Leveraging Compact Satellite Embeddings and Graph Neural Networks for Large-Scale Poverty Mapping

2025-11-03

AI Summary

High-resolution, fine-grained poverty mapping remains scarce across the Global South. While Demographic and Health Surveys (DHS) provide high-quality ground truth, their sparse spatial coverage and deliberate coordinate perturbation—introduced for privacy protection—severely hinder model generalization. Method: We propose an end-to-end framework integrating low-dimensional satellite embeddings (AlphaEarth), a graph neural network (GNN), and a Gaussian kernel-based fuzzy label loss. It constructs a spatial graph over survey points to explicitly model geographic dependencies and probabilistically mitigates the adverse impact of coordinate displacement on supervision signals. Contribution/Results: Evaluated on 37 DHS datasets across sub-Saharan Africa, our method significantly outperforms image-only baselines. It achieves marked improvements in zero-shot extrapolation to unlabeled regions, demonstrating that lightweight remote-sensing representations—combined with structured spatial modeling—enable scalable, robust socioeconomic prediction at continental scale.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Computer Vision: Remote Sensing / Geospatial AI
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

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.

Problem

• Mapping poverty in Global South using satellite embeddings
• Addressing limited spatial coverage of survey data
• Improving wealth prediction accuracy with graph networks

Innovation

• Uses graph neural networks for spatial poverty mapping
• Leverages compact satellite embeddings for wealth prediction
• Introduces fuzzy label loss for coordinate displacement handling