● Toolkit
Most firms in this field will not tell you how they reach a conclusion. We will, in detail, including the data each tool is built on and the question it cannot answer. A method you are not allowed to inspect is a method you have no way to check.
Public airborne lidar is accurate to a few centimetres and almost nobody uses it for search work. We do.
We hydrologically fill the bare-earth surface across a corridor and subtract the original ground. What is left is every closed basin, measured: how deep it is, how far its floor sits below the road grade, how much canopy or water is over it, and how far it lies from any road that people actually use.
Two exclusions matter as much as the ranking. Basins in the median, the shoulder ditch and interchange infields are thrown out — a vehicle there is found in the first hour, so it cannot explain a search that has run for weeks. And standing water is handled separately, because lidar returns a flat water surface: a pond has zero fill depth and is invisible to depression analysis unless you go and get it from the hydrography layer.
“We searched that ridge” is almost never true. We take the track a team actually walked and compute, from the terrain, what was visible from every point they stood — then report the fraction they could not see.
In steep ground the numbers are stark. A test line covering 101 hectares on a map had real sight of 19 of them. The other 81% was ground the team walked past and could not have seen into.
A sensor that did not fire is evidence. Handled properly, silence across a network is one of the sharpest constraints available in an open-source case.
We map known reader and camera positions along the relevant routes, identify the unmonitored stretches, and work out which of them are consistent with the timeline. If a vehicle passes one reader and never reaches the next, the ground between them becomes the highest-value corridor in the case.
We model path loss from known mast positions over the real ground — earth curvature at the standard refraction allowance, and a diffraction term so that “just behind a ridge” is distinguished from “deep in a hole”. The output separates served ground from ground where a handset genuinely could not reach a tower.
This is the difference between the phone was switched off and the phone went somewhere it could not transmit from. It is also honest about its own limits: run over flat country it will correctly tell you terrain explains nothing, which is a useful answer in itself.
When an agency obtains carrier records, a registration is usually read as “somewhere in this sector” — a wedge kilometres across. Three constraints tighten it: the antenna azimuth and beamwidth, the timing-advance value which is a distance ring, and terrain, because ground with no path to that antenna could not have registered on it. Intersecting all three routinely reduces a sector to a few tens of hectares.
From an entry point we trace the flow network downstream and rank the places transport tends to stop: slack-water reaches, the inside of confluences, and the entry to any standing water. Gradient is judged against the stream’s own profile rather than a fixed threshold — on a coastal-plain creek every metre looks flat, and in a mountain river none of it does.
The tools above each produce a ranking. Handing a commander seven of them is not help. We combine them into a single probability distribution over the ground and update it as the search proceeds.
This is the standard framework for search under uncertainty — the one used to locate the USS Scorpion and, after two failed seasons, the Air France 447 wreckage. A prior is built by simulating how a person of that category moves over the actual terrain. Each sensor that did not fire, and each area already searched, updates the surface by the probability it would have produced a detection.
The output is the thing an incident commander can actually use: what share of the probability has been eliminated, where the remainder now sits, and how few hectares it has been squeezed into. From that we can also plan aerial tasking that buys the most probability per battery, rather than flying a grid over ground worth nothing.
A meaningful number of missing people have already been located — in another county, another state — and nobody ever connected the two records. We score every missing-person record against every unidentified-decedent record on sex, the age the person could have been when they died, height, build, recorded ethnicity, and the distance between where they vanished and where the remains were found. It runs against the national set, not one case, and re-runs when new records are published.
Endangered adults routinely surface alive in a jail, an emergency department or a shelter under no name at all, and nobody checks twice. We run a structured sweep of every custody, health and shelter system for the relevant jurisdictions, on a clock, and re-run it — a jail roster checked a fortnight ago tells you nothing about today.
Most unidentified-person cases now resolve through genealogy rather than a database hit. We support that work — match lists, relationship bands, tree building and, just as importantly, a record of which branches have been eliminated.
The most common unforced error in a missing-persons case is that the data expired before anyone asked for it. Every custodian holding something perishable — a business recorder, a carrier, a bank, a toll authority, a dispatch centre — gets a row, an expected retention window and a countdown, with a preservation request ready to send.
Some of those windows are far shorter than people assume. Message content at a carrier is measured in days. A small-business recorder often holds a fortnight. By the time a case feels serious enough to escalate, several of these have already run out.
All public, all lawful, all citable. Currency matters as much as accuracy: elevation may have been flown years before a disappearance, and ground changes.
| Source | Used for | Note |
|---|---|---|
| USGS 3DEP | Bare-earth elevation for terrain, sightline and radio modelling | Lidar-derived; flight dates vary by county and can predate a case by years |
| USGS National Hydrography | Streams, ponds and wetland extents | Full-resolution layers |
| MRLC / NLCD | Land cover and canopy screening | Updated on a multi-year cycle |
| OpenStreetMap | Road geometry, barriers, mast positions | Crowd-sourced — good coverage of roads, incomplete for infrastructure |
| NamUs | Missing, unidentified and unclaimed person records | US Department of Justice. Not affiliated with or endorsed by NamUs |
| Public camera registries | Reader and camera positions | Crowd-sourced and incomplete; a snapshot, not an inventory |
| Copernicus Sentinel-1 | Radar change detection — experimental | Under evaluation; not offered as a capability |
| Court, custody and vital records | Record sweeps and cross-matching | Availability varies by state and county |
Our working limits on conduct — what we will and will not do to obtain information — are set out separately in our standards.
There is no charge to ask, and we will tell you honestly whether open sources can still add anything to the case.
