Universities Space Research Association (USRA) contributed planetary science expertise to the newly released NASA-IBM Lunar Foundation Model, an open-source AI model designed to analyze diverse lunar datasets. Developed through collaboration led by NASA and IBM Research, the model was pretrained using Som Bench, a multimodal lunar dataset containing nearly two million co-registered data bundles spanning 11 modalities and two spatial scales. The model integrates imagery, topography, illumination geometry, thermophysical properties, mineralogy, radar, gravity, and other geologic data.

USRA's contribution came from Dr. Rachel Slank, associate scientist with USRA's Science and Technology Institute on assignment at NASA's Marshall Space Flight Center. She served as planetary science subject-matter expert, bridging lunar science priorities with model development decisions.

The model was evaluated across three benchmarks: crater detection at regional and meter scales, segmentation of irregular mare patches, and regression of lunar polar ice prospectivity. Across all benchmarks, the pretrained model matched or outperformed comparison models based on ImageNet pretraining. The study demonstrated strong label efficiency in crater detection, suggesting lunar pretraining reduces task-specific labeled data requirements.

By releasing the pretrained model, fine-tuning code, and benchmark datasets openly, the team aims to provide a reusable foundation for lunar research applications. Som Bench supports standardized evaluation of lunar machine learning applications.