TY - JOUR
T1 - A global survey for dust devil vortices on mars using MRO context camera images enabled by neural networks
AU - Conway, Susan J.
AU - Bickel, Valentin T.
AU - Fenton, Lori K.
AU - Patel, Manish R.
AU - Carson, Helen C.
AU - Blouin, Antoine
AU - Crevier, Justin
AU - Blanc, Evan
AU - Nguyen, Bao Nhi
AU - Holmes, James A.
AU - Jackson, Brian
AU - Roelofs, Lonneke
N1 - Publisher Copyright:
© 2025 The Authors
PY - 2025/5
Y1 - 2025/5
N2 - Dust devils are atmospheric vortices that loft dust from the ground, typically at the hottest times of the day and year on Mars. They contribute dust to the atmosphere and so indirectly affect the global atmospheric circulation. Their size, shape and velocity can provide indications of the weather on Mars. Hence, tracking their occurrence in time and space provides useful data for understanding Mars’ current climate. They are also of relevance to landed missions as they can clean solar panel surfaces, extending mission lifetimes. Despite the numerous observations and surveys of dust devils to date, the global extent, distribution and occurrence of dust devils is not yet consistently constrained, mainly due to the relatively limited spatial and temporal scope of manual analyses. To provide the most comprehensive global catalogue of active martian dust devils to date, we applied a RetinaNet convolutional neural network to existing remote sensing images of Mars to identify the distinctive signature of the light-toned lofted dust cloud and dark shadow formed by active dust devils on Mars. The algorithm used ∼6 m/pixel Context Camera (CTX) images from Mars Years 28–36, scanning through a total of 132 359 images. False positives were manually removed with the help of the Zooniverse platform, resulting in 13 409 detections. This survey presents the most spatially and temporally exhaustive global catalogue of dust devils to date. We confirm many trends revealed in disparate previous studies. For example, approximately half of the detections are concentrated in the Amazonis Planitia monitoring site – a hotspot identified from previous imaging campaigns. In addition, we confirm that orbital observations are not well-suited for detecting dust devils at landing sites, despite the ubiquitous detection of vortices with in-situ data. Our study reveals previously understudied hotspots where dust devil lofted clouds can be seen from orbit, most notably southern Hellas Planitia where only dust devil tracks had been previously extensively reported. Importantly, our results reveal latitudinal clusters of dust devils, and in particular large dust devils, at around 60°N and 60°S during local summer solstice, which had only been hinted at by previous work. This concentration is at a much higher latitude than previous modelling suggests, indicating that dust devil generation on Mars is controlled by more factors than are currently accounted for.
AB - Dust devils are atmospheric vortices that loft dust from the ground, typically at the hottest times of the day and year on Mars. They contribute dust to the atmosphere and so indirectly affect the global atmospheric circulation. Their size, shape and velocity can provide indications of the weather on Mars. Hence, tracking their occurrence in time and space provides useful data for understanding Mars’ current climate. They are also of relevance to landed missions as they can clean solar panel surfaces, extending mission lifetimes. Despite the numerous observations and surveys of dust devils to date, the global extent, distribution and occurrence of dust devils is not yet consistently constrained, mainly due to the relatively limited spatial and temporal scope of manual analyses. To provide the most comprehensive global catalogue of active martian dust devils to date, we applied a RetinaNet convolutional neural network to existing remote sensing images of Mars to identify the distinctive signature of the light-toned lofted dust cloud and dark shadow formed by active dust devils on Mars. The algorithm used ∼6 m/pixel Context Camera (CTX) images from Mars Years 28–36, scanning through a total of 132 359 images. False positives were manually removed with the help of the Zooniverse platform, resulting in 13 409 detections. This survey presents the most spatially and temporally exhaustive global catalogue of dust devils to date. We confirm many trends revealed in disparate previous studies. For example, approximately half of the detections are concentrated in the Amazonis Planitia monitoring site – a hotspot identified from previous imaging campaigns. In addition, we confirm that orbital observations are not well-suited for detecting dust devils at landing sites, despite the ubiquitous detection of vortices with in-situ data. Our study reveals previously understudied hotspots where dust devil lofted clouds can be seen from orbit, most notably southern Hellas Planitia where only dust devil tracks had been previously extensively reported. Importantly, our results reveal latitudinal clusters of dust devils, and in particular large dust devils, at around 60°N and 60°S during local summer solstice, which had only been hinted at by previous work. This concentration is at a much higher latitude than previous modelling suggests, indicating that dust devil generation on Mars is controlled by more factors than are currently accounted for.
KW - Climate
KW - Dust devils
KW - Machine learning
KW - Mars
KW - MRO CTX
KW - Weather
UR - http://www.scopus.com/inward/record.url?scp=85218454262&partnerID=8YFLogxK
U2 - 10.1016/j.pss.2025.106072
DO - 10.1016/j.pss.2025.106072
M3 - Article
AN - SCOPUS:85218454262
SN - 0032-0633
VL - 259
JO - Planetary and Space Science
JF - Planetary and Space Science
M1 - 106072
ER -