Inferring Mars' Surface Winds by Analyzing the Global Distribution of Barchan Dunes using a Convolutional Neural Network

Lecture / Seminar
Time: 10:00-11:00
Lecturer: Lior Rubanenko
Organizer: Department of Earth and Planetary Sciences
Details: Department of Geological Sciences Stanford University
Abstract: Sand seas on Mars are riddled with eolian landforms created by accumulating sand ... Read more Sand seas on Mars are riddled with eolian landforms created by accumulating sand particles. When the sand supply is limited and the wind is approximately unidirectional, these landforms take the shape of crescentic barchan dunes, whose slip-faces are approximately perpendicular to the dominant wind direction, and their horns are oriented downwind. The morphology of barchan dunes is thus routinely used to infer wind conditions on Mars by manually analyzing aerial or satellite imagery. Despite the effectiveness of this technique on a local scale, employing it on a global scale remained challenging thusfar - as manually outlining individual dunes globally is impractical, and automatic detection methods have been largely ineffective at accurately segmenting dunes in images. Here we use Mask R-CNN, an instance segmentation convolutional neural network, to detect and outline dunes globally on Mars in images obtained by the Mars Reconnaissance Orbiter Context Camera (MRO CTX). We measure the morphometrics of dunes from their detected outlines, and infer the direction of the winds that formed them. By comparing the global wind distribution we derived to a global climate model, we study Mars' past and recent climate, and constrain global sand mobility thresholds which offer insight into the erosion and dust lifting capabilities of the atmosphere of the Red Planet.
Close abstract