NASA and IBM Launch Open Source Lunar Model Utilizing 17 Years of Data

NASA and IBM have released an open-source lunar model that integrates 17 years of data from the Lunar Reconnaissance Orbiter (LRO) and other missions. This model includes 2 million tile bundles, 1 million high-resolution images from the Narrow Angle Camera, and 964,000 multispectral images from the Wide Angle Camera, along with 30 spatially aligned data layers from various instruments. The collaboration began in early 2022 under a Space Act Agreement, culminating in the release of the first Prithvi model on Hugging Face in August 2023.

The model demonstrates a 22 percent reduction in prediction error for ice deposits and a 19 percent improvement in coarse-scale crater detection compared to the baseline SwinV2-B model. Additionally, it shows a 3 percent lead over SwinV2-B on Irregular Mare Patches. Juan Bernabé-Moreno, director of IBM Research Europe, noted that the model connects observations across instruments, revealing patterns that are difficult to spot in isolation.

NASA's chief science data officer, Kevin Murphy, emphasized the importance of this model in leveraging decades of lunar data for scientific advancement. The model is part of the AI for Science initiative, which aims to enhance scientific research through advanced AI techniques. However, Bernabé-Moreno cautioned that the model is not suited for absolute geodetic positioning. This follows previous developments in lunar exploration technology, including IBM's TerraMind model, which was developed in 2025 in collaboration with the European Space Agency and Forschungszentrum Jülich.