Materal and Child Health Monitoring Using Satellite Imagery and Open Data in LMICs

Abstract

Traditional data on maternal and child health (MCH) relies on nationally representative household surveys, which are often costly and infrequently conducted in low and middle income countries. Our project explores the use of machine learning with satellite imagery and other remotely sensed variables to estimate these key MCH indicators, such as women’s mean BMI, skilled birth attendant rate, under 5 child mortality rate, and more. We integrate this MCH information from Demographic and Health Surveys (DHS) with geospatial information from Google Earth Engine and imagery from Landsat and Sentinel satellites. In order to predict MCH information, we build complementary pipelines: a tabular pipeline that combines DHS data with geospatial features from Google Earth Engine, and a satellite pipeline that applies convolutional neural networks and vision transformers to raw Landsat imagery. Beyond building accurate models, our goal is to evaluate where and when these approaches generalize best across countries, years, and local contexts. By assessing both performance and limitations, this work advances the potential of satellite-informed prediction as a tool for global health monitoring and equity.

More results/details to be shared soon!