SAR Polarization: How Many Channels Do You Need?

Radar polarization is the one specification that almost never makes it into a headline. Resolution does, revisit does, frequency does. Polarization sits a few lines further down the datasheet, and that placement misleads anyone buying radar for the first time.

What the four channels actually measure

A SAR sensor transmits a pulse whose electric field is oriented either horizontally (H) or vertically (V), and it can listen in either orientation as well. That gives four combinations: HH, VV, HV, and VH. The first two are co-polarized, the second two cross-polarized, and the difference is not cosmetic .

Co-polarized returns are dominated by surface scattering from bare soil, water, and smooth ground, together with the double-bounce returns that come off the corner geometry of buildings, bridges, and ship hulls. Cross-polarized returns come mostly from volume scattering inside a canopy , where the signal ricochets among leaves or air bubbles and arrives back with its polarization rotated.

The four polarization channels, arranged by transmit and receive orientation, and the scattering mechanism each one favors. Source: own diagram, built on the SAR fundamentals published by NASA Earthdata.

That is the entire physical basis for the word polarimetric. A sensor recording all four channels captures the full scattering matrix and can, in principle, tell you which mechanism produced a given pixel. A sensor recording one channel gives you brightness, and brightness alone says nothing about mechanism.

So four channels really are better than one. The open questions are what they cost you, and whether the target needs them .

The commercial fleet has dropped the full matrix

Line up today's commercial SAR imagery providers by band and channel count, which is what this side-by-side breakdown from observationdata.com does, and the pattern is hard to miss. Almost the entire fine-resolution fleet now flies one channel.

TerraSAR-X is the interesting case in that list. It sells single and dual polarization as standard products, and quad polarization exists only as an experimental mode outside the standard catalog , collected in a split-antenna configuration. The capability is real, but you cannot simply order it the way you order a Spotlight scene.

Which leaves RADARSAT-2 as the only sensor in that group selling HH, HV, VH, and VV in routine operations, and the polarimetry pages on this site still describe the mission in the terms it was designed around.

Full polarimetry has not disappeared from orbit, to be clear. It has moved to L-band and to agency programs: ALOS-2 flies quad-pol, and the Argentine-Italian SAOCOM pair offers single, dual, or quad polarization across its imaging modes. What has thinned out is quad-pol you can task commercially at fine resolution on short notice.

It would be easy to read that table as decline. It is closer to specialization: the X-band operators that launched over the past decade optimized hard for resolution and revisit, and polarimetry was what gave way .

What dual-pol already buys you

Add one cross-polarized channel to a co-polarized one and most of the contrast gain is already in hand. The step from two channels to four is much smaller than the step from one to two.

Sea ice classification is the standard example. HH separates open water from ice well enough on its own, and adding HV separates deformed multi-year ice from smoother first-year floes, because older ice full of air bubbles and rough surfaces throws far more volume scatter . The ice application pages here were built on exactly that behavior.

Crop discrimination works along the same lines. VV responds to the vertical stalk structure of cereals, VH picks up broader-leaved canopies, and the ratio between the two tracks growth stage across a season considerably better than either channel alone. Vessel detection benefits as well, since cross-pol suppresses sea clutter more aggressively and lifts small hulls clear of the background.

None of that needs four channels. It needs two channels, chosen sensibly .

What the full matrix costs

Quad-pol is not free, and the bill arrives as coverage . To collect all four combinations, the radar alternates its transmit polarization pulse by pulse, which doubles the pulse repetition frequency it has to run.

A higher PRF shrinks the unambiguous range window, so a full polarimetric mode ends up with roughly half the swath of a single- or dual-pol mode at the same resolution. On RADARSAT-2 the Fine Quad-Pol mode covers about 25 kilometers .

The doubled PRF is also what pushes wide-swath quad-pol out of reach on high-frequency systems, which is why the X-band fleet offers no polarimetric equivalent of ScanSAR. SAOCOM does manage quad-pol beyond a 100-kilometer swath in its TopSAR modes, but it buys that width with resolution measured in tens of meters, and it works at L-band. Data volume roughly doubles as well, and the processing chain gets heavier at every step.

For a single site under repeat observation, none of this matters much. For anything regional, it is the whole problem .

When you genuinely need all four

There is a real threshold, and it is polarimetric decomposition. Methods such as Freeman-Durden split each pixel into surface, double-bounce, and volume components, and they need the complete matrix to do it.

That is what makes quad-pol worth its swath penalty in forest biomass estimation, in separating urban double-bounce from forested volume scatter across mixed terrain, and in soil moisture retrieval beneath vegetation, where the canopy has to be modeled away before the ground becomes visible.

Dual-pol decompositions do exist, and the H-Alpha approach is standard in most processing toolboxes. They work with a reduced coherency matrix, which is enough to sort scattering into broad classes but not enough to separate three mechanisms quantitatively. If your plan calls for the three-component model, you need four channels and your supplier list is short.

A short way to decide

Work back from the job rather than from the specification sheet, and the channel count follows on its own.

• Detecting something against a uniform background, vessels at sea or flood extent, works fine on a single co-polarized channel
• Separating two classes that differ in structure, ice types or crop types, wants one co-pol and one cross-pol channel
• Tracking change in one target over time depends on whichever channel stays consistent across the entire archive , which matters more than which channel it is
• Attributing a pixel to a scattering mechanism is the one job that truly requires four channels
• Any of the above across a wide area should settle the swath question before the polarization question

Those five cases cover the great majority of commercial radar orders, and only one of them points at the full matrix.

Comparing a spec nobody advertises

Polarization does not appear in the first line of a provider's specification sheet, and it rarely survives into the comparison tables buyers build for themselves. Resolution and revisit crowd it out, partly just because they are single numbers and polarization is a set.

The practical consequence is that you have to go and ask. Ask which channels are available in the specific imaging mode you intend to order, rather than which channels the satellite supports in principle, because those two answers diverge more often than they agree. Ask what the swath is in that mode. And ask how the archive behind a time series was collected, since a polarization change partway through shows up as change on the ground when nothing on the ground has changed.

The physics has not moved. The market has, and the fleet you can buy from today is mostly single-channel. For most of what radar gets bought for that is a smaller loss than it sounds – the second channel is usually available where you need one, and the fourth has become a research instrument that happens to be sold commercially.