Open population datasets and open challenges

This post details the creation and public release of high-resolution population datasets derived from satellite imagery and computer vision techniques. It describes the technical approach of using convolutional neural networks to identify buildings globally, the trade-offs made for scalability, and the collaboration with Columbia University and the World Bank for population estimation and validation. It also presents an analysis of population aggregation near urban areas using clustering algorithms and discusses the implications for designing communication technologies that can bridge distances up to 63 km to connect 99% of the population in analyzed countries. The post highlights the use of these datasets for humanitarian aid and disaster response planning.