Abstract: The data-driven research paradigm brings a strong demand for training data sharing in geospatial artificial intelligence (GeoAI). The training data content and organization from different GeoAI applications are diverse. A unified information model will lay the foundation for GeoAI training data sharing and interoperability. By analyzing the common features and core attributes of different GeoAI training data, an information model for training data was proposed, and the training data quality elements and evaluation methods were explored. The results provide a reference for development of GeoAI training data stores and sharing services.