Deep learning four decades of human migration

A global annual migration-flow dataset spanning 1990 to 2024 has been produced using deep-learning models and diverse data sources. This dataset offers estimates of human movement across 230 countries, featuring enhanced temporal resolution, broader coverage, and more refined uncertainty estimates compared to previous datasets. The research, published in Nature on June 10, 2026, leverages advanced AI techniques to reconstruct migration patterns over four decades. The methodology integrates various data streams, including census data, surveys, and digital traces, to provide a comprehensive view of international and internal migration. This initiative aims to equip policymakers and researchers with more accurate and granular data to understand and address the complexities of human mobility in the 21st century. The improved dataset is expected to facilitate more effective planning for infrastructure, social services, and humanitarian aid in regions experiencing significant population shifts. The study highlights the potential of deep learning to tackle large-scale data integration and inference challenges in social sciences.