Improving Childhood Respiratory Infection Risk Mapping with AlphaEarth Embeddings

Childhood acute respiratory infection (ARI) remains a leading cause of preventable child death in low- and middle-income countries, with risk strongly shaped by air quality and climate. Yet surveillance is often weakest where burden is highest, limiting reliable risk mapping. Geospatial foundation models offer an alternative by encoding landscape structure from global Earth-observation data. We test whether AlphaEarth Foundations embeddings improve cluster-level ARI prediction beyond gaseous pollutants using Demographic and Health Surveys data from 11 countries. Under pooled cross-validation, adding embeddings increases test $R^2$ from 0.098 to 0.164. However, gains depend on sample size: embeddings improve prediction in data-rich countries but overfit where survey clusters are sparse. These results show the promise—and data requirements—of geospatial foundation models for scalable respiratory-risk mapping.