AI could help identify how and where landslides will occur

New modelling could help identify how and where landslides will occur in the event of an earthquake or heavy rainfall.

Earth Sciences New Zealand and Massey University are leading the Sliding Lands Hōretireti Whenua work, which will use existing landslide databases with machine learning and artificial intelligence.

The model aims to work out the probability of a landslide given a certain amount of rain or seismic activity to trigger it.

Engineering geologist at Earth Sciences New Zealand Saskia de Vilder told Nine to Noon the goal is to have national scale hazard models for a particular type of landslide.

"What we're really talking about here is these kind of fast-moving rapid-velocity landslides that we see triggered by earthquake and rainfall events.

"These landslides are the ones where you're going to have limited or no evacuation time."

She said the tool was not going to give "site specific information" of where a landslide might take place but would act as a screening tool of where authorities would need to look further.

"For example... because we can link it to rainfall, we can take in rainfall forecasts coming up to an event, we can produce these maps and that might be able to say help preposition where you might need response crews prior to an event.

"It might help you determine where you might need to look more during an event and afterwards it might help where you need to go for a reconnaissance."

The model divided regions into "landslide domains" that considered factors like climate, geology and rainfall.

de Vilder said getting more location specific information would depend on how the information is used but it was an improvement on what had been used before.

She said the model aimed to be transparent, and users would be able to see where machine learning tools were used and which landslide informed the modelling.

"We want to be able to, so you can interrogate that model and see what's actually driving what's, you know, what's driving it."

De Vilder said part of their project was also to work with different end users including councils to see how this model can be applied.

"How can what we're producing be helpful to councils and feed into that hazard mapping process? So that's definitely ongoing conversations."