Engstrom, RyanHersh, JonathanNewhouse, David2024-01-172024-01-172021-07-31The World Bank Economic Review0258-6770 (print)1564-698X (online)https://openknowledge.worldbank.org/handle/10986/40907Can features extracted from high spatial resolution satellite imagery accurately estimate poverty and economic well-being The present study investigates this question by extracting both object and texture features from satellite images of Sri Lanka. These features are used to estimate poverty rates and average expected log consumption taken from small-area estimates derived from census data, for 1,291 administrative units. Features extracted include the number and density of buildings, the prevalence of building shadows (proxying building height), the number of cars, length of roads, type of agriculture, roof material, and several texture and spectral features. A linear regression model explains between 49 and 61 percent of the variation in average expected log consumption, and between 37 and 62 percent for poverty rates. Estimates remain accurate throughout the consumption distribution, and when extrapolating predictions into adjacent areas, although performance falls when using fewer households to calculate estimates of poverty and welfare.en-USCC BY-NC-ND 3.0 IGOPOVERTY ESTIMATIONSATELLITE IMAGERYMACHINE LEARNINGBIG DATAINEQUALITYJournal ArticleWorld Bank10.1596/40907