Problems to delete the speckle (GRD S1)

Hello Everyone! I really need help with the workflow of my GRD Images.

I’m trying to detect the formation of slums with S1 GRD Images (Monthly images (2020), Descending, same relative orbit)

I’ve been working with de workflow of Filipponi (1) and the timeseries (2)

  1. Orbit file > Thermal Noise Removal > Border Noise Removal > Calibration > Speckle Filter (Refined Lee) > Range Doppler Terrain correction > Conversion to dB
  2. Thermal Noise Removal > Orbit file > Calibration > Range Doppler Terrain correction> Stack > Multi´-temporal speckle filter (refined lee)

My problem is that I can’t delete the speckle of the images. Oddly, with the coregistration I’m having worse results.

With these results I pretend to calculate the velocity and acceleration of the formation of slums, without coregistration the results can slightly identify this formation, but with a lot of accelerations in places where there hasn’t, the results with the coregistration appears to be accelerations everywhere.

I’m starting to think that there’s the presence of white noise. Anyone had ever have a similar problem? Or have any idea of how to solve it?

The code of my images are:
AOI: POLYGON ((-70.1821194355951 -20.1625259834406,-70.0374695163088 -20.1651619728581,-70.0401055057263 -20.3068464040496,-70.1831079316267 -20.3042104146321,-70.1821194355951 -20.1625259834406))

The resultant image (with the workflows mentioned) have this histogram:

Those negative values coincide with the areas where I have problems.


I’m not sure if this could be the problem, but the Lee filter is made to preserve edges in images, which means that it doesn’t remove as much speckle where there are lots of small objects in an area. I am not sure if this is the root of the issue here, but it could be worth trying to use a simple boxcar filter instead of the refined Lee, and see if it gives better results.

I hope this helps!

Thanks for your answer Paul! I change the filter to boxcar (3x3), but my problem still occuring. Once that I applied the serie (Thermal Noise Removal > Orbit file > Calibration > Range Doppler Terrain correction (Alos Palsar)> Stack > Multi-temporal speckle filter(Boxcar)) I calculated the acceleration of change (SNAP, band math) (for example, (March -2*February + January /2)) my results look like this:

However, when I export each monthly band separately and then calculate the absolute value of acceleration (QGIS) I got this:

The calculation on snap results accelerations everywhere. The calcule with separate bands looks better, but with a lot of noise in places where there hasn’t have any changes. I don’t know if this is a problem of the speckle filter or the calculation. Anyone has ever have a similar problem with this? I will appreciate some help! Thanks

I can see that in the series of transformations you indicate, you do not transform the data to dB before applying band maths using the formula you indicate, which would be required for additions and subtractions of the received intensities.

I think it could also help to see if there is an underlying issue to try to detect change between only two images, with for example March/February before transforming to dBs (or March-February for dB values).

Multi-temporal speckle-filter smears out structural changes in time. You are probably better off using single-image speckle-filtering.

Im sorry, the results that I published up are based on this workflows, both with the Terrain correction in base to Alos Palsar (12.5 m) (EPSG:4326):
a) Workflow for time series (Boxcar filter) (

B) Standar Workflow, using single speckle filter (Refined-lee filter)
I been working on this and still can’t remove the noise. You suggest to apply the workflow until the terrain correction, make the band math (for accelerations and velocities) and then convert the values to dB?

Even if I pretend to make a time series images to evaluate the apparition of slums? There’s any filter in particular that you recomend? I’ve been using the Lee Sigma, Refined Lee and Boxcar and the problem with the noise still appearing.

You cannot really expect speckle to disappear completely without filtering heavily, which will degrade resolution one way or another. I don’t really know which single speckle filter to recommend but the Lee filters should be ok. Anyway, you should drop the multi-temporal filtering as it is made for situations where the structures stay stable for the whole period.