Shipborne radar activity monitoring from orbit
Passive detection of X-band and S-band navigation radar emissions from LEO satellites can confirm ship presence and operational status independently of AIS, exposing vessels that are dark by choice rather than by accident.
Sensors
- HawkEye 360 RF constellation: A cluster-based LEO constellation using time-difference-of-arrival and frequency-difference-of-arrival (TDOA/FDOA) to geolocate emitters. Published geolocation accuracy is typically 1–5 km for cooperative emitters; maritime radar detection is a stated mission. Revisit depends on cluster passes, roughly 2–4 times per day over most latitudes, improving as the constellation grows toward its planned 20-plus satellite architecture.
- Kleos Space RF constellation: Operates in formation-flying clusters designed for TDOA geolocation of VHF through X-band signals. Kleos has published detection of maritime radar bands as part of its activity-based intelligence offering. Geolocation accuracy is broadly comparable to HawkEye 360 at 1–10 km depending on geometry and signal duration.
- Spire Global RF payload (LEMUR-2 satellites): Spire's multi-mission satellites carry software-defined radio payloads that can be tasked across a wide frequency range. The constellation exceeds 100 satellites, giving revisit rates well under 2 hours at mid-latitudes for signal-of-opportunity collection, though geolocation precision from a single satellite is limited without cluster geometry.
- Future sovereign RF payloads: Several national programmes are procuring hosted or dedicated RF monitoring payloads on LEO platforms. These typically combine wideband receivers with on-board signal characterisation, targeting X-band (9.2–9.5 GHz) and S-band (2.9–3.1 GHz) navigation radar windows specifically.
Why a navigation radar is a hard signal to hide
Commercial ship navigation radars transmit in two internationally allocated bands: X-band, centred near 9.3–9.5 GHz, and S-band at 2.9–3.1 GHz. Both are pulsed, high-power signals, typically 10–25 kilowatts peak for X-band magnetron sets on larger vessels. That is not a whisper. A rotating antenna broadcasting at those power levels is detectable from several hundred kilometres in free space, which is roughly the geometry a LEO satellite enjoys on each pass.
The fundamental physics is straightforward. The radar's signal propagates outward in all directions, not only toward the sea surface. A satellite passing overhead intercepts the upward lobe and sidelobes of that emission. Pulse repetition frequencies for common navigation radars run between roughly 500 and 3,000 pulses per second, and antenna rotation periods are typically 2–4 seconds. A satellite receiver that dwells over a vessel for even a few seconds will observe multiple rotation cycles and can characterise the signal's pulse width, repetition interval and carrier frequency.
The rotation-cycle problem and what it means for detection confidence
The intermittent nature of radar rotation introduces a detection probability that is less than one. When the antenna beam points away from the satellite's position in azimuth, the received power drops by the antenna's front-to-back ratio, which can be 20–30 dB on a directional array. A satellite pass lasting 60–90 seconds over a given point may catch only a handful of full rotation cycles. If the geometry is unlucky and the pass is short, a radar that is genuinely active could be missed.
This is an honest limit, not a caveat to be buried. Detection confidence improves with longer dwell, multiple passes, and cluster architectures that allow simultaneous reception from different look angles. It also improves when the vessel is stationary or slow, since a moving platform introduces Doppler shifts that help distinguish the radar emission from background interference, but the rotation-cycle gap remains a real source of false negatives.
Sea state adds a secondary complication. Heavy weather causes vessel motion that varies the antenna's elevation angle relative to the satellite. In extreme cases, wave-induced rolling can temporarily point the antenna's main lobe below the horizon, reducing upward radiated power. Practically, this effect is modest compared with the rotation-cycle problem, but it is worth noting for high-latitude winter operations.
Emitter fingerprinting: linking a pulse train to a vessel class
Every navigation radar has a characteristic pulse signature. Pulse width, repetition frequency, carrier frequency stability, and the specific modulation scheme (simple magnetron pulse versus frequency-modulated continuous wave) together form a fingerprint. Published work in the open literature, including studies in journals such as the MDPI Remote Sensing collection, has demonstrated that these parameters can be used to classify radar types and, by extension, narrow down vessel classes.
Magnetron-based radars, which remain common on fishing and smaller commercial vessels, have carrier frequencies that drift measurably with temperature and magnetron age. Solid-state broadband radars, increasingly fitted on newer vessels, have far more stable carriers but distinctive chirp patterns. A library of known signatures, built from controlled observations or from cross-referencing AIS-correlated detections, allows an analyst to move from 'a radar is present' to 'this radar is consistent with a particular class of fishing vessel or tanker'. That is not the same as identifying a specific IMO number, and it should not be presented as such without corroborating data.
Correlating RF detections with AIS gaps to flag intentional dark behaviour
The operational value of radar monitoring is greatest when it is fused with AIS data. A vessel that is transmitting AIS normally and also running its navigation radar is unremarkable. A vessel with no AIS signal but an active navigation radar is a detectable contradiction, and that contradiction is the product.
The workflow is: collect RF detections from a pass, geolocate each emitter using TDOA or FDOA, cross-reference the position against the AIS record for the same time window, and flag any emitter that has no corresponding AIS transmission within a reasonable position tolerance. The geolocation uncertainty of 1–5 km from current commercial RF constellations means that close-proximity vessels can be ambiguous, particularly in busy straits. At sea in open ocean, that ambiguity largely disappears.
A single RF detection of a dark radar is suggestive. A pattern of detections across multiple passes, consistent in position drift with a plausible vessel track, is operationally significant. This is why archive depth matters: HawkEye 360 has been collecting data since 2019, giving analysts a multi-year baseline against which anomalous behaviour can be assessed.
What this method cannot do
Passive RF monitoring from LEO does not produce imagery. It produces geolocated signal events with associated metadata. It cannot confirm cargo, crew, or flag state. Geolocation accuracy at 1–5 km is sufficient to cue a vessel-of-interest report but not to identify a specific ship in a dense anchorage. Vessels that switch off their navigation radar entirely, relying on lookout or GNSS chart overlay alone, are invisible to this method. Some operators in sensitive operations do exactly that.
Revisit is also a constraint. Even with growing commercial constellations, a given ocean area may be observed only a few times per day. A vessel that runs its radar only between passes, or that operates in a region with sparse satellite coverage, may evade detection for extended periods. Combining RF monitoring with SAR imagery and optical tasking is the more complete approach, and each method has its own gaps.
Putting it into an intelligence workflow
For a maritime domain awareness programme, shipborne radar detection is most useful as a cueing layer rather than a primary identification tool. An RF alert at a position with no AIS history triggers a tasking request for SAR or optical imagery. The imagery confirms vessel presence and, at sufficient resolution, type. The combination produces an evidence chain that is far harder to dismiss than any single sensor alone.
Satellize structures analytics pipelines of this kind for government clients, fusing open-constellation data with licensed commercial RF feeds. The Tonga crop-estimation programme is a different domain, but the underlying architecture, automated ingestion, anomaly detection against a baseline, and structured alert delivery, applies directly to maritime RF monitoring. The daily Overhead column covers cases where this kind of multi-layer fusion has produced publicly documented results.
Typical figures
| Target frequency bands | X-band 9.2–9.5 GHz (primary); S-band 2.9–3.1 GHz (secondary) |
| Geolocation accuracy (TDOA/FDOA cluster) | 1–5 km CEP for current commercial constellations; 5–10 km for single-satellite intercepts |
| Revisit rate | 2–4 passes per day per area for cluster-based systems (HawkEye 360, Kleos); sub-2-hour average for Spire's 100+ satellite constellation |
| Minimum detectable peak power | Commercially undisclosed; physics supports detection of signals above roughly 1 kW EIRP from LEO at ~500 km slant range under clear-sky conditions |
| Detection probability per pass | Less than 1.0 due to antenna rotation cycle; improves with dwell time and multi-satellite geometry |
| Signal characterisation outputs | Pulse width, pulse repetition frequency, carrier frequency, modulation class (magnetron vs. solid-state FMCW) |
| Archive depth | HawkEye 360 collecting since 2019; Kleos and Spire from 2020 onwards |
| Latency (detection to alert) | Typically 30–120 minutes post-pass for commercial RF analytics pipelines; near-real-time for some direct-downlink architectures |
| Coverage | Global, including beyond VHF-AIS range; polar coverage dependent on orbital inclination of specific constellation |
Analytics Satellize can run
| Dark vessel RF alert | TDOA/FDOA geolocation cross-referenced against AIS gap analysis | Timestamped position report with confidence radius, AIS status, and radar signature class; delivered as JSON feed or PDF alert |
| Vessel class probability estimate | Emitter fingerprinting against pulse-parameter library (pulse width, PRF, carrier stability) | Ranked list of probable vessel classes with supporting signal metadata, delivered as structured report |
| Persistent track reconstruction | Multi-pass RF detection sequence fitted to kinematic vessel track model | GIS layer (GeoJSON or shapefile) showing inferred vessel track with detection timestamps and position uncertainty ellipses |
| Area activity baseline and anomaly score | Statistical comparison of current RF detection density against historical archive for same area and season | Weekly or monthly heatmap report flagging areas of elevated or suppressed radar activity relative to baseline |
| SAR/optical cue list | RF detections filtered by dark-AIS status and geolocation confidence, ranked by anomaly score | Prioritised tasking list for SAR or optical collection, formatted for direct submission to commercial tasking portals |
| Fleet behaviour change detection | Time-series analysis of radar-on/radar-off patterns across a defined vessel population correlated with AIS history | Quarterly intelligence summary identifying fleets or routes where dark-radar behaviour has increased, with supporting evidence table |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.