Algorithms can use seismic data and structural knowledge of buildings to judge which parts of a city will be most at risk and prioritise rescue efforts
Where to start? (Image: Daniel Berehulak/New York Times/Eyevine)
WHEN an earthquake hits, emergency responders need to know which areas to go to first, as quickly as possible.
Can artificial intelligence help? Ahmad Wani thinks so. His start-up,, has developed technology that, within minutes of a tremor, can predict what areas of a town are likely to be worst affected.
“I think the product is going to save lives,” says Ray Mueller, who is on the emergency services council for San Mateo county in California, which sits over the San Andreas fault. “It’ll allow you to identify areas that are most affected by an earthquake so that first responders will know where to go.”
To develop the algorithm, One Concern first loaded it with public data including the age, type and construction materials of individual buildings in a town. It was then trained to understand the ways in which earthquakes can damage buildings. By combining this knowledge with seismic data following a quake, Wani says his system can effectively predict how buildings will react to the shock waves.
Once the AI has made its assessment, it plots a damage map that emergency responders can analyse. The streets where buildings are expected to be most devastated are highlighted, and the map also shows the areas where the highest number of people are likely to be affected.
“The real benefit is emergency responders can be very rapidly updated on an emerging situation”
“We know the population of every block,” says Wani. “We multiply the damage and population to get a response priority. So a school with 100 children – it makes more sense to go there.”
One Concern’s software is currently being trialled by emergency services in San Mateo, who are checking how it handles data from small tremors and what maps it produces. For years there have been concerns over the resilience of housing in at-risk areas along the San Andreas and Hayward faults. The government has a list of 1500 older concrete buildings in Los Angeles, for example, which are believed to be vulnerable.
Two other counties in the region are said to be interested in testing the software. Eventually, says Mueller, responders will also be able to use the system to train their staff by simulating mock quakes and seeing how buildings respond.
Other AI strategies are proving useful in less-developed areas of the world where there is not much data available about the buildings present. Following the earthquake that hit, for example, members of the UK’s ORCHID project, an AI research initiative, were involved in trials that ended up changing the actions of response teams on the ground.
In one case, online volunteers were shown satellite images of the affected area after software had pre-scanned them and decided there might be settlements. Volunteers rated images as high priority if they thought houses were present and the AI engine worked out whether there seemed to be a consensus. Certain sites were then flagged for a closer look by specialist response teams.
Two villages, not previously known to teams on the ground, were identified, and responders were able to provide assistance to the inhabitants.
“AI initially identified a whole batch including those two villages and then humans went through that batch for the cross check,” says David Jones, chief executive of Rescue Global, a charity involved with the ORCHID project.
The episode demonstrates how this sort of technology could be best deployed – as a data-crunching tool harnessed by human operators who ultimately call the shots.
“I think the real benefit is if responders can be very rapidly updated on an emerging situation, and AI might be able to do that,” says Peter Sammonds of the Institute for Risk and Disaster Reduction at University College London.
After a disaster, it pays to find out fast which areas have suffered most damage. Sending out drones fitted with cameras can help, but how do you sift through all the data they bring back?
Patrick Meier, at the Qatar Computing Research Institute, is working with Kathmandu University in Nepal to solve that problem. Meier and his team have developed an algorithm to scan the images that the drones take and highlight anything that appears to be settlements.
The researchers are also training it to spot damaged buildings by showing the algorithm example photographs so it can recognise similar sights elsewhere.
This article appeared in print under the headline “Quake AI plans disaster rescue”
This entry passed through the Full-Text RSS service – if this is your content and you’re reading it on someone else’s site, please read the FAQ at fivefilters.org/content-only/faq.php#publishers.