Pauli Decomposition

Hello again, and welcome to another week of Arctic Resolution.

Last post, we explored the new 2016 RADARSAT-2 Acquisitions we obtained over the Expedition Fiord region of Axel Heiberg Island, Nunavut.  I showed you derived Circular Polarization Ratio (CPR) images of our study area.  This week is similar.  Since producing the CPR maps, I have compiled a set of Pauli decomposition images for our site.

What is a Pauli decomposition?  Pauli decompositions are used to show scattering mechanisms affecting the roughness of a surface.  In this way, they are very similar to CPR images; albeit more qualitatively interpreted.  CPR is calculated as a single integer, in which surfaces can be categories as smooth (CPR <<1), moderately rough (CPR ~0.5-1), or blocky (CPR >1).

In contrast, Pauli decompositions are more interpretive.  They are represented as RGB composite images, with the colour red being derived from the HH and VV bands: |HH-VV|^2; green from the HV band: |2HV|^2; blue also from HH and VV: |HH+VV|^2.  Recall that we are using quad-polarimetric radar, which takes advantage of RADARSAT-2's capability of transmitting and receiving horizontal and vertical signals.

What does this ultimately mean?  Areas of the images that feature more red highlight areas of double-bounce scattering, like blocky surfaces or ice.  Green areas show multiple-bounce scattering, which are rougher surfaces.  Blue shows single-bounce radar, which will indicate smoother surfaces.  Of course, these images are composites, so very few places will show solely one colour, with lots of purples and yellows indicating multiple types of radar return.  This fact exemplifies that Pauli decompositions are qualitatively interpreted; they look pretty, but caution is advised when characterizing surfaces.

Like the last post, the images are terrain corrected in SNAP, and uploaded into ArcGIS.  Unlike the previous CPR images, we did not use PolSARPro at any step during the processing.  All of the processing was performed in SNAP, as follows:

• Radiometric calibration
• Multilooking (we went with 4 azimuth looks this time, instead of 6)
• Pauli decomposition
• Terrain correction (Range-Doppler)
• Export to ArcGIS

For some reason, though, some of our images look silly!  Instead of looking like a regular Pauli decomposition, a few took on very bright pinks, blues, and greens.

Pauli decomposition produces an RGB composite image, with purple, yellowish and greenish colours most common.  The bright pink and aqua is the result of a stretching issue.

I was able to fix this problem by setting NoData values to 0 in each of the images, which removed pixels that were offsetting the histograms used in stretching the images.  I haven't stretched the images on the same scale, yet, but you can see they look much better.

Pauli decomposition of 2016 RADARSAT-2 acquisitions, masking out areas of 0 elevation (i.e., water).  Images are not mosaiced or stretched on the same scale.  

Pauli decomposition of 2016 RADARSAT-2 acquisitions, including water.  Images are not mosaiced or stretched on the same scale.

Comparing differences between the water-inclusive and water-exclusive compliations, as well as colour tone discrepancies among images within the same compilations, shows the importance that image stretching plays when we interpret these images.  Remember, since Pauli decompositions are qualitatively interpreted, consistency is very important.

A preliminary assessment, though, shows that our large salt dome is yellow, which suggests that this area is rough, with a combination of double-bounce scattering (red band) and multiple bounce scattering (green band).  This is consistent with the high CPR signature seen previously.

RADARSAT-2 Data and Products (c) MacDonald, Dettwiler and Associates, Ltd. (2016) - All Rights Reserved. RADARSAT is an official trademark of the Canadian Space Agency.