Recognition and Classification of Martian Chaos Terrains Using Imagery Machine Learning: A Global Distribution of Chaos Linked to Groundwater Circulation, Catastrophic Flooding, and Magmatism on Mars

Abstract

Martian chaos terrains are fractured depressions consisting of block landforms that are often located in source areas of outflow channels. Numerous chaos and chaos-like features have been found on Mars; however, a global-scale classification has not been pursued. Here, we perform recognition and classification of Martian chaos using imagery machine learning. We developed neural network models to classify block landforms commonly found in chaos terrains—which are associated with outflow channels formed by water activity (referred to as Aromatum-Hydraotes-Oxia-like (or AHO) chaos blocks) or with geological features suggesting volcanic activity (Arsinoes-Pyrrhae-like (or AP) chaos blocks)—and also non-chaos surface features, based on >1400 surface images. Our models can recognize chaos and non-chaos features with 93.9% ± 0.3% test accuracy, and they can be used to classify both AHO and AP chaos blocks with >89 ± 4% test accuracy. By applying our models to ~3150 images of block landforms of chaos-like features, we identified 2 types of chaos terrain. These include hybrid chaos terrain, where AHO and AP chaos blocks co-exist in one basin, and AHO-dominant chaos terrain. Hybrid chaos terrains are predominantly found in the circum-Chryse outflow channels region. AHO-dominant chaos terrains are widely distributed across Aeolis, Cydonia, and Nepenthes Mensae along the dichotomy boundary. Their locations coincide with regions suggested to exhibit upwelling groundwater on Hesperian Mars.

Extraction of stratigraphic exposures on visible images using a supervised machine learning technique

“…Another application example is the detection of volcanic rootless cones (Palafox et al, 2017). Compared with support vector machine approaches, Shozaki et al (2022) demonstrated the recognition and classification of Martian chaotic terrains using convolutional neural network models. Their models showed over 90% accuracy for the classification and contributed to revealing the global distribution of chaos features on Mars.…”

MarsMapNet: A Novel Superpixel-Guided Multiview Feature Fusion Network for Efficient Martian Landform Mapping

“…Many studies have been conducted in order to identify a specific type of Martian landform, such as impact craters [12], dunes [13], dark slope streaks [14], volcanic rootless cones and transverse aeolian ridges [15,16], CO2 jet deposits [17] and chaos terrains [18]. The morphology and distribution, together with material composition analysis, of these landforms can be used to reveal their causes.…”