Australia is progressively running out of readily accessible resources close to the surface.
Many mines reaching the end of their lifespan now were prospected several decades ago and most near surface resources have either already been found – or we know roughly where they exist.
However, while the near-surface mineral deposits are drying up, demand for resources for both domestic use and for export shows no sign of slowing. Prospecting must now go further underground, below the rocky sediments that obscure the deep deposits of these much-needed minerals.
But deep exploration is difficult and expensive. It needs to be underpinned by detailed knowledge about the subsurface from geophysical data. It’s crucial to get the most accurate and meaningful information possible, before the physical prospecting begins.
That's where CSIRO's Deep Earth Imaging future science platform team comes in.
These scientists are developing the next generation of tools to craft models of the complex geology that lies deep beneath the surface. To do this, they use as many sources of geoscientific information as possible – including electromagnetic, gravitational and seismic datasets.
Imaging the deep Earth using integrated geophysics
Different geophysical methods will give different clues about what's below the surface and combining them can give researchers even more information. But finding meaningful information from the vast amount of data is challenging.
As part of the Deep Earth Imaging team, CSIRO postdoctoral fellow Cericia Martinez, is automating two dimensional (2D) seismic velocity models from seismic data to better understand Australia's crustal geology and build a picture of the subsurface.
"We can't always go and dig or drill a hole to figure out what’s in the subsurface, but we can use physics and maps and data to help us do that," Dr Martinez says.
"One way we can understand what is beneath the surface is to look at how seismic vibrations travel in the subsurface. Just like my voice travels through the air via sound waves, seismic waves travel through rocks in the Earth's crust," she explains.
By sending a vibration into the Earth and measuring the time that it takes for a wave to travel through the subsurface from one location on Earth to another location on Earth, Dr Martinez says we can start to try to understand what lies hidden beneath us.
Because rocks and other subsurface features have different properties and densities, seismic waves travel at different velocities and are subject to other physical properties of waves such as reflection and refraction.
Dr Martinez has been mining the data acquired from seismic transects across the continent done by Geoscience Australia. Her research focuses on looking at seismic travel-time data – the time it takes for seismic waves to travel through the subsurface – to identify the probable crustal architecture and geological features below.
"The work that I'm doing, this geophysical inversion with seismic travel-time data, is trying to isolate the seismic velocity data which is just one of the many properties of a rock type," Dr Martinez explains.
Identifying subsurface mineral, water, oil and gas reserves
Dr Martinez says that seismic velocity data can help to identify different subsurface features and geological units – such as water, oil and gas reservoirs and orebodies.
But there's a problem with this data: there's a huge volume of it, it's very complex, and analysis of it requires lots of time and effort from experts who have very specialised skills-sets and domain knowledge.
Dr Martinez's research incorporates geology, physics, mathematics and computer science to develop a new geophysical inversion algorithm for seismic travel-time data. She is focusing on increased automation, classification and modelling to take away some of the time-consuming tedium of wrangling raw seismic data.
"Hopefully with the algorithm that I'm developing, this process will be faster and require fewer processing hours," she says.
One major challenge in trying to work with big datasets such as those Dr Martinez is using, is accessing the 'good' data while filtering out the uncertain data that contains background noise that distracts and hides what information is wanted.
"When looking at data like this, you first need to know something about what you're looking for. One person's signal can be another person's noise, which is actually the challenge," she says.
Automation and machine learning to improve accuracy
"We are trying to develop this algorithm that doesn't require as much domain knowledge by using automation and machine learning to handle the data uncertainty," Dr Martinez says.
"The goal of our algorithm is to be quick, agile and responsive in generating regional scale velocity models that give us an idea of the big-picture architecture of the subsurface."
Dr Martinez says by correlating her research with known geologies, she can extract greater meaning and offer greater predictability of the location of subsurface resources.
"We want to be able to compare the results that we get from a variety of geophysical analyses to something that’s already known," Dr Martinez says.
"If you had four holes in the region that you would already be able to pull rock samples from, and you could then compare it to the modelling formed by our algorithm, that'd be the ideal outcome."
The research is relatively new and there is a lot more work to be done, Dr Martinez warns, adding it's a long way until the model can be used in the field in real-time.
"After proving that the algorithm works in controlled datasets, we'd eventually like to get these algorithms more streamlined so that once you collect the data, it's automatically processed and able to be used immediately."