Twenty Minutes to 96% Accuracy
How Kansas City is Outpacing the Storm
After a cataclysmic storm strikes, a clock begins to tick. For city officials, the first few hours are a frantic race to evaluate impact, mobilize first responders, and secure the disaster designations required for state and federal aid.
In most cities today, this race is run on foot. Volunteers and first responders fan out across debris-strewn streets with clipboards and pens to manually record the wreckage. It is a heroic effort, but one tethered to a slow, analog process.
My Bellwether colleagues and I recently joined Kansas City’s Emergency Management team to see if we could help cities understand disaster damage in real-time, getting urgent help to those who need it faster. Alongside Andrew Ngui, Kansas City’s Chief Digital Officer, we put our geospatial AI to the test in a high-stakes simulation to see if we could turn a days-long manual slog into a 20-minute digital sprint.
A team of volunteers surveyed the Kansas City neighborhood on foot
The Data Delay in Disaster Recovery
At Bellwether, our mission is to build the first prediction engine for the Earth. We train our models on massive Earth observation datasets, enabling the system to "see" granular changes to the physical world—from bridges washed out by floods to a damaged structural integrity of a roof—in a matter of seconds. Part of the work we do is devoted to helping communities understand the extent of damage caused by a weather emergency, so that they can respond and recover faster.
In March, Kansas City invited us to participate in a neighborhood-wide disaster response simulation. The City’s Emergency Management team randomly designated specific homes as "damaged" in a hypothetical storm and set up a head-to-head race.
The question was simple: Which damage assessment approach is faster, more accurate, and more efficient? A team of 60 volunteers surveying the neighborhood on foot, or the Bellwether platform analyzing aerial imagery?
The Result: Twenty Minutes to Clarity
Kansas City’s Fire Department flew a single drone to capture overhead imagery of 407 homes. Kansas City’s Emergency Management team then randomly swapped some of those images with generated images of the structures in various states of damage from a tornado. After the images were uploaded, Bellwether’s AI-powered geospatial tools went to work, identifying structural anomalies that indicated damage. After a 25-minute drone flight, our system analyzed the entire neighborhood and calculated damage in just 20 minutes, with 96% accuracy. The volunteer team required a full afternoon to finish the same assessment, at similar levels of accuracy. In the wake of a disaster, when every moment counts, the difference between 20 minutes and four hours can literally change lives.
Bellwether’s AI-powered geospatial tools identify structural anomalies that indicate damage
A Blueprint for Resilience
Kansas City is proving that technology can help emergency responders focus on what matters—helping the homes and families that need them most, rather than manually sorting and labeling data.
This work builds on Bellwether’s ongoing collaboration with the National Guard and Civil Air Patrol, which use our tools to help first responders locate damage to critical infrastructure after extreme weather events like hurricanes and flood zones. Our goal is to take the lessons learned in Kansas City and create a blueprint for every community facing the increasing threat of extreme weather.
If your community is interested in exploring how to improve your disaster response strategy, we want to hear from you. Send us an email at hello-bellwether@google.com.