Australian researchers have created a new data-driven tool that can accurately predict landslides up to a year before they occur, on a regional scale.
The proof-of-concept research, published last month in the Journal of Geophysical Research by a team at the University of Melbourne, uses satellite observations coupled with machine learning to identify not only if a landslide is going to occur, but where to look.
The tool could be a powerful way to protect people against deadly natural disasters: around 25-50 people are killed by landslides per year in the US alone, and the worldwide yearly death toll is estimated in the thousands.
The two most lethal landslides in Australia’s history were the 1997 Thredbo disaster, which killed 18 people, and the 1929 Briseis Dam disaster which killed 14 people.
So why are landslides so difficult to predict, and what does this new tool do differently?
“Contrary to what an ordinary person might believe, these things don’t just happen spontaneously,” says Antoinette Tordesillas, a professor of mathematics at the University of Melbourne and one of the authors of the new study. “There’s technology for early prediction of landslides, and for the best of them there’s about two days’ lead time.”
But those techniques are expensive, require expert judgement, and you need to already be looking in the right place.
“What we’ve done is target two things: one, we aim for early prediction, so more than those couple of days lead time, because the idea is that then you could do some remediation to prevent the landslide.
“So, our best predictions right on the location of an impending landslide now come a year in advance.”
But the second key feature of the team’s new tool is that you can predict landslides month ahead on a regional scale.
“If you don’t know where to look, and there’s so many slopes, and many of them are moving, but movement doesn’t necessarily mean collapse, then how do you tell which one will actually collapse?”
So, Tordesillas applied her particular field of expertise to the problem, which is the study of failure in granular materials. Tordesillas says all granular materials, at whatever scale, behave in similar ways: so, a hillside slope that moves and shifts will behave much like any other granular material, mathematically. In fact, there is a unique pattern of motion before a slope collapses – a kind of signature.
The groundbreaking feature of the team’s new tool, then, is that it applies recently acquired wisdom in the mathematical field of granular failure to look at various points of motion.
“It’s like a dance, a choreography,” explains Tordesillas. “It’s a pattern of motion, but it’s a collective kind of dance.
“Imagine you’ve got 1,000 dancers. If I blocked your view of 999 dancers, and all I’m giving you is the ability to watch what happens to one dancer, it’s going to be very difficult for you to identify this choreography just by tracking the motion of that one dancer.”
That analogy, Tordesillas says, describes most landslide predictions today. “When they try and predict a landslide, what they’re tracking and modelling is one point. That’s the limitation we’ve overcome.
“You cannot watch one location on a mountain slope – that won’t work, and that’s how you miss events, or get false alarms,” she says. “So really being able to know what aspect of that total choreography is going to lead to catastrophic collapse is the key to this.”
Doing this requires combing through and processing huge troves of satellite data about various points on the surface of the planet to make these predictions: this study is the first to demonstrate success at a regional scale.
Moreover, the new technology can be used with freely available satellite data, so it could be an accessible tool for people living in high-risk situations.
And predictions from the model can be used to monitor and manage natural hazards that will be compounded in the future by the rising sea levels and volatile weather associated with climate change – like landslides and avalanches, which can also trigger tsunamis.