New AI tool developed by Ohio State and Kuwait University researchers can assess earthquake damage using drone photos, which could help first responders make critical rescue decisions, study finds.
A new artificial intelligence tool that can imagine what earthquake damage looks like from the street, using only drone photos taken from above, could give first responders and planners a powerful new way to save lives and rebuild smarter.
Engineers at The Ohio State University, in collaboration with Kuwait University, have developed a model called the LoRA-Enhanced Ground-view Generation (LEGG) diffusion model, which uses aerial images to create highly realistic, three-dimensional reconstructions of buildings and streets. The system is designed to predict where structures are cracked, tilted or partially collapsed — even in dense city neighborhoods where damage can be hard to see from the sky alone.
The work, published in the International Journal of Remote Sensing, aims to close a critical information gap that often slows rescue and recovery after major earthquakes.
Traditional surveys rely heavily on drones, satellites or lidar to scan damage from above. Those tools are essential, but they miss something important: people on the ground make life-or-death decisions based on what they can see at street level. Manually collecting that information can take days or weeks, especially when roads are blocked or buildings are unsafe to enter.
The team set out to bridge that divide.
Co-corresponding Rongjun Qin, a professor of civil, environmental and geodetic engineering at Ohio State, explained that the AI is trained to link what a building looks like from the air with how it appears from the ground.
“What our algorithm does is generate thousands of pairs of semi-realistic photos of what a building looks like on the top and from the ground,” Qin said in a news release. “Having such data is vital, as drones gather important information from above, but people actually make emergency decisions from ground-level views.”
The researchers built LEGG using real drone imagery and ground photos so the model could learn complex visual patterns — the subtle cues that signal structural distress. Once trained, the AI can take new aerial images and synthesize photorealistic street-level scenes that highlight likely damage.
In effect, it gives emergency teams a way to see around corners and into streets they cannot immediately reach.
To test the approach, the team applied its framework to the 2023 Kahramanmaras earthquake in Turkey, a 7.8 magnitude disaster that destroyed hundreds of thousands of buildings and damaged many more. They compared drone imagery from 2015 with photos taken in the days after the quake, capturing dramatic changes in the built environment, from collapsed structures to temporary shelters in open spaces.
After being shown a dataset of only 3,000 city structures from the affected area, the AI was able to generate street-level images that sharpened the recognition of key problems, including façade cracks, leaning buildings and partial collapses. That performance suggested the model could extract fine-grained clues from limited data and turn them into high-resolution, realistic ground views.
The strength of the system comes from combining perspectives. By feeding the model both aerial and ground imagery during training, the researchers gave it a strong starting point for inferring how damage at roof level or in building outlines might translate into what a person would see standing on the street.
Qin noted that makes the tool especially promising for time-critical missions.
“This simulation is essentially a map, but an experienced and well-trained AI could offer an additional supply of information that would be really helpful for emergency crews in making quick decisions about where to go when the clock is ticking,” he said.
In large earthquakes, that kind of guidance can be crucial. Search-and-rescue teams must decide where to send limited crews, heavy equipment and medical support. A system that flags likely hotspots of severe damage, or identifies buildings that are leaning or at risk of collapse, could help prioritize neighborhoods and routes.
Beyond immediate response, the researchers see broader potential. The same framework could be used to study how different types of buildings and city layouts fare in major quakes, informing building codes, retrofits and land-use planning in earthquake-prone regions such as Japan, California or Turkey.
“As long as you have good data, AI can serve as a very generous predictor of past and future outcomes,” Qin added. “It’s a tool that can be incredibly helpful.”
Co-author Halil Sezen, a professor of structural engineering in civil, environmental and geodetic engineering at Ohio State, emphasized the value for engineers and policymakers who may be far from the disaster zone.
“This work presents a great opportunity for engineers and other decision makers to remotely assess the damage in structures soon after a disaster,” Sezen said in the news release.
In practice, the team expects their algorithm to be used alongside other emergency and resource-planning tools, not as a stand-alone solution. Ground crews, satellite data, sensor networks and local knowledge will still be essential. But AI-generated ground views could become a new layer of information that ties those pieces together.
The researchers also stress that the model’s performance depends heavily on the quality and diversity of the data it is trained on. To make the system robust across different cities and building styles, they say more experiments and larger datasets are needed, including imagery from other regions and types of disasters.
“There is still a lot of work to be done to bring in the kind of perspective AI offers,” Qin added. “But the more good quality data that we have, the faster we’re going to achieve our goals.”
Looking ahead, the team hopes their approach will inspire governments, aid organizations and engineers to rethink how they prepare for and respond to earthquakes. By fusing imaginative AI with real-world imagery, they envision a future in which communities can assess damage faster, design more resilient infrastructure and, ultimately, reduce the human toll when the ground starts to shake.
Source: Ohio State University