Scientists seek to solve mysteries of ancient Native city
Few technological developments have captured the minds — and fear — of humanity like artificial intelligence. Whether it’s robots rising up to subdue their makers like in the Westworld series, or the malevolent computer Hal 9000 from the classic movie 2001: A Space Odyssey, machines that can learn are depicted as threats to the world as we know it.
Obviously, these futures are the work of imaginative screenwriters. In fact, artificial intelligence, or AI, is at work in the field of geological science right now helping to preserve the world and save lives. That topic is the focus of a virtual seminar series throughout the summer called “Machine Learning in Solid Earth Geoscience,” a series that has been hosted in Santa Fe in pre-COVID-19 years.
We haven’t yet reached the point where machines can think for themselves. So when we speak of AI, it’s about the capacity for computer algorithms to take in massive amounts of data and uncertainty and then identify patterns, learn from those patterns, and make predictions thousands of times faster than a human can.
This is especially helpful in the realm of geological science, where we look for patterns deep below the Earth’s surface to understand Earth’s processes and resources. In geology, there are a range of unknowable possibilities, and separating one slice of information among the cacophony of natural noise is impossible for the human mind.
Earthquakes are a good example, because while we can predict many of the world’s natural disasters, alerting cities to an oncoming earthquake remains elusive. But the answer may lie in AI.
To find that answer, we study fault zones, which constantly tremble as one tectonic plate pushes against another, creating vibrating signals that look much like acoustic waves. Occasionally, deep underground, the plates subtly shift, causing a “slow slip” movement that is almost imperceptible to humans.
Leading up to each slow slip is a seemingly random jumble of these acoustic waves. For years, scientists had dismissed this as noise, useless random data. Recently, however, Los Alamos National Laboratory applied
AI to this data and discovered it was hiding a trove of information about seismic behavior and earthquakes. For instance, two years ago, a Los Alamos study trained an AI algorithm on slow slips along the Cascadia subduction zone in the Pacific Northwest. By focusing on these seemingly random signals as the plates pushed together, the algorithm detected an obvious pattern preceding a slow slip, a kind of signature, that allowed scientists to predict an event to within a few days.
This science is still nascent, but it has opened an exciting new pathway that may one day lead to global earthquake prediction systems.
This same principle is being applied to carbon capture and storage efforts, in which we remove CO² from an industrial process or the atmosphere and store it underground to reduce its impact on climate change. The goal, obviously, is to trap this gas thousands of feet beneath the earth’s surface under layers of rock. Unfortunately, CO² is buoyant, so it must be stored underground in well-sealed areas to prevent it from leaking back to the atmosphere. And that raises a question: how long will it remain underground?
First, with the use of sensors, AI can help sniff out CO² leaks, looking for consistent patterns that can then trigger an alert of seepage. Second, AI can help us predict the fate of CO² over thousands of years. Will the CO² gas eventually transform into a solid, making the site more desirable and less prone to leaks? Along with geological data and historical information supplied by scientists, AI can provide a range of scenarios, then home in on the most likely outcome.
This ability to peer into the future is also helping to predict water shortages in New Mexico. In our state, about half of drinking water comes from beneath the ground. Farmers tap wells to raise vegetables, orchards, and feed for animals. In a future with less rainfall, it’s vital that we know how much water is in the ground, and how much is leaving on any given day or month.
This can be done with the aid of groundwater modeling, which helps us understand where underground water moves. In one spot, there may be a vast aquifer below. But 100 yards in any direction the water may be forced to flow around an underground rock formation, rendering the ground beneath that location dry. And this can all change depending on how much water is pumped from one location.
Linked with groundwater flow and aquifer models, an AI algorithm can tell us that if Farmer Joe pumps X number of gallons, he may disrupt his neighbor’s well. The AI can then use expected precipitation for any given area to predict how this will change in a year or a decade. AI can also provide a daily look at how much water is being removed from the aquifer and that information can help municipalities set standards that will benefit Farmer Joe and the people who expect water to flow from their faucets each time they turn the nob.
The complex interactions of humans with their environment, and especially the underground resources we depend on, challenge our understanding, making them hard to predict. And it’s only becoming more complicated with growing populations and changes to the environment and climate. Luckily, complex problems with unknown outcomes are exactly what AI is best positioned to solve.