Researchers have developed a new tool to assess earthquake damage using artificial intelligence, a leap that could ultimately help first responders make critical rescue decisions, a new study suggests.
The team’s AI, called the LoRA-Enhanced Ground-view Generation (LEGG) diffusion model, is trained on real drone aerial imagery and used to create highly photorealistic 3D reconstructions of the ground. By creating images detailed enough to fully capture the physical features of an area, this synthetic model can be characterized to recognize complex visual patterns and predict where structures may be damaged, even in densely populated urban areas.
“What our algorithm does is generate thousands of sets of semi-realistic photos of the top of the building and what it looks like from the ground,” he said. Hata Eishun Co-author of the study and professor Civil Engineering, Environmental Engineering, and Geodetic Engineering at The Ohio State University. “Drones collect important information from the air, but humans actually make urgent decisions from a ground perspective, so capturing such data is critical.”
A similar study of the aftermath of a devastating earthquake unmanned aerial vehicle or LIDAR-based detection method While researchers had surveyed collapsed buildings and structures from the air, no one had considered what the damage looked like on the ground before the lengthy rescue operation. Additionally, depending on the severity of the earthquake, manual damage assessments can take days to weeks to fully complete, which is not ideal for rapid recovery missions.
In this paper, Hata et al. introduce a framework to fill these gaps using AI-generated imagery, with the aim of laying the foundation for more accurate disaster assessments and better earthquake preparedness.
“This simulation is essentially a map, but an experienced, well-trained AI could provide additional information that would be very helpful for emergency personnel to quickly decide where to go over time,” Qin said.
This study International Journal of Remote Sensing.
To test the applicability of their proposed algorithm, the researchers conducted a case study on a real-world disaster. 2023 Kahramanmaras, Türkiye, earthquakea powerful 7.8 magnitude earthquake destroyed 280,000 buildings and damaged at least 700,000. Comparing drone images from 2015 with photos taken in the days after the quake revealed dramatic changes to the local built environment, including collapsed buildings and temporary shelters in vacant lots.
After presenting a dataset of just 3,000 of these urban structures to AI, the model demonstrated that it was able to create images that enhanced recognition of many building issues, such as facade cracks, building lean, and partial collapse, and that it could extract subtle cues from multiple sources to generate high-resolution, photorealistic street-level views.
This advanced capability emerged from a combination of drone and ground imagery that researchers injected to ensure the model had a strong starting point for understanding potential structural damage and its impact on communities, Hata said.
“As long as you have good data, AI can be very generous in predicting past and future outcomes,” he said. “This is a very helpful tool.”
In the future, the team’s framework could be applied to new scenarios and fields to help governments and engineers design more resilient infrastructure and reimagine post-disaster assessments and emergency management policies.
“This study provides engineers and other decision makers with an excellent opportunity to remotely assess structural damage immediately after a disaster,” he said. Khalil Sezenco-author of the paper and professor of structural engineering Civil Engineering, Environmental Engineering, and Geodetic Engineering at The Ohio State University.
That said, their algorithm will likely be used in conjunction with other emergency and resource planning tools, Hata said, noting that with more detailed experiments, the model could help predict levels of destruction in other earthquake-prone environments, such as Japan and California.
“There is still a lot of work to be done to embrace the kind of perspective that AI provides,” Qin says. “But the more quality data we have, the faster we can reach our goals.”
Co-authors include Ningli Xu, Abdullah Türer and Abdulmajeed Batarfi of The Ohio State University, and Hessah Albanwan of Kuwait University. This research was supported by the Turkiye Scientific and Technical Research Council, the Ministry of Environment, Urbanization and Climate Change of the Republic of Turkiye, the Information Advanced Research Projects Activity (IARPA) and the Naval Research Office.
