Submarine snorkel mast and periscope detection using SAR
High-resolution spotlight SAR can detect the faint radar returns of submarine snorkel masts and periscopes, but sea clutter at moderate sea states makes discrimination genuinely hard. This page explains the physics, the processing chain, and the honest limits of what satellite SAR can and cannot do.
Sensors
- Capella Space SAR (X-band): Spotlight mode delivers ground-range resolution of approximately 0.5 m, which is sufficient to resolve the physical extent of a snorkel mast head. Revisit over a fixed point depends on constellation size; with eight or more satellites, sub-daily revisit over priority areas is achievable. Single-pass coherent processing supports CFAR-based point-target detection.
- ICEYE SAR (X-band): Spotlight Ultra mode achieves approximately 0.25 m azimuth resolution, making it among the finest commercially available. Very short dwell times reduce motion blur from wave action. ICEYE publishes latency figures of under one hour from collection to delivery, which matters when cueing other assets.
- RADARSAT Constellation Mission (C-band): Three-satellite Canadian government constellation with a revisit of up to four days at mid-latitudes in standard modes, improving with wider swath selections. C-band (5.4 GHz) penetrates rain better than X-band but suffers higher sea-clutter return at the same sea state, which is a meaningful disadvantage for sub-metre targets. Quad-polarisation (HH, HV, VH, VV) modes support polarimetric decomposition, which is the main compensating advantage.
- Sentinel-1 (C-band, ESA): Free, open-access archive back to 2014. Resolution in Interferometric Wide Swath mode is 5 x 20 m, which is far too coarse to detect a snorkel mast directly. Useful for contextual ocean-state characterisation and for calibrating clutter models that inform higher-resolution tasking decisions. Not a detection sensor for this application.
- Umbra SAR (X-band): Commercial spotlight mode with published resolution to 0.16 m in some collection geometries, making it the finest commercially available at time of writing. Coherent change detection between passes is a documented capability. Small target detection in high-clutter environments benefits from this resolution margin.
What a snorkel mast actually looks like to a radar
A diesel-electric submarine running on its snorkel presents a radar cross-section (RCS) that published research places in the range of roughly 0.1 to 10 square metres depending on mast geometry, aspect angle, and sea state. That is not nothing, but it is comparable to the return from a breaking wave crest or a small piece of flotsam. The detection problem is not sensitivity. Modern SAR systems are sensitive enough. The problem is discrimination: separating one genuine point target from thousands of clutter candidates in a single image.
At X-band (around 9.6 GHz), the Bragg-resonant wavelength for sea surface scattering is short, which means even a sea state 2 to 3 surface produces a textured, high-variance clutter background. The snorkel mast return is a bright point target, but so is every whitecap that happens to align with the radar look direction. C-band (5.4 GHz) has a longer Bragg resonance and generally lower normalised radar cross-section (NRCS) for the sea surface, but the mast's own RCS also tends to be lower at C-band because the wavelength is closer to the mast's physical dimensions, reducing corner-reflector effects.
CFAR detection and why the threshold is never comfortable
The standard detection algorithm for point targets in SAR imagery is Constant False Alarm Rate (CFAR) processing. A sliding window estimates local clutter statistics around each candidate pixel; the target is declared if its return exceeds the local mean by a factor set to hold false-alarm probability at an acceptable level. For ship detection in open ocean, this works well because large vessels return tens of thousands of square metres of RCS. For a snorkel mast, the signal-to-clutter ratio (SCR) may be only a few decibels, and the CFAR threshold that keeps false alarms manageable will also suppress genuine detections.
Multi-look processing averages independent looks at the same scene to reduce speckle variance, which improves SCR at the cost of spatial resolution. For a target this small, that trade is painful. Sub-metre spotlight imagery from Capella or ICEYE preserves the point-target peak while reducing the clutter floor through averaging, which is why very high resolution is not merely a nice feature here but a functional requirement. Published studies in journals such as IEEE Transactions on Geoscience and Remote Sensing have explored adaptive CFAR variants, including cell-averaging CFAR, ordered-statistic CFAR, and Weibull-distribution clutter models, specifically for small maritime targets. None of them make the problem easy. They make it tractable under favourable conditions.
What polarimetry adds, and what it cannot fix
A snorkel mast is a metallic, roughly cylindrical object. Its polarimetric scattering behaviour differs from that of a distributed sea-surface clutter patch. In dual- or quad-polarisation SAR, the ratio of cross-polarised to co-polarised return (HV/HH or VH/VV) tends to be lower for a specular metallic point target than for a breaking wave, which has a more complex depolarising structure. Polarimetric decomposition methods, including the Pauli decomposition and entropy-alpha analysis documented in the open literature, can therefore provide a discriminant that intensity alone cannot.
The honest limit is that this advantage shrinks as sea state rises. At sea state 4 and above, breaking waves generate strong cross-polarised returns that overlap with the target's polarimetric signature. Incidence angle matters too: shallow incidence angles (below about 20 degrees) increase sea-surface return and reduce the polarimetric contrast. Most commercial SAR satellites collect at incidence angles between 20 and 45 degrees, which is workable but not uniformly ideal. Polarimetric capability is also not universal across commercial tasking options; ICEYE and Capella currently offer primarily single or dual polarisation in their highest-resolution modes.
The geometry problem: where to look
Even if detection algorithms are well-tuned, satellite SAR cannot stare. A single pass covers a strip and moves on. A submarine transiting at four to six knots covers roughly 7 to 11 km per hour. If the revisit interval is twelve hours, the submarine's possible position has expanded to an ellipse covering tens of thousands of square kilometres. Satellite SAR for snorkel detection is therefore a cueing and confirmation tool, not a persistent surveillance system.
Effective use requires prior intelligence to constrain the search area: acoustic data, patrol boundaries, known transit routes, or tip-off from other sensors. The satellite then provides a time-stamped, geo-referenced detection that can be passed to maritime patrol aircraft or surface assets. The value is in the precision of the fix and the covertness of the collection, not in the revisit rate. A government client ordering this analytic needs to understand that the satellite component of the kill chain is the confirmation step, not the search step.
Operational sensitivity and the open literature
This application sits at the edge of what commercial satellite operators will discuss publicly. Detection of submarine snorkel masts using airborne SAR has been a documented research area since at least the 1990s, with work published in IEEE and MDPI Remote Sensing journals covering CFAR variants, polarimetric discriminants, and clutter modelling. The extension to satellite SAR at sub-metre resolution is a natural continuation of that literature and is openly discussed in academic contexts.
Collection tasking for this purpose is subject to export controls, end-user licensing, and in some jurisdictions specific government authorisation. Satellize operates as a non-aligned analytics provider and structures engagements through appropriate licensing frameworks. Clients with sovereign mandates should expect a structured discussion of collection authority before any tasking is placed. The analytic methods themselves, CFAR processing, polarimetric decomposition, and coherent change detection, rest on published, peer-reviewed foundations and are not proprietary.
Honest performance envelope
Under favourable conditions, meaning sea state 2 or below, X-band spotlight SAR at 0.5 m or finer resolution, incidence angle between 25 and 40 degrees, and a mast presenting a broadside aspect, detection probability in the published literature reaches into the 70 to 90 percent range with false-alarm rates that are operationally manageable. At sea state 4, those figures degrade substantially and no published method reliably recovers them without additional information.
Archive depth for commercial X-band spotlight imagery is limited. ICEYE has operated since 2018; Capella since 2020. Neither constellation was tasked for speculative snorkel detection, so historical coverage of specific areas of interest is sparse. Coherent change detection, which compares two passes of the same geometry to identify new point targets, requires repeat collections planned in advance. This is a forward-looking analytic, not a retrospective one.
Typical figures
| Spatial resolution (detection mode) | 0.16 to 0.5 m (Umbra, ICEYE, Capella spotlight); 3 m (ICEYE Spot); C-band systems 1 to 3 m in fine modes |
| Radar frequency | X-band: ~9.6 GHz (Capella, ICEYE, Umbra); C-band: ~5.4 GHz (RADARSAT-C, Sentinel-1) |
| Polarisation options | Single (HH or VV), dual (HH+HV or VV+VH), quad (HH/HV/VH/VV on RADARSAT-C); highest-resolution commercial modes typically single or dual |
| Minimum detectable RCS (favourable conditions) | Approximately 0.1 to 1 m² at sub-metre resolution, sea state ≤2; rises sharply with sea state |
| Revisit (priority area) | Sub-daily achievable with multi-operator tasking; single-operator revisit 6 to 24 hours depending on constellation and latitude |
| Latency (collection to delivery) | Under 1 hour for ICEYE direct downlink; 2 to 6 hours typical for processed GEC/GRD product |
| Swath width (spotlight mode) | 5 to 10 km typical; limits area coverage per pass and requires prior cueing to a constrained search zone |
| Archive depth | Commercial X-band spotlight: 2018 (ICEYE) to 2020 (Capella) onwards; tasked coverage of specific areas is sparse prior to a named programme |
| Sea-state operating limit | Detection probability degrades significantly above sea state 3; sea state 4+ makes discrimination unreliable with current published methods |
| Delivery formats | GEC (Geo-Ellipsoid Corrected) GeoTIFF, SICD complex imagery, NITF; detection reports as GeoJSON or shapefile |
Analytics Satellize can run
| Point-target detection layer | Ordered-statistic CFAR on calibrated sigma-nought imagery with Weibull clutter distribution fit | GeoJSON alert layer with candidate detections, RCS estimate, local SCR, and confidence score per detection |
| Polarimetric discriminant score | Pauli decomposition and entropy-alpha analysis on dual- or quad-pol collections to separate metallic point targets from sea clutter | Per-detection polarimetric feature vector appended to detection GeoJSON; analyst-readable PDF summary |
| Coherent change detection (CCD) report | Complex coherence comparison between two same-geometry passes; new point targets appear as coherence drops in a stable background | Change map GeoTIFF with new-target candidates flagged; requires pre-planned repeat collection geometry |
| Clutter model and threshold advisory | Sea-state estimation from Sentinel-1 NRCS and ECMWF wind fields; CFAR threshold pre-computation for planned collection windows | Collection-window recommendation report with predicted SCR range and recommended incidence angle |
| Multi-pass track hypothesis | Kinematic association of detections across multiple passes using constant-velocity motion model bounded by submarine speed envelope | Track hypothesis shapefile with position uncertainty ellipses and estimated heading/speed range |
| False-alarm triage report | Contextual filtering using AIS vessel positions, wave-height model, and wind-shadow geometry to suppress likely non-target detections | Filtered detection list with triage rationale per candidate; delivered within agreed latency window |
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.