Using Data Science for Subsurface Imaging
From its largest to smallest scales, we use large and complex datasets to infer properties of and understand processes within the Earth. In this talk, I will show three examples of how we can use emerging tools from Data Science in different applications of subsurface imaging. The first example illustrates how machine-learning algorithms can be used to infer relationships between data and rock properties when we have only a limited number of labelled data, and applies this work to correlate facies with well data. The second example shows how we can reduce our model size to allow for more complicated analyses, such as in uncertainty quantification applied to time-lapse seismic imaging. The final example shows how we must carefully choose which data to collect and analyze using laboratory experiments as proxies for large-scale post-seismic velocity recovery. Together, these examples illustrate a diverse set of subsurface imaging problems, showing the efficient use of tools and ideas from Data Science to improve our understanding of the Earth.