Wednesday, June 14, 2017

PALSAR - Hiccups Part II

I'm very excited to say that next week I have the pleasure of attending the Earth Observation Summit in Montreal!  The summit is a combination of three meetings: the 38th Canadian Symposium on Remote Sensing (CSRS), the 17th Congress of the Association Québécoise de Télédétection (AQT) and the 11th Advanced SAR (ASAR) Workshop.  My main purpose in attending is for the ASAR Workshop, but I am also very keen on learning all about the cutting edge remote sensing techniques and satellite technologies!

I have a presentation in one of the sessions on Advanced Polarimetric Methods. Mike and Hun will also be attending, and they have talks in the Geology session (wah! I'm geology, too! But my topic fits in either session :) )  Next post, I'll tell you all about the cool things I learned!

In preparation for my talk and our upcoming field season to Axel Heiberg Island, I was going through some of my data with a fine picked comb.  Remember those PALSAR Hiccups I was having before? It turns out things are a bit worse than I previously thought!  Not only are the images poorly registered (i.e. they don't line up with the other maps properly) but there is also some severe distortion to a process called 'layover', which is a common, unfortunate challenge with radar imaging. I wrote a little bit about layover in a previous post last year.  Usually terrain correction helps lessen the distortion.

Compare these HH-Intensity image pairs:

Figure 1 Two images of South Fiord Diapir. Above, RADARSAT-2, water is masked out. Below, PALSAR-1 with water included.  Note how the shape of the dome is distorted in the PALSAR image, with a thick white line on the eastern margin.


Figure 2: Above, RADARSAT-2, below PALSAR-1. Again, water is masked out in the RADARSAT but not the PALSAR image. Pink outlines delimit salt diapirs as identified in  other spectral datasets, including (west to east) Surprise Central Diapir, Bastion Ridge Central Diapir, and Glacier Fiord Diapir. Notice how the mountains are more defined and '"3D" looking in the top image, but appear almost sideways or overturned in the lower image. Additionally notice how combination of the layover distortion and poor image registration results in the pink outlines being offset from the actual features.

The cause for this distortion is the incidence angle. The incidence angle is the angle between the radar beam and the normal (perpendicular to) to the surface (Figure 3).  

Figure 3: Illustration of angle of incidence, from Wolfram

For the RADARSAT-2 images the incidence angle is ~40▫, where as the PALSAR-2 images have an incidence angle of 21.5▫.  For a flat surface, this wouldn't be much of a problem.  However, as shown in Figure 2, Axel Heiberg Island has a lot of hills, mountains, and diapirs!  The angle of repose for most materials is around 30▫, and that can be less for bigger blocks.  It is likely that a lot of the mountainous areas are sloping close to the incidence angle of the PALSAR-1 images. This means that the radar beams are inflicting the surface almost perpendicular to the mountains, making them look flat!  And oddly sideways!  To help me visualize this better, I made some sketches and set up a physical model using my bus pass, an external hard drive and a pen, but hopefully you won't need to do that. 

There are two main types of slope angle effects, forshortening and layover. Most radar images (unless you are looking at extremely flat terrain!) will experience these to some extent. The effects are exacerbated in mountainous areas, and with smaller incidence angles. Recall how radar works - we are sending a beam to a surface, and measuring the amount that bounces back.  The amount of time it takes for the beam to bounce back can affect its position in the image, so if a radar beam is taking different amounts of time to reflect off of different parts of an object, that can affect what the object looks like in the image.

Foreshortening occurs with a radar beam reaches the bottom of a slope feature before the top. In foreshortening, slope facing the radar beam will appear "shortened" in the resulting image. The lower image in Figure 1 shows foreshortened western flank of the diapir dome.  Layover is similar, but occurs when the radar beam hits the top of the slope before the base.  This makes it look like the sloped feature is tilting towards the radar position. The lower image of Figure 2 is a better example of layover, because it looks like those mountain ranges are tilted sideways towards the east (I think - that's what it looks like to me, but I'm still learning how to recognize these effects. :) )

Ultimately, these distortions have shown that not all of our PALSAR-1 data is reliable.  I noticed that a lot of the salt polygons were misplaced from the distortions.  However, I am happy to say I was able to perform a quick band-aid solution.  I went through each of the polygons and did a quick visual assessment to see if the radar looked reliable. I manually moved some of the polygons in ArcGIS so they would be overlying the right features even if the shapefiles aren't registered the same. This has given me enough patched-up data to still present some PALSAR-1 results at the conference next week, but it is a short term solution until we either re-register the data, or scrap it.

So, the 'new' results show that the salt diapirs more rough in PALSAR-1's L-Band than in C-Band.  In C-Band, the CPR average values are ~0.4, but in L-Band they are ~0.6! The non-diapiric salts have similar signatures.  In C-Band non-diapiric salt has CPR ~0.21, and in L-Band it is ~0.23.  Previously we thought there were some radar-rough areas of non-diapirc salt in L-Band, but I have confirmed those were registration issues.

This is still good, and we can work with this!

I'll let you know how the Earth Observation Summit goes!

For more information on radar slope effects, this page is awesome: 

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.