Furthering Geophysics in Exploration Using Machine Learning

Geophysics is an unusual animal when it comes to data science. Wavelets are still math, being sine and cosine waves, represented digitally as time and amplitude pairs, and we perform lots of data science on them, but it's never been seen as a probabilities problem before now.

That's not to mean it never was. While checking geometry solutions, we look to see if the values are within some tolerance (about 75-80% accuracy). When running reflection statics, we set a convergence level of 5% on the iterative steps of a multi-variate linear regression model. And as geologists, we say, "the Amplitude Versus Offset (AVO) response is very high in this zone, and with other factors in place we think there's a 92% chance of oil here." It's just that a lot of it has been very subjective and limited to a few attributes.

Enter machine learning. We now have the ability to delve deeper into multiple bits of information with the touch of a computer run. Load up multiple 3D seismic surveys with horizon maps, multiple oil wells, sonic logs, gravity wells, velocity fields, energy absorption characteristics, AVO attributes, and a whole host of other variables. Let the computer examine all of it, from the little specs to the biggest structures, and it can find patterns in the data for you. Think there's oil there? Check the outcomes and see what the probabilities are. If enough of the right attributes agree, you'll settle on black gold.

If you are interested in learning more, check out Geophysical Insights (led by my acquaintance Deborah Sacrey) on how their Paradise software is using deep learning to find new insights. https://www.geoinsights.com/seminars/