AI for Good

Chapter 30Mapping Population Movement Using Satellite Imagery

—Tammy Glazer, Gilles Hacheme, Amy Michaels, and Christopher J.L. Murray

Executive Summary

Population estimates are imperative for understanding the potential impact of interventions designed to address sustainability, humanitarian action, and health issues. However, population estimates are just that—estimates, ones often challenged by lack of recent and accurate census data, changing fertility rates, and migration patterns.

Here, we used a multi-step process that combined satellite data (providing information on building density), census data, and household surveys to estimate the number of people per building, as well as deep learning modeling techniques with model-based geostatistics to estimate the number of people living in the Sahel region of Africa, an area experiencing extreme effects of climate change and humanitarian crises.

While our findings are preliminary, we found our population estimates to be accurate and to be replicable in other regions. We also were able to combine the human structure maps we created with environmental data to predict risk and migration in Kenya, finding that humans in Kenya are moving away from areas of heat, rainfall, and days of extreme heat and that ecological niche is highly predictive of malnutrition there.

The methods that we developed show great promise. We learned that collaboration across institutions and the ability to leverage the expertise and data that each institution ...