Comparative planetology – Comparing cirques on Mars and Earth using a CNN

Abstract

New approaches for comparative planetology are vital to recognize and assess possible shared formative processes that shape landforms on different planetary bodies, and in the case of this study to identify possible cirques preserved from past martian climates that may have been more favorable to equatorial ice. We used “RetinaNet”, an augmented Convolutional Neural Network (CNN) initially trained with digital elevation data from over 1950 terrestrial cirques, to identify similar forms along a 50x50km area of the equatorial martian Dichotomy Boundary east of Gale Crater. A significant number of false positive detections associated with CTX stereo pair derived DEM artifacts and tile edges as well as craters were automatically removed. High confidence levels assigned by the machine learning algorithm suggest that the 16-remaining cirque-like forms may have formed and evolved under similar conditions to their terrestrial analogs. Morphometric data extracted for each of these cirque-like forms was statistically compared to terrestrial analog morphometric measures. While cirque length and width were significantly larger than their terrestrial counterparts and the area of these forms was ~150 % larger, on average headwall heights were ~25 % shorter. Allometric relationships suggest that the shape of these martian features changes proportionally as they grow in contrast to terrestrial forms where length increases faster than width and height. This could be due to the weaker gravitational force on Mars (3.721 m/s2) and other effects related to martian climate and environmental conditions, such as cold based glaciation. The identification of cirques in an equatorial region has implications for cryosphere distribution/redistribution and martian geologic and climatic history.