Machine Learning Tools for Automatic Mapping of Martian Landforms

Express. These orbiters are surveying the entire Martian surface to understand the planet’s past and the geological, climatic, and other processes responsible for its present state. They gather ever-increasing amounts of spatially extended data. Science data archived from all planetary missions prior to the 2001 Mars Odyssey mission totals approximately 5 terabytes. NASA expects that number to double over the Odyssey’s duration and to increase by a factor of 10 with the newest Mars Reconnaissance mission. This data deluge presents difficult transmission, storage, and distribution challenges. Less publicized is the scientific community’s challenge to process, analyze, and ultimately turn a significant portion of the collected data into knowledge. Only a small fraction of the data is analyzed at this time because analysis still involves traditional methods that rely on visual inspection and descriptive characterization. Automated or semiautomated tools for Martian data analysis can substantially broaden the scope of scientific inquiry. Recognizing this opportunity, we’ve undertaken research to apply pattern-recognition and machine-learning tools to automatic analysis and characterization of the Mars surface. This research includes machine surveys of specific landforms, such as impact craters and valley networks, and automatic generation of geomorphic maps.