New Software Can Predict Landslides Weeks Before They Happen

An anonymous reader quotes a report from Smithsonian: Australian researchers may have found a way to detect landslides as far as two weeks in advance, giving residents time to evacuate and engineers the opportunity to shore up slopes. Using AI and applied mathematics they've developed a software that can identify the subtle signs of an impending slide, signs that would be invisible to the naked eye. "Right now, a lot of the predictions [about where landslides will happen] are based on someone's gut instinct on the location," says Antoinette Tordesillas, a professor at the School of Mathematics and Statistics at the University of Melbourne, who co-led the research. "We don't rely on gut instinct. We want to develop an objective method here." To develop the software, Tordesillas and her team used radar data from mining companies, which produce extremely detailed information of the surface movement of slopes. The team took the data and looked for patterns, eventually figuring out which networks of movements indicated unstable locations. They also used data from a landslide-prone Italian volcano to help develop the algorithm. The software can also incorporate data about other landslide risk factors, like rainfall and erosion, making the targeting even more precise. The data used for monitoring can come from radar based on the ground, on satellites or even in drones.

But can it not predict landslides

[Geoffrey.landis]

More important, though, is whether it can NOT predict landslides before they DON'T happen.

low rate of false positives is also important

Re:

[110010001000]

We already have monitoring, including wireless pressure and flow sensors and satellite imagery. But this is a new generation of people discovering neural networks so expect to see a lot of "amazing" progress. Next, someone will have software that predicts rainfall based on feeding a neural network.

Re:

[110010001000]

That is coming in Phase II of the 5 year research.

Re:

[Anonymous Coward]

More important, though, is whether it can NOT predict landslides before they DON'T happen.

low rate of false positives is also important

As of now, it can't do anything. The headline is absolutely false, the software has not predicted a single landslide. They are working on it and hope it eventually will.

Re:

[Type44Q]

Actually, a wide range of can *not* predict landslides before they don't happen. This list includes (but isn't limited to):

aircraft carriers

dragonflies

dingle-berries

Total BS

[110010001000]

There is already open source landslide detection software (one written by NASA). If these "researchers" would contribute to already existing solutions we would be much better off.

Mudslide

[Oswald McWeany]

Predicting Landslides has always been a challenge. Predicting Mudslides less so- just look for spikes in sales of Kahlua and Vodka in college towns.

About time

[Opportunist]

So far every election day every politician and statistician is surprised by the way people vote.

Oh, THOSE landslides...

[K. S. Kyosuke]

My first thought was of UNIVAC I. :)

So can I

[OneHundredAndTen]

The issue is, how accurately? Not very, in my case. This system, we do not know yet.

Predicts them? Or causes them?

[mark-t]

(Conspiracy theories.... gotta love 'em)

Comment removed

[account_deleted]

Comment removed based on user account deletion