Ellen Phiddian
As with just about every other field, AI and machine learning has been making waves in geology. So is it helpful, or hype? And is it likely to put any geologists out of their jobs?
The mood at the Geological Society of South Australia (GSSA)’s Discovery Day and Western Gawler workshop held at the end of November is that, while AI and machine learning are changing some games, you still need an expert in the room.
The GSSA has been using AI and machine learning techniques to help explore and understand the western Gawler Craton, an ancient geological province that takes up much of the centre of the state and houses a variety of critical minerals.
“I see it as another tool,” Dr Mark Pawley, a senior geologist with the GSSA, tells Cosmos.
Pawley has done both manual and automated “lineament mapping”: looking at linear patterns in a landscape that hint at geological features. He says that the unbiased machine learning program can be a suitable check against what a person generates.
“We tend to link lines, I’ve found, because we want a nice big single structure. Rather than having a series of segments, which is what the algorithm would do. We tend to join them.
“I think that it’s going to be a reality check.”
But Pawley says that expertise in geology is still needed to work with the program. He’s been working on interpreting aeromagnetics: surveys of the magnetism of an area carried out from an aircraft. Different rocks have different levels of magnetism.
“[If] you have a gradient, you don’t know if that’s the boundary of a rock type, or the boundary between a rock and the structure that’s been demagnetised. You just know that there’s a change from magnetically high to magnetically low,” says Pawley.
“What caused that is the interpretation. At the moment, I don’t know that the machine learning could do that. I think you’d still need someone to look and go ‘oh, it’s this’.”
Dr Yusen Ley Cooper, a senior geophysicist with Geoscience Australia, has also been working with machine learning and airborne electromagnetic surveys.
Ley Cooper’s work is slowly filling in electromagnetic data for the whole of Australia – all done via aircraft surveys. Planes cross the country in straight lines, scanning all the electromagnetism beneath them.
“We have these lines of data that we acquire every 20 kilometres,” he says.
Figuring out what lies between those neat 20-kilometre lines is a challenge – and it’s where the geologists have been testing machine learning.
“There’s been lots of ways of drawing these connections, there was a method called ‘Kriging’ that’s been used and very successful.
“Now what we’ve done is try to use ancillary datasets to find these correlations.”
Ley Cooper says that the technique works – but geologists need to be careful interpreting it.
“You need to know where these relationships hold and where they don’t.”
Dr Ian Roach, program leader of stratigraphic drilling at Geoscience Australia, agrees that machine learning needs “very, very careful supervision” in geology.
“It’s not a magic bullet, and it can’t be allowed to steer decision making,” says Roach.
“Some people would tend to just push a button and take the result without questioning.”
But Roach adds that the ability to crunch reams of data is an advantage.
“[Machine learning can] deal with very large data sets to that ordinary mortals just can’t cope with. So I think it’s fantastic so long as it’s well trained – like a dog.”
