NASA and IBM's 17-year lunar model goes open source

nasa and ibm research have developed an open-source lunar foundation model that utilizes 17 years of observation data from the lunar reconnaissance orbiter. the model, trained on nearly 2 million tile bundles across 11 modalities, can predict ice deposits at the poles and detect craters. it was trained from scratch using sombench, a large co-registered multimodal lunar corpus. the model provides a foundation for lunar science, making decades of lunar observation data usable for machine learning.