Predicting poverty using indicators from satellite imagery - An application to Antananarivo (Madagascar)

This paper contributes to the empirical literature on ne scale estimation of poverty using open-access satellite imagery in urban settings. Our main hypothesis is that urban housing morphology reects socioeconomic status, as households within similar physical environments share comparable characteristics. We rely on census data of Madagascar from 2018 that registered a total population of 1.27 million inhabitants living in 323 297 households distributed in 192 neighborhoods of the city of Antananarivo, to assess the performance of indicators produced from satellite imagery during the period 2014-2021 using spatial econometrics. We show that the building detection confidence metric provided by Google Open Buildings, combined with other readily available indicators of satellite imagery, can explain 74% of the variations of poverty at the neighborhood level. This approach demonstrates that open access satellite-derived indicators can serve as alternative source of data that bypass complex machine learning methods thereby signicantly reducing both financial cost and the technical expertise required for implementation.