A satellite RF feed usually arrives in a GIS the way most feeds do, as a point layer with timestamps and a handful of attribute columns. The container is convenient, but extracting value is possible only with good context.
What the receiver actually recorded
Spaceborne RF geolocation is passive. The satellites carry wideband software-defined receivers and listen for energy radiated by transmitters within line of sight: marine radar, VHF sets, SATCOM uplinks, navigation and surveillance radars, and GNSS jammers among them. The satellites send nothing of their own.
Two architectures are in commercial service, and they differ in how many spacecraft a detection requires. Difference-of-arrival methods compare the same emission across two or more separated receivers with synchronized clocks, which is why some operators fly their satellites in tight clusters. Direction-finding works from a single spacecraft, whose antenna array measures the bearing the signal arrived on and takes successive bearings along the pass.
What a record of RF data contains
Both routes end at the same deliverable, and it arrives in numerical table format. A record of RF data carries the estimated position, the carrier frequency and bandwidth, the pulse characteristics that make up the emitter’s signature, and a confidence grade of high, medium, or low, all of it timestamped.
Those columns are what a query can reach, which is why RF data drops into an existing workflow more easily than its reputation suggests. Teams who want it beside the imagery that confirms it can purchase RF data from Sfera Technologies alongside optical and SAR acquisitions over the same area of interest.
The uncertainty is measured in kilometers
Geometry decides how good that position estimate gets. When Space Flight Laboratory deployed HawkEye 360’s Cluster 11, attitude control and formation flying were the stated reasons the builder was chosen, because a difference-of-arrival fix depends on where the receivers sit relative to one another during the pass.
With a strong wide-bandwidth signal and satellites well spread across the sky, sub-kilometer accuracy is achievable in principle. Those conditions are simply uncommon for emitters that are not cooperating.
For non-cooperative transmitters the operational range runs from one to five kilometers. Poor geometry or a weak signal pushes errors into the tens of kilometers, or leaves no usable fix.
The only publicly available independent assessment of a commercial RF constellation comes from ESA’s eoPortal for the Unseenlabs BRO satellites, which geolocate by single-satellite direction of arrival. It records a mean error of 5.4 km, a median of 2.5 km, a standard deviation of 6.7 km, and a maximum expected error near 30 km.
Median, mean, and maximum geolocation error for single-satellite direction-of-arrival geolocation, drawn at a common scale. Source: ESA eoPortal assessment of the Unseenlabs BRO constellation, 2026. Own diagram.
Averaging several passes over the same persistent emitter tightens that figure, at the cost of waiting for the passes. For a workflow built around alerting, the waiting is the expensive part.
What each layer can answer
The practical consequence is that no single feed closes the question, and the layers a geospatial team already carries answer different halves of it.
| Question | RF detection | AIS or ADS-B | SAR or optical |
| Position | estimated, kilometer scale | reported, meter scale | imaged, meter scale |
| Identity | emitter fingerprint, no hull | declared by the vessel itself | inferred from hull, wake, and size |
| Target stops reporting | still detected if it emits | invisible | still imaged |
| Night and cloud | unaffected | unaffected | SAR unaffected, optical blocked |
| Needs cooperation | no | yes | no |
Read across a row and the complementarity is obvious. Read down the identity column and the central limitation is too. RF says something is there, and the electromagnetic signature is distinctive enough to recognize the same emitter on a later pass, but the registered vessel behind that emitter has to come from somewhere else.
The confirmation step is a tasking order
An uncertainty radius of a few kilometers is awkward for a map symbol and quite convenient for an acquisition request. A tasking order wants an area of interest, and that is the shape an RF fix already produces.
The usual sequence runs from the fix to a correlation against reported traffic, then to imagery over whatever remains unexplained. Where a vessel has stopped transmitting on AIS, which the IMO mandates for most commercial shipping, the silence beside a live emission is the anomaly worth spending an acquisition on. SAR earns its place at that step because the answer cannot wait for a clear sky.
Cross-validation also guards against a failure mode that appears in exactly the places operators care about. In sparsely trafficked waters, false detections make up a larger share of the total, and one unconfirmed detection can lead to a waste of resources in pursuit of the wrong target. Remote areas compound this, since the charts underneath are often thin as well, a gap GISuser covered when Fugro resurveyed Alaskan approaches for NOAA.
What the constellation will not hear
Detection depends on radiated power, signal duration, spectral shape, and orbital geometry. Several categories of emission fall outside that envelope, and no amount of tasking recovers them:
- Direct-sequence spread-spectrum signals, whose power is deliberately spread below the noise floor and cannot be picked out without the spreading code
- Fast frequency-hopping waveforms, where the dwell time per hop is shorter than the receiver needs to integrate
- Low-power IoT transmitters such as LoRa and NB-IoT, which radiate well under a watt and stay below the threshold of a wideband survey receiver, even though purpose-built gateways do pick up the same protocols over a cooperative link
- Emissions timed against known overpasses, or suppressed entirely, which is a straightforward countermeasure for anyone who expects to be watched
- Message content of any kind, since commercial services are built and licensed to detect and characterize, with interception left to government systems
That last boundary is a product decision as much as a regulatory one, and it is worth stating plainly to anyone who arrives expecting government-grade signals intelligence.
Holding it in a GIS
Most of the damage happens at import. A fix stored as a point, symbolized as a point, and joined to other layers by nearest neighbor has lost the one attribute that mattered.
Keeping the reported uncertainty as a field, and drawing the detection as an area, costs one column and changes what analysts conclude from the same file. The join to AIS or ADS-B should be spatial and temporal together, with a window wide enough to cover both the error radius and the reporting interval.
And the attribute worth carrying furthest is the signal description. It survives a bad position, which the coordinate cannot, and it is what lets the same emitter be recognized on a later pass.
RF detection is a genuinely useful cue, and treating it as a fix is the fastest way to waste it.