Journal Article

Lunar Surface Model Age Derivation: Comparisons Between Automatic and Human Crater Counting Using LRO‐NAC and Kaguya TC Images

Fairweather, J. H., Lagain, A., Servis, K., Benedix, G. K.

2023

Overview

Dating young lunar surfaces, such as impact ejecta blankets and terrains associated with recent volcanic activities, provides critical information on the recent events that shaped the surface of the Moon. Model age derivation of young or small areas using a crater chronology is typically achieved through manual counting, which requires a lot of small impact craters to be tediously mapped. In this study, we present the use of a Crater Detection Algorithm (CDA) to extract crater populations on Lunar Reconnaissance Orbiter—Narrow Angle Camera (LRO‐NAC) and Kaguya Terrain Camera images. We applied our algorithm to images covering the ejecta blankets of four Copernican impact craters and across four young mare terrains, where manually derived model ages were already published. Across the eight areas, 10 model ages were derived. We assessed the reproducibility of our model using two populations for each site: (a) an unprocessed population and (b) a population adjusted to remove contaminations of secondary and buried craters. The results showed that unprocessed detections led to overestimating crater densities by 12%–48%, but “adjusted” populations produced consistent results within <20% of published values in 80% of cases. Regarding the discrepancies observed, we found no significant error in our detections that could explain the differences with crater densities manually measured. With careful processing, we conclude that a CDA can be used to determine model ages and crater densities for the Moon. We also emphasize that automated crater datasets need to be processed, interpreted and used carefully, in unity with geologic reasoning. The presented approach can offer a consistent and reproducible way to derive model ages. Plain Language Summary Studying young lunar surfaces, such as impacted areas or volcanic activity, helps us understand recent events that have shaped the Moon's surface. Determining the model age of these areas generally involves manually counting small craters, which is time‐consuming and variable. This study presents a machine‐learning approach to detect craters on images acquired by the Lunar Reconnaissance Orbiter‐Narrow Angle Camera and the Kaguya Terrain Camera. Four impact craters and four young mare terrains were analyzed, where model ages had already been determined manually. When comparing our automatic counts to the manual counts, we observed that our results became more consistent with the published surface ages when we excluded secondary or buried craters from our crater populations. We also outline that automatic crater detection methods can be used to determine the age of lunar surfaces in a reliable and consistent manner when used correctly. Key Points Automatic crater counting of the Moon was achieved using a Convolutional Neural Network architecture and applied to LRO‐NAC and Kaguya Terrain Camera images Testing of the automatic counts against manual counts across the same count areas is required to provide confidence in the results Surface ages resulting from automatic crater counts are within acceptable error of model ages for the same area found using manual counts

• Publisher: John Wiley & Sons, Inc,American Geophysical Union (AGU)
• Subject: Age / Algorithms / automatic crater counting / Cameras / Chang'E‐5 / Datasets / Geology / impact craters / Investigations / Moon / planetary chronology