Machine learning as a tool to classify extra-terrestrial landslides: A dossier from Valles Marineris, Mars

Document Type


Publication Date



Department of Geological and Mining Engineering and Sciences


Many researchers studied Martian landslides, and most of these studies are in the Valles Marineris (VM). Those studies are mainly morphologic analysis, estimates of landslide deposit volumes, thermal properties of the landslides, spectral analyses, and mapping and classifying landslides. But, this study created a robust landslide inventory of 682 incidents by collation of data from different sources, apart from our own mapping. This mapping was done for the completeness and accuracy of the landslide inventory. This inventory consists of 141 debris flows, 291 rock avalanches, and 250 slumps. This classification was based on the available literature, and these landslides covered Hebes, Ophir, Candor, Melas, Coprates, Ius and Tithonium chasmas. Landslides were mapped with the aid of images from Mars Reconnaissance Orbiter (MRO) Context Camera (CTX), Mars Orbiter Mission (MOM) Mars Colour Camera (MCC), and high-resolution Google Mars. Digital Elevation Model (DEM) from the Mars Express (MEX) High-Resolution Stereo Camera (HRSC), Thermal Inertia (TI) from Thermal Emission Imaging System (THEMIS), and available geological map were used to define 15 critical parameters of the landslide inventory viz., length (L), width (W), relative relief (RR), area, L/W ratio, RR/L ratio, TI of landslide (mean, minimum, maximum), and slope statistics (mean, minimum, maximum), TI of the scarp, the slope of the scarp, and geology. These 15 parameters were subjected to a multicollinearity testing, which excluded the dependent variable TI (mean). Further, the rest of the variables were analyzed using multiple discriminant analysis (MDA), where the parameters RR/L, L/W, and RR showed higher canonical discriminant function coefficients of 1.245, 0.461 and 0.159, respectively, suggesting that these are the parameters that primarily discriminates the landslides. The classification of landslides was evaluated using a 10-fold cross-validation method using different machine learning algorithms. The five models that provided the best results are simple logistic, logistic regression, meta-classifier, multi-layer perceptron, and Sequential Minimal Optimization (SMO). Of these, simple logistics has the maximum classification accuracy of 81.09%. Thus, this study demonstrates the potential of using machine learning models to classify extra-terrestrial landslides, especially the long-runout landslides that are characteristics of other planets and satellites.

Publication Title