Leveraging Compact Satellite Embeddings and Graph Neural Networks for Large-Scale Poverty Mapping
This work addresses the scarcity of poverty maps in the Global South, though it is incremental in improving generalization for socioeconomic mapping.
The paper tackled the problem of creating accurate, fine-grained poverty maps in Sub-Saharan Africa by using graph neural networks with satellite embeddings, resulting in slight accuracy improvements over image-only baselines on 37 DHS datasets.
Accurate, fine-grained poverty maps remain scarce across much of the Global South. While Demographic and Health Surveys (DHS) provide high-quality socioeconomic data, their spatial coverage is limited and reported coordinates are randomly displaced for privacy, further reducing their quality. We propose a graph-based approach leveraging low-dimensional AlphaEarth satellite embeddings to predict cluster-level wealth indices across Sub-Saharan Africa. By modeling spatial relations between surveyed and unlabeled locations, and by introducing a probabilistic "fuzzy label" loss to account for coordinate displacement, we improve the generalization of wealth predictions beyond existing surveys. Our experiments on 37 DHS datasets (2017-2023) show that incorporating graph structure slightly improves accuracy compared to "image-only" baselines, demonstrating the potential of compact EO embeddings for large-scale socioeconomic mapping.