Noise in TerraSAR-X Images

Hi. I am processing some TerraSAR-X images using SNAP over an area of ocean north of New Zealand. We chose the Wide ScanSAR mode in MGD format because we wanted to cover a large spatial region. I have applied radiometric calibration to sigma0 but the images are very noisy compared to other SAR sensors that are using similar modes/resolutions. For example, see the attached image where you can see lots of noise. There is also a strange spatial structure despite this being an area of open open with no land or structures to reflect the radar signals (caused by weather or wave conditions?).

I have tried to apply speckle filters and additional multi-looking to the image, but this doesn’t help. After some digging around online, I found some old presentation slides from 2008 titled “TerraSAR-X Products – Tips and Tricks” which talked about these noise patterns in low backscatter areas which are typical for ScanSAR, but I wasn’t able to find any further information.

Does anyone on this forum have experience processing TerraSAR-X images using SNAP that may have some advice? My usual processing chain for processing (non-Sentinel) SAR images in ScanSAR mode over flat ocean is very simple. I just apply radiometric calibration to sigma0 and then geocoding using the Ellipsoid Correction.

Thanks

Andrew

This product TerraSAR-X MGD, is already multilooked with reduced speckle filtered,

Source: https://spacedata.copernicus.eu/documents/12833/14537/TerraSAR-X_ProductGuide

In SNAP, did you try up SpeckleFiltering of different filters and different window size, and number of looks, also try up to reduce the sigma, Might this could helps,

Also you could find the other close date less noise or without or same date of other track, if it’s available, if this applicable to you.

1 Like

do you have the chance to get the SLC image insted of MGD?

Sometimes these patterns are only subtle, but are enhanecd by multi-looking. Please have a look at this example: Crop classification using SENTINEL-1A SLC data

If you had the SLC product, you could do the multi-looking yourself and maybe reduce this pattern a bit.

Thank you both for your responses.

Yes, I tried using a variety of different parameters for the speckle filtering, but found no improvement in the results. We have 3 separate TerraSAR-X images for different dates and regions (all over open ocean) and see the same noisy data in all cases, compared to other SAR systems that are operating in similar resolution modes.

Unfortunately we don’t have access to the SLC format or any other images because this region is not part of the standard imaged regions for TerraSAR, so images are only collected when we specifically task them from Airbus.

@Abraun I found a 2017 post of yours where you suggested using multi-looking after the radiometric calibration stage instead of speckle filtering (TerraSAR X image quality and building measurement). However, since the MGD product I am using is already multi-looked, I’m guessing that this SNAP mult-look processing is just smoothing the pixels by averaging over a X-by-X square, centred on each pixel. The attached image shows the original image without multi-looking (left) and after this additional multi-looking has been applied (right - using 7-by-7 GR Square Pixel settings). The resultant image is still grainy (although not as bad as before) but with the corresponding loss in spatial resolution.

I’m also curious about what the structure in the image might be. There are distinct regions of high and low backscatter but, since the dividing line is roughly diagonal across the image, this is not related to the SAR geometry. The imaged region is an area of open ocean in the Pacific with no terrain or structures to reflect the radar signals. I’m guessing that this may be a result of a rain or wind front but I am not an expert on SAR - do you think that meteorological factors could be the cause this kind of image structure?

I don’t think that is atmosphere-related. It rather looks like it is caused by the ScanSAR pre-processing (to MGD).
I don’t know exactly where it originates from but I doubt that you can remove it after it is already introduced to the image.

Maybe an expert on TerraSAR-X can clarify?

1 Like

The structured pattern (high backscatter in top left corner and low backscatter in bottom left) is probably due to meteo. See for instance http://www.sarusersmanual.com/ManualPDF/NOAASARManual_CH17_pg355-372.pdf for examples of raincells over water in SAR images.

As for the noise, looking at the documentation it appears that the noise floor (the noise equivalent sigma 0 or NESZ) of TSX wide ScanSAR is higher than S1’s or RS2’s similar modes. The higher noise floor will be specially visible in areas of low backscatter, e.g. oceans, in the form of more intense salt and pepper noise as well as more pronounced “scalloping” (this is the horizontal stripes in your images, resulting from the fact that in the ScanSAR mode different points on the ground are illuminated by different parts of the antenna pattern). You can always try spatial filtering, but you will need to accept trade-offs.

The information about the NESZ comes from:

The lower the NESZ, the better, all other things equal.
Mind you, there may be more up to date product spec documents somewhere on the internet, but numbers should not change that much.

2 Likes

@css Thank you for this explanation. It is really helpful and I will read through the information contained in the links you posted to learn more details.

Through the SNAP processing of SAR images from Sentinel-1, RADARSAT-2, ICEYE and NovaSAR, it’s been interesting to see the differences between the output images when using the different instruments and acquisition modes. The historical TerraSAR-X images you see on the Airbus GeoStore (using the same Wide ScanSAR mode and over the ocean) don’t have the same noise issue that I have found with our tasked images. So I guess they must be doing some kind of smoothing or spatial filtering, with a resultant reduction in spatial resolution. As @ABraun suggested, I will try to make contact with someone at Airbus who can maybe shed some light on their processing procedures, and I will update this post if I get any further information.

One thing comes to mind now, if you have any chance to use L band HV polarization could be a good solution over open water area, continuing in your research instead of TerraSAR-X,

Please post the name of your TerraSAR-X WS MGD directory e.g. TSX1_SAR__MGD_RE___SC_S_SRA_20150313T232932_20150313T233002.

TSX WS has 6 beams 1: 15-23 degr, 2. 23-27, 3. 27-31.5, 4. 31.5-35, 5. 35-38, 6. 38-41.
In beam 1 at 15 degrees of incidence angle, you will be getting a very high return from the water surface.
Your diagonal looks like a storm front. You could work with a software package, that allows for a localized histogram adjustment to look for features in the image.

Thanks for your inputs. The TerraSAR-X WS MGD directory is named TDX1_SAR__MGD_RE___SC_S_SRA_20191109T065739_20191109T065801. Looking in the metadata (SNAP screenshot attached) it seems that the beam incidence angles for this image are in the range 41.6 - 48.5 degrees. This is at the far extreme of the “Full Performance Range” for the Wide ScanSAR mode (15.6 - 49 degrees according to the TSX Product Guide).

What you are seeing are the 4 beams (wide vertical stripes) and the fine horizontal stripes are the bursts.

Display: default percentiles will show the stripe patterns. You can play with 2,3, 4, … standard deviation stretches and the image becomes darker by and by. If you cutoff the lower end of the histogram to e.g. 15,


These pattern are visible, since there is very low backscattering from a smooth water surface. The diagonal pattern is a storm front as said earlier. Waves have been whipped up and cause rough surface scattering and thus a grey texture.
Sample ship:
2020-02-14_142038_ship
Yet we know, that the data can be exploited: You can adjust the histogram by setting the lower boundary e.g. to 15 and the upper boundary to 40 and the water surface becomes dark and ships pop up (~200-300m in length).

To exploit the data automatically, you would have to work with an adaptive threshold throughout the image: the incidence angle, the standard deviation and the mean of the clutter could be used to do so.

Software for SSC processing is rare on the market.