Introduction Urban heat represents one of the most critical and inequitable manifestations of climate change, with mounting impacts on human health, energy systems, and urban sustainability. Bridging the gap between observation and inference requires scalable approaches that enable all-weather, continuous urban heat mapping at decision-relevant resolutions. Methods This study introduces a multimodal Geospatial Artificial Intelligence (GeoAI) pipeline that fuses atmospheric reanalysis and Earth observation to generate hourly, super-resolved land-surface temperature (LST) estimates for urban heat and health-risk assessment. The pipeline integrates three complementary foundation models—Prithvi-WxC, Prithvi-EO, and Granite-LST—to capture interactions between atmospheric dynamics and surface morphology. The system is implemented over Indianapolis, Indiana. Results The pipeline produces continuous temperature fields at 10–30 m resolution with sub-2 °C error, reproducing realistic diurnal heat-island dynamics across the Indianapolis study area. The fused model captures fine-scale thermal heterogeneity driven by impervious surface fraction, vegetation cover, and building morphology, resolving intra-urban temperature gradients that single-source products miss. Hourly temporal continuity enables characterization of heat exposure timing and duration, including nocturnal heat retention in historically underserved neighborhoods. Discussion Beyond technical performance, the framework demonstrates how foundation-model fusion can bridge environmental monitoring and health analytics, offering a scalable tool for exposure mapping, early-warning systems, and equitable climate adaptation. This work establishes a reproducible blueprint for AI-enabled urban climate twins, advancing the integration of environmental intelligence into public health resilience planning.
Daniel P. Johnson (Mon,) studied this question.