LOC8 Plus Skydio Run Coconino’s AI Search Engine

The Coconino County Sheriff’s Office in Flagstaff, Arizona is using the latest AI-capable Skydio drones with the LOC8 G2 image-analysis platform to compress search and rescue photo review from thousands of drone images down to about twenty per mission.

The agency also works with a Scotland-based company to build environment-specific AI models for Northern Arizona terrain. Deputy Paul Clifton, Assistant Search and Rescue Coordinator and UAS Program Manager, leads the program, profiled in AZ Family coverage published May 20, 2026.

How The Workflow Actually Changes

The bottleneck in drone-assisted search and rescue has historically not been the flight. It has been the post-flight review of thousands of high-resolution images by human eyes, a task that fatigues searchers within about ten minutes and degrades accuracy from there.

Deputy Paul Clifton explained the math directly: “As opposed to looking at thousands of photographs with the human eye that fatigues in 10 minutes, we’re leveraging automation tools.” The result he cited is plain: “Maybe it reduces the number of photos you have to look at from a thousand down to twenty.”

That ratio is the entire operational thesis. A drone flies a defined grid, captures imagery at known overlap, and the AI filters the set down to the small number of frames that contain probable anomalies.

The pilot and the search coordinator review the short list. The hours saved go directly into either covering more ground or starting a ground team faster on the actual target location.

The Skydio AI Layer

The Sheriff’s Office confirmed in interview that the unit now flies the latest AI-capable Skydio drones. Coconino’s program started in 2018 with two DJI Mavic 2 Enterprise units, and the migration to Skydio reflects the agency’s preference for on-board autonomy in canyon and forest terrain where a manually flown drone is a liability for a pilot watching a screen.

The current generation of Skydio enterprise hardware (the X10 line) carries six navigation cameras around the airframe for 360-degree obstacle avoidance, AI-driven autonomous flight in complete darkness through the NightSense system, and a Teledyne FLIR Boson+ thermal sensor on the X10 variant. That stack handles the in-flight half of the search problem.

The Coconino terrain rewards autonomy. Deep canyons in the Coconino National Forest, sudden elevation changes around the San Francisco Peaks, and dense ponderosa pine canopy are exactly the conditions where a tethered radio link and a tired pilot produce missed coverage.

The LOC8 G2 Image-Analysis Engine

As Arizona’s Family reported, LOC8 is the ground-side half of the workflow. The software was released in 2019 by Loc8 LLC, a subsidiary of Unmanned Systems Research, and its current version is LOC8 G2 with a redesigned interface and a wideband search engine.

The platform is co-founded by Anthony Lockly (CEO), PJ Kirkpatrick, Peter Menet, and Gene Robinson, with development originating from a real-world failure: a missing-person search where 1,500 drone images were reviewed by human eyes and the victim was missed, even though the subject was captured in several frames and only obscured by evergreen brush.

The patent-pending algorithm scans imagery at the individual pixel level for user-defined color clusters. A single pixel of the target color is enough to flag the frame. The software runs against drone, manned-aircraft, and terrestrial-camera footage and is sold on subscription from $45 to $70 per month, which puts it inside the budget of a single county SAR unit without state funding.

The most documented case study of LOC8 G2 in practice came out of Moose Lake, Minnesota in 2024, when an older man with dementia went missing. The drone flew the search grid, LOC8 G2 scanned the imagery for the red color of the man’s clothing, and the system flagged a grouping of red pixels as an area of interest within ninety minutes. Closer analysis confirmed it was the missing man.

That is the workflow Coconino is now running on a 140-to-150-mission-per-year cadence.

The Scottish Partnership

The third piece is the part that the AZ Family segment touched on but did not fully unpack. Coconino has a partnership with a Scotland-based company to develop environment-specific machine learning models tuned to the Northern Arizona landscape.

Clifton named the reason: “Northern Arizona versus Southern Arizona, the context, the environments are very different.” A generic missing-person color model trained on Scottish moorland is not going to perform the same in pine forest at 7,000 feet of elevation with ponderosa shadows and red rock substrate. The Scottish partner provides the model-training pipeline. Coconino provides the local terrain data.

The exact identity of the Scottish partner was not specified in the AZ Family coverage. Scotland is a recognized hub for SAR drone AI work, with the University of Glasgow and the SARDO project (developed by British Mountain Rescue volunteers) both publishing in adjacent areas in 2024 and 2025.

Coconino’s SAR Footprint

Coconino County is one of the largest counties in the United States by area, covering portions of the Grand Canyon, the Coconino National Forest, Sunset Crater, Wupatki National Monument, and large stretches of federally managed wilderness. The Sheriff’s SAR unit runs roughly 140 to 150 missions per year, with about 90 percent on federally managed public lands.

That mission volume matters for the technology decision. A SAR unit with five missions per year cannot justify the operational learning curve of a multi-vendor AI stack. A unit running three missions per week, in terrain that defeats ground patrol routinely, can.

DroneXL’s Take

No sugarcoating this, the Coconino program is the cleanest example in 2026 of what a layered AI SAR workflow actually looks like in practice. On-board autonomy from Skydio in the air, pixel-level filtering from LOC8 G2 on the ground, and terrain-specific model training from a Scottish partner stacked on top of both.

Three observations matter. First, the bottleneck in drone SAR has shifted from flight time and thermal optics to image-review labor. The 1,000-to-20 ratio that Clifton cited is the kind of operational gain that justifies a full procurement rework, not just a hardware upgrade.

Second, LOC8 at $45 to $70 per month is the most interesting number in the entire story. That price puts forensic image analysis inside the budget of a single county SAR unit without state funding. The historical barrier to AI SAR has been software cost. LOC8 has effectively removed it for departments willing to learn the workflow.

Third, the migration path Coconino walked is the template most US sheriff offices should study: start with DJI Mavic Enterprise hardware in 2018 for initial drone operations, build pilot competence and procedures, then graduate to Skydio AI-capable airframes once on-board autonomy becomes the bottleneck. Skipping the cheap-hardware learning year and going directly to a Skydio fleet is how programs fail before they fly.

Watch the Northern Arizona terrain models that come out of the Scottish partnership. Once they exist, they will likely become commercially available, which lowers the barrier for any other Western US department to skip the partnership step entirely and just license the trained models.

Photo credit: Coconino PD, Concord Police Department Facebook.