Seeing the informal economy from above
Aerial imagery is rising steeply in the AI journals, and economists already use night-time lights to measure growth. Yet none of the papers on the informal economy uses either. Can image models measure informal activity that surveys miss?
What prompted this#
In the AI dashboard trends, "aerial imagery" had 9 papers in July–September 2026 against about 1 expected from its earlier share. In economics, night-time lights are an established proxy for activity: 26 papers carry the keyword since October 2025 (24 in economics journals), none of them with a machine-learning keyword. And of the 55 papers with an informal-economy, informal-employment or informal-sector keyword, none carries a satellite, night-light, aerial or remote-sensing keyword.
Questions#
- Can models trained on high-resolution imagery (market stalls, rooftops, vehicles, building materials) predict informal employment where surveys exist, and fill in where they do not?
- Do such measures change growth or inequality estimates in countries with large informal sectors?
- Do the models carry over from one country to another? Cross-dataset generalization is another rising AI keyword.
Data that could help#
- Free Sentinel-2 and Landsat imagery, VIIRS night-time lights, and commercial imagery for a few cities.
- Geocoded household surveys (LSMS, DHS) as ground truth; both are in the dataset ranking's list.
Next step#
Pick one country with geocoded survey clusters and test whether image features add to night-time lights in predicting informal employment shares.