Towards an AI Cube for EO data inference in a distributed infrastructure

Geospatial data cubes show great promise to facilitate integrated and efficient analysis of big Earth Observation (EO) data. Existing geospatial data cube processing focuses more on traditional geoprocessing algorithms. It is still unknown how to enhance cube analytics with geospatial artificial intelligence (GeoAI) models. The paper presents an AI cube approach to improve the cube capabilities from geospatial data storage and processing to the cube inference. The approach includes the proposal of a GeoAI model warehouse using a cube organization, a matchmaker for model selection, and a cube inference pipeline. It is developed using the on-the-fly and batch modes of model inference in a cube infrastructure-based EO cloud computing platform, the Open Geospatial Engine (OGE). The implementation illustrates that the approach enhances traditional data cubes from physical models to GeoAI analysis and contributes to the development of an AI-ready Spatial Data Infrastructure (SDI). The on-demand model matchmaker and parallel inference improve the performance of GeoAI inference by achieving the best accuracy from available models and reducing the inference time by over 80%.