P042-0011
Classification of Martian chaos terrains using imagery machine learning: Implications for ground ice distributions and dynamics
Abstract:
Here, we perform machine learning for recognition and classification of Martian chaos terrain imagery to distinguish the proposed formation mechanisms. Based on the fine-tuned trained model, we constructed three classifiers of visible images in grayscale, images of thermal inertia map, and colored images of the digital elevation model. We prepared 1240 images as training data for each modality. We defined two types of chaos terrains: Ones are proposed to have been formed through melting of ground ice (hereafter we call ice-melting chaos), and the others are suggested to have been generated in association with magma excavation and tectonic tiles (hereafter we call volcanic chaos). We also collected 800 images of non-chaotic features, such as valley networks and impact craters.
Our constructed classifiers can distinguish chaotic terrains with non-chaotic features with ~95% accuracy. They can also classify ice-melting chaos and volcanic chaos with ~97% accuracy. By applying our classifiers to uncategorized chaotic terrains, we obtained the spatial distribution of ice-melting and volcanic chaos over Mars. We find that ice-melting chaos are the most abundant type on the Martian surface. Ice-melting chaos mainly occur near the dicotomy boundary, including locations where emissions of CH4 are suggested (Giuranna et al., 2019). Our results also show that most of volcanic chaos seem to occur surrounding ice-melting chaos in large depressions. These results would provide insights into the distribution of ground ice and the dynamics of ice melting events on Mars.