Abstract Rock abundance, defined as the surface's fractional area covered by rocks, is used to characterize the Moon's regolith, decipher its impact history, and assess potential landing sites. By definition, it should be agnostic to rock‐size distributions. However, it has been suggested that rock abundances, derived from surface temperatures measured by the Lunar Reconnaissance Orbiter (LRO) Diviner radiometer, are not. Systematic analyses of the effect of rock‐size distributions on Diviner‐derived rock abundances have been hindered by the laborious nature of manually mapping individual boulders from optical images. Here, we conduct such an analysis from automated boulder detections in high‐resolution LRO Narrow Angle Camera images using BoulderNet, a boulder segmentation machine‐learning model. We present a comparison of BoulderNet‐ and Diviner‐derived rock abundances and find that, although both data sets are strongly correlated, rock abundances derived from thermophysical modeling are underestimated in areas with a greater proportion of smaller rocks relative to larger ones.
Amaro et al. (Sun,) studied this question.