Hollows on Mercury: Creation and Analysis of a Global Reference Catalog With Deep Learning

AbstractHollows are geologically young depressions on Mercury, most likely associated with the loss of volatile species. The distribution and morphometric properties of hollows provide information about the overall volatile budget of Mercury's (shallow) subsurface, with significant implications for our understanding of the evolution of Mercury and airless planetary bodies in general. Here, we use a convolutional neural network to map the global geographic distribution and morphometric properties of hollows in MESSENGER orbital images and assess their geostatistical relationships with the thermophysical environment. We identify up to 19,110 hollows in the MESSENGER MDIS Narrow‐Angle Camera data set and discover previously unidentified hollows in more than twenty large‐scale geographic regions. Globally, the detected hollows are predominantly located in the northern hemisphere, where MESSENGER image coverage and spatial resolution are highest. Hollows are preferentially detected in impact craters, at low elevations, on low slope angles, and cluster toward the maxima of ejecta mass production by micrometeoroid bombardment. We observe that hollows tend to be located on equator‐facing s