Nature·10 June 2026
Deep learning four decades of human migration
Researchers built the first year by year picture of how many people moved between every pair of countries from 1990 to today, covering 230 countries and regions. Migration data is usually patchy, years apart, and defined differently everywhere, so they trained an ensemble of neural networks on official statistics, censuses and past estimates to produce consistent annual flows with error bars. The model beat existing five year estimates on data it had never seen, and all data, code and models are public.
A globally consistent, year by year map of human migration across 230 countries from 1990 to the present, reconstructed with deep learning.
- The number
- 230 countries and regions, annual resolution, 1990 to present.
- The caveat
- These are model estimates, not head counts. The authors flag the regions where uncertainty stays high and real data is still missing, and the work does not explain why people move.
Verified DOI resolves live, no retraction on record. Read the source