EGU22-4026
https://doi.org/10.5194/egusphere-egu22-4026
EGU General Assembly 2022
© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Globally Tracking Dust Devil Vortices on Mars Using Neural Networks

Susan J. Conway1, Valentin T. Bickel2, Manish R. Patel3, Lori Fenton4, and Helen Carson5
Susan J. Conway et al.
  • 1Nantes Université, Université d’Angers, Université du Mans, CNRS UMR 6112 Laboratoire de Planétologie et Géosciences, France (susan.conway@univ-nantes.fr)
  • 2ETH Zurich, Zurich, Switzerland
  • 3School of Physical Sciences, Open University, Milton Keynes, UK
  • 4Carl Sagan Center at the SETI Institute, Mountain View, CA, USA
  • 5Department of Materials Science & Engineering, University of Washington, Seattle, WA, USA

Dust devils are atmospheric vortices driven by daytime dry convective circulations and are visible because of the dust entrained from the ground. They are common in deserts on Earth and globally on Mars. They appear as tubular or conical light-coloured clouds of dust that cast a dark shadow which is particularly distinctive in orbital images. They can reach much larger sizes on Mars (several km in height), compared to Earth, perhaps because their size is limited by the depth of planetary boundary layer. Here, we perform a global survey for dust devil vortices by using a neural network to search through the database of Context Camera (CTX) images aboard NASA’s Mars Reconnaissance Orbiter spanning Mars Years 28-35.

We used an off-the-shelf convolutional neural network (CNN) architecture (RetinaNet) as used successfully for previous planetary studies. After training and testing (average precision AP ~0.7) we processed the whole database of CTX images (n=111,547 images) for dust devil detections using the JMARS-served CTX images. Every detection with a CNN confidence level (CT) greater than 0.5 (n=57,051) was verified by a human operator. The effective diameter of the dust devil was estimated from the bounding box size by measuring the diameter of the “cloud” in a sample of 33 dust devils to generate a linear scaling relationship.

3,747 images were found that contained validated dust devils at CT >0.5, comprising 11,201 individual detections. The images spanned MY 28 starting at Ls 114° through to MY 35 at Ls 114°. Trends in frequency and size of dust devils with season agree with previous studies, where higher densities and larger sizes of dust devils are found in local summer for each hemisphere and low levels of activity occur in local winter. Valles Marineris and Elysium Planitia (InSight, MSL) are notable areas lacking dust devils despite good temporal image coverage. We confirm the hotspots of Chryse and Hellas Planitiae noted in some, but not all previous studies. We find two notable hemispherical asymmetries in the data: (a) The peak in size/density occurs just after the solstice in the southern hemisphere, but at the solstice in the northern hemisphere. (b) Excluding known hotspots in Amazonis and Arcadia Planitiae we find that two broad latitudinal zones seem to exhibit both higher frequency and size: 55-70°N at Ls 120-150° and 50-70°S at Ls 300-330°, agreeing with observations of dust devil tracks. We attribute the hemispherical asymmetries to the dominance of the southern summer Hadley circulation and are investigating this further using data from the OpenMARS climate database.

Acknowledgments: we thank the JMars team at ASU for hosting map projected CTX image products used in this work. SJC acknowledges the French Space Agency CNES for supporting her Mars work.

How to cite: Conway, S. J., Bickel, V. T., Patel, M. R., Fenton, L., and Carson, H.: Globally Tracking Dust Devil Vortices on Mars Using Neural Networks, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-4026, https://doi.org/10.5194/egusphere-egu22-4026, 2022.