On paper, Arcelia looks like a typical, moderately poor Mexican town. Located in Guerrero — one of Mexico’s poorest states — it carried an official Human Development Index (HDI) score of 0.714, placing it in the UN’s “high development” band. When researchers used satellite imagery and artificial intelligence to estimate local development, the picture changed: the model gave Arcelia an HDI of 0.617, which falls into the “medium development” category. That shift matters for roughly 33,000 people.
That contrast is not an isolated anomaly. A 2026 study by researchers at Stanford University and the UN Development Programme published in Nature Communications found that more than half of the world’s population (about 58%) is assigned the wrong development tier when national averages are used. Much of the problem comes from relying on infrequent or outdated household surveys and censuses: roughly half of the world’s poorest countries haven’t had a census in the past decade, making current, local-level measurement difficult.
Why local HDI scores matter
The Human Development Index is more than a ranking. By combining income, education and life expectancy into a single 0–1 scale, the HDI influences how resources are allocated internationally and which regions get prioritized for aid. But HDI was designed as a national indicator; averaging across an entire country hides local disparities. When aid or policy targets are set using only national scores, resources can miss pockets of need.
Earlier efforts to add resolution produced the Subnational Human Development Index (SHDI), which mapped HDI at the province level for 1,739 provinces in 159 countries. That was a step forward: Mexico, for example, went from one national score to 32 provincial scores. But even province-level averages can obscure variation inside provinces. Guerrero, for instance, contains dozens of municipalities with very different realities.
How satellites and AI add detail
The Stanford team trained a machine-learning model on satellite images aligned with known provincial HDI scores, letting the algorithm learn which visual features correlate with development. Patterns such as road density, building morphology and nighttime lights emerged as useful proxies for income and education; health outcomes were harder to infer from space.
Applying the model at the municipal level in Mexico produced a much more varied map than province-level averages. Areas that looked uniformly “higher development” by provincial data broke into a patchwork of differing scores when seen at municipal resolution. In a simulated aid program aimed at the poorest 10% of Mexico’s population, incorporating municipal-level estimates improved the researchers’ ability to identify target populations by more than 11 percentage points.
Limitations and what satellites cannot see
Despite these gains, the study and outside experts caution that satellites do not tell the whole story. The model explained only about 29% of within-province HDI variation in Mexico, and vital components of human development — such as malnutrition, certain health conditions, and nuances of education quality — are poorly captured from space. As Sabina Alkire, director of the Oxford Poverty and Human Development Initiative, notes, an undernourished child cannot be detected from nightlights.
Satellites are therefore best understood as a complementary tool. They can fill information gaps where surveys are infrequent, costly or slow, and they can flag local hotspots that merit ground verification and targeted data collection. But they are unlikely to replace household surveys and censuses that directly measure income, health and education.
Conclusion
The study demonstrates that satellite imagery plus machine learning can add a valuable new layer to global poverty measurement, revealing local development differences that national averages obscure. For policymakers and aid agencies, that finer resolution can mean better-targeted interventions. Yet the technology has limits: it captures visible infrastructure and activity well but misses many aspects of human well-being that only ground-level data can reveal. The most reliable approach will combine satellite-derived estimates with timely, on-the-ground surveys to ensure resources reach the people who need them most.