IBM and NASA Release Open-Source Lunar AI Model
IBM and NASA have released an open-source artificial intelligence model to help researchers analyze lunar observations. The model, part of the Prithvi family of open foundation models for geospatial, weather, and other scientific applications, was trained on over 30 layers of data collected by nine instruments on four NASA missions.
The model is designed to perform tasks such as crater mapping, volcanic-feature analysis, and identification of areas that may contain ice. It brings together observations at different resolutions and under varying lighting conditions, giving scientists a reusable starting point for lunar remote-sensing work.
Initial tests showed the model's effectiveness in detecting ice prospectivity near the poles, crater detection, and mapping irregular mare patches - volcanic features that help study the Moon's geological history. The team reported a 22% reduction in error compared to a baseline SwinV2 transformer for ice-prospectivity benchmark.
The release is part of IBM and NASA's efforts to make open-source models available for research purposes, not for mission certification. Researchers can adapt the model using the released checkpoint and code, with the possibility of broadening lunar research in areas such as potential resource utilization and surface history studies.