. It's dependency-free vanilla JS (MapLibre GL from a CDN, two canvas overlays) — no build step.

To run locally: clone, then serve the folder over HTTP (e.g. python -m http.server 8000) and open http://localhost:8000/ — opening index.html directly won't work because the app fetches its JSON data. The data/ files are committed, so you only need Python and the raw source downloads (see the docstrings in scripts/) if you want to regenerate data.

Code is MIT; by contributing you agree your contributions are too. The underlying datasets keep their own licenses (see Credits).

Credits

• Concept and original visualization: Max Galka (Metrocosm)
• Creator of this rebuild: Michael van Diermen
• Data (1990–2020): Abel, G.J. & Cohen, J.E. (2019), "Bilateral international migration flow estimates for 200 countries", Scientific Data (updated 2025)
• Data (2020–2024): United Nations, Department of Economic and Social Affairs, Population Division — International Migrant Stock 2024 (POP/DB/MIG/Stock/Rev.2024)
• Data (1960–1990): Standaert, S. & G. Rayp (2022), "Where Did They Come From, Where Did They Go? Bridging the Gaps in Migration Data", UNU-CRIS
• Data (years 1990–2023, period 2020–2024): Gaskin, T. & G.J. Abel (2025), "Deep learning four decades of human migration" (arXiv:2506.22821), GPL-3.0
• Data (model input we previously used directly): Chi, G. et al. (2025), "Measuring global migration flows using online data", PNAS — © Meta, CC BY 4.0, via HDX
• Basemap tiles: © OpenStreetMap contributors, © CARTO
• Map rendering: MapLibre GL JS
• Country centroids: world-countries-centroids; ISO code mapping: lukes/ISO-3166-Countries-with-Regional-Codes

Code is MIT licensed (see LICENSE). The UN data is © United Nations, made available for use under its terms.