How AI Can Be Useful In Disaster Reaction

I was dubious when I first learned that artificial intelligence (AI) might be beginning to assist catastrophe response, including in the wake of the earthquake that devastated Turkey and Syria. We frequently hear grand (and unrealistic) promises about the potential of AI to solve the world’s ills.

However, xView2 appears to be a successful US Department of Defense initiative. Although it is still in the early stages of deployment, this visual computing initiative has already aided in Turkey’s ground rescue operations and disaster logistics.

In 2019, the Pentagon’s Defense Innovation Unit and Carnegie Mellon University’s Software Engineering Institute sponsored and created the open-source project xView2. It has since worked with numerous research collaborators, including Microsoft and the University of California, Berkeley. It identifies and categorizes building and infrastructure damage in the disaster area much more quickly than is currently feasible using machine-learning algorithms in conjunction with satellite imagery from other providers.

The Defense Innovation Unit’s chief AI scientist and a researcher at Berkeley, Ritwik Gupta, tells me that this means the program can directly assist first responders and recovery specialists on the ground get an assessment that can help find survivors and help coordinate reconstruction efforts over time.

Gupta frequently collaborates in this process with major foreign institutions like the World Bank, the United Nations, and the US National Guard. The Australian Geospatial-Intelligence Organisation and the California National Guard have both used xView2 in the past five years to fight wildfires. More recently, it was used to aid in the rebuilding process after flooding in Nepal, where it helped locate damage caused by ensuing landslides.

In Adiyaman, Turkey, which has been devastated by the earthquake and where residents have been frustrated by the delayed arrival of search and rescue, Gupta claims that at least two different ground teams of search and rescue personnel from the UN’s International Search and Rescue Advisory Group have used xView2. He notes that the World Bank, the International Federation of the Red Cross, and the United Nations World Food Programme have all used the platform in response to the earthquake. xView2 has also been used elsewhere in the disaster zone and has been successful in assisting workers on the ground be “able to find areas that were damaged that they were unaware of,” he says.

Gupta tells me, “If we can save one life, that’s a good use of the technology.”

Why AI is useful

The algorithms use a method called “semantic segmentation,” which is akin to object recognition in that it assesses each individual pixel of an image and its connections to neighboring pixels to reach conclusions.

The satellite pictures of the damage are on the left, and the model’s assessment is on the right. The darker the red, the worse the wreckage, is how this appears on the platform in the images below. This same level of assessment, according to Atishay Abbhi, a catastrophe risk management expert at the World Bank, used to take weeks instead of hours or minutes.

This is an improvement over more conventional disaster assessment systems, which depend on eyewitness reports and calls for rescue and emergency responders to quickly determine where assistance is required. However, this can still take days, if not longer. In some more recent instances, fixed-wing aircrafts like drones have flown over disaster areas with cameras and sensors to provide data evaluated by humans. The reality that various responding organizations frequently have their own siloed data catalogues slows down the typical reaction even more, making it difficult to develop a uniform, shared understanding of which areas require assistance. Organizations can organize and prioritize their responses with the aid of xView2 in a matter of minutes, saving time and lives.

The obstacles

Of course, this technology is not a panacea for disaster reaction. xView2 faces a number of significant difficulties that take up the majority of Gupta’s study time right now.

The model’s dependence on satellite imaging, which only provides clear images during the day, when there is no cloud cover, and when a satellite is overhead, is the first and most crucial factor. Three days after the initial earthquake, on February 9, the first usable pictures from Turkey finally arrived. Additionally, there are significantly fewer satellite pictures captured in isolated and less developed economic regions, such as Syria, which is just across the border. Gupta is investigating novel imaging methods like synthetic aperture radar, which produces images using microwave pulses rather than light waves, to solve this.

Second, because satellite images have an aerial viewpoint, even though the xView2 model is up to 85 or 90% accurate in its precise evaluation of damage and severity, it also can’t really detect damage on the sides of buildings.

Last but not least, Gupta claims it has been challenging to convince local organizations to employ and believe in an AI solution. He claims that first rescuers are very conventional. “They’re not going to trust it at all when you start telling them about this fancy AI model, which isn’t even on the ground and is looking at pixels from like 120 miles in space.”

Next steps

xView2 supports catastrophe response efforts at various stages, from identifying damaged areas right away to determining potential locations for secure temporary shelters to determining the scope of long-term reconstruction. At the World Bank, Abbhi expresses his hope that xView2 “will be really important in our arsenal of damage assessment tools” going forward.

Anyone could use the software because the code is free and open source. And Gupta wants to maintain the status quo. He declares, “I detest it when companies come in and start saying, We could commercialize this. “This ought to be a government service run for the benefit of all.” Gupta is developing a web application that will enable any user to conduct evaluations; at the moment, businesses request analysis from xView2 researchers.

According to Gupta, researchers should concentrate on the varieties of AI that can have the greatest positive effects on society rather than discounting or exaggerating the part that emerging technologies can play in complex issues. He inquires as to how the discipline of AI can turn its attention to these extraordinarily challenging issues. In my view, [these] are much more difficult than, say, creating new text or images.

How AI Can Be Useful In Disaster Reaction

Why AI is useful

Next steps

Download The Radiant App To Start Watching!