Our method uses machine-vision (AI) techniques to process daylight and night-time satellite imagery. Using additional features and proprietary data, we are able to both now-cast and forecast poverty shares through 2030.
Working under the auspices of the Asian Development Bank and in collaboration with local National Statistical Offices, we used machine learning to develop highly granular poverty models and maps of Thailand and the Philippines. These models reveal the incidence of poverty and extreme poverty at the village level throughout each country.
In partnership with DANE (National Administrative Department of Statistics of Colombia), the Government of Colombia, and PARIS21, we are creating granular poverty estimates of Colombia at the 4 km² level so that decision-makers can targets regional areas where poverty rates are still high. We used daytime satellite images to predict the share of the population that is poor using a machine learning algorithm for computer vision to detect features relevant for poverty detection.
Working in collaboration with the International Fund for Agricultural Development (IFAD), the Government of Niger and the Organisation International de la Fracophonie, we are piloting the development of a village level poverty model for Niger. Data from the project will be used to inform the further development of IFAD’s forthcoming 2020 Sahel strategy.