Mapping Aeolian Landforms and Uncovering Multi-Generational Wind Imprints on Mars Using Machine Learning

This study presents a large-scale mapping framework for identifying and characterizing periodic decameter aeolian landforms on Mars using CTX mosaics and machine learning. To test whether the mapped features align with the current aeolian regime, we mapped barchan slipfaces and used Global Climate Model (GCM) outputs to characterize present-day aeolian activity across an extended region, encompassing the Thaumasia and Bosporos planae as well as the lowlands located in the northwestern edge of Noachis Terra. Morphology and data clustering allowed us to define eight regional units, presenting distinct sets of aeolian landforms. The spatial distribution of these units is influenced by long-wavelength topography, with the Thaumasia highlands and Coprates rise forming a major divide. The bedforms in Thaumasia and Bosporos planae are morphologically simpler and possibly younger. Yet, their trend deviates by 35° from the orientation of GCM-predicted bedforms forming under current surface conditions. In addition, the units located in the eastern lowlands display complex morphologies and paleo-wind streaks indicative of multi-phase aeolian imprints. These features imply an ancient SE-NW paleo-wind regime, opposing the current southeastward transport trend. These major discrepancies may reflect several reorganizations of Mars’ atmospheric circulation in the past. Our systematic mapping results underscore the richness and complexity of Mars’ aeolian record, revealing extensive and continuous patches of likely ancient wind-shaped landforms. Our results suggest that a considerable number of multi-generational aeolian imprints (i.e. wind-formed features created over multiple periods) may have been preserved on the surface of Mars, under the form of decameter-scale aeolian landforms.