This $299,574 federal Project Grant awarded by the National Science Foundation's Computer and Information Science and Engineering (CISE) program (CFDA 47.070) aims to develop a comprehensive cyberinfrastructure solution for training large-scale Graph Neural Networks (GNNs) to support spatiotemporal prediction and modeling of geographically distributed and heterogeneous data. The project led by Emory University will address key research challenges in formulating spatiotemporal prediction within a geographically inspired graph deep learning framework, enabling efficient and accurate spatiotemporal prediction across vast distributed datasets, and integrating spatial correlation, heterogeneity, computing parallelism, and geographic communication. The award supports the creation of centralized and decentralized spatiotemporal graph learning infrastructure to leverage multiple edge micro-datacenters for collaborative GNN model learning and address spatial heterogeneity through decentralized geographical multitask learning. This project seeks to provide a cyberinfrastructure solution to overcome computational and communication bottlenecks for a broad range of domain science applications relying on large-scale spatiotemporal prediction.