The Moon's surface is covered in craters that make landing hazardous, and mapping them requires processing the millions of images NASA's Lunar Reconnaissance Orbiter has captured since 2009. According to ESA, an AI model it developed marked 3,192 potential craters, ranging from roughly 6 to 180 metres across, in a 1.25 by 1.27 km patch of lunar terrain near 44.12°N, 307.80°E.

Because no public high-quality database of small lunar craters exists, ESA turned to players of its IMPACT game, who mark craters on real satellite images in a minigame. Patrick Bauer, a PhD student working on AI-based crater detection using IMPACT data, said each crater marked by a player is "a small but important contribution to mapping the Moon." Player-labelled data formed a training set of 672 tiles, each 512 by 512 pixels.

ESA says the results are promising but the model is still being developed and its detections are not perfect, with degraded craters that have eroded rims especially difficult to detect. Future research will examine whether digital elevation models could improve detection. The featured tile is an extract from image M1219279265LC, captured by the Lunar Reconnaissance Orbiter's narrow angle camera and credited to NASA/GSFC/Arizona State University and ESA.