Tracking Welfare from Space: Longitudinal Evidence from Rural Ethiopia

Satellite imagery and machine learning can predict local poverty with considerable cross-sectional accuracy, but whether these methods can track welfare over time remains largely untested. We provide a comprehensive first test using a true household panel: 1,787 rural Ethiopian households observed with GPS coordinates across three survey rounds (2021, 2023, 2025), combined with time-varying geospatial features. The outcome is consumption expenditure per adult equivalent. Cross-sectional models achieve cross-validated Spearman rank correlations of 0.53–0.67 across rounds.

Available at: https://doi.org/10.55158/DEEPWP38