Using Machine Learning to Map Western Australian Landscapes for Mineral Exploration

Landscapes evolve due to climatic conditions, tectonic activity, geological features, biological and sedimentary dynamics. Geological processes at depth ultimately control are linked the resulting surface features. Large regions in Australia, West Africa, India, China blanketed by cover (intensely weathered material and/or later sediment deposition, both up hundreds of metres thick). Mineral exploration through poses a significant technological challenge worldwide. Classifying understanding landscape types their variability is key importance for mineral covered regions. Landscape expresses how near-surface geochemistry underlying lithologies. Therefore, mapping should inform geochemical sampling strategies exploration. Advances satellite imaging computing power have enabled creation large geospatial data sets, sheer size which necessitates automated processing. In this study, we describe methodology enable pattern domains using machine learning (ML) algorithms. From freely available digital elevation model, derived data, sample landclass boundaries provided domain experts, our algorithm produces dense map model region Western Australia. Both random forest support vector classification achieve approximately 98% accuracy with reasonable runtime 48 minutes on single Intel® Core™ i7-8550U CPU core. We discuss computational resources study effect grid resolution. Larger tiles result more contiguous map, whereas smaller detailed and, some point, noisy map. Diversity distribution landscapes mapped previous results. addition, results consistent trends main basement features region. Mapping scale can be used globally as fundamental tool guiding efficient programs under cover.