University of Miami researchers used NASA derived hyperspectral imaging and AI to detect shallow water unexploded ordnance in the Florida Keys in a peer reviewed demonstration.
An airborne imaging system that pairs NASA-derived hyperspectral sensors with AI can now spot corroded shallow-water unexploded ordnance (UXO) from a single drone pass, without putting divers in the water. The first peer-reviewed demonstration, over the Florida Keys, points to a faster way for ports and coastal agencies to refresh hazard maps that in many places have not been updated since the conflicts that seeded them.
The technique was published this month in Frontiers in Marine Science by a team led by Ved Chirayath, the Vetlesen Endowed Chair of Earth Sciences at the University of Miami Rosenstiel School. Chirayath, also affiliated with NASA Ames through his work on airborne remote sensing, combines two technologies his group has refined for ocean imaging: MiDAR (Multispectral Imaging, Detection, and Active Reflectance) and Fluid Lensing, an ACES/UMiami-developed method that uses wave motion itself as a focusing element to see through a disturbed water surface.
The hard part of airborne UXO mapping is not the sensor. It is the ocean. Waves act as moving lenses, scrambling the signal from anything beneath them. Fluid Lensing inverts that problem, treating the wave field as a known distortion that can be computationally removed. A hyperspectral camera records the same scene across hundreds of narrow wavelength bands, and a trained model looks for the spectral fingerprints of corroded metal, sediment disturbance, and the algae and sand buildups that bury a 50-year-old shell. The Rosenstiel announcement describes the work as the first operational test on live UXO targets.
Independent coverage, including a trade-media piece in GeoConnexion, presents the demonstration as a starting point rather than a finished system. The Florida Keys flight gave the team a controlled environment to validate the detection pipeline. The open question is whether the same pipeline holds in murkier Atlantic channels, deeper Pacific bays, and surf zones where sand migration and biofouling can shift the spectral signature within a season.
That open question is the structural reason this story matters. Many national UXO charts in shallow water still reflect post-conflict surveys from the 1940s through the 1970s, drawn up by divers working piece by piece over years. Those surveys are not wrong, but the seabed has moved. Tides have shifted sand over devices, marine algae have grown over hulls, and in some cases whole munitions have been relocated by storms. The bottleneck has been logistics: redoing the work at the same fidelity, across thousands of kilometers of contested coastline, has been too expensive and too slow to be routine.
An airborne technique that completes a survey in a single flight changes that calculation, even before the algorithm is fully generalized. For ports, fisheries ministries, and the coast agencies that publish public hazard charts, the relevant shift is operational. Routine refresh cycles become plausible, and the next storm does not have to wait for the next decade's survey ship.
There are limits worth naming. The published detection performance is bounded to the conditions in the Florida Keys test. Sand migration, biofouling, and changes in water clarity are explicitly called out in the paper as variables the team has not yet characterized across sites. The system is also not a clearance tool. Locating a candidate munition is the first step; identifying, exposing, and disposing of it still requires human teams on or near the water.
What changes now is the pace at which shallow-water hazard maps can be checked against reality. The Rosenstiel team says broader-environment validation is the next phase. Whether the approach scales into a routine coastal-monitoring service, rather than a research demonstration, will depend on how those murkier and deeper sites respond to the same pipeline.