Compare backscattering coefficients Sigma, Beta, Gamma for different landcover

Microwave Toolbox
1

I have taken a Sentinal 1 Image and calibrated it to find three backscatter coefficients sigma, beta, and gamma.
Using Pin Manager and the ground location points extracted the three coefficients for Urban areas, Villages, apartments, water, agricultural lands, forest.
30 pointsapartments_Data.txt (5.4 KB) Data_agri.txt (4.7 KB) each have been collected.
how can we find the threshold for a particular individual class for each backscatter coefficient?
can we separate them from each other based on the statistics or spectral sepratibilty ?
I have calculated average, standard deviation and coefficient of variation for all classes.
I don’t know how can I separate them based on some calculations .
Please find attached two files

2

I am not sure if that makes so much sense. These calibration measures are mostly compensating radiometric effects caused by variation of the global local incidence angle. I don’t know if a comparison of all tree gives you enough variation to discriminate different land cover classes because most of the differences will be constant offsets based on the topography and not related to the classes.
A more promising aporoach would be calculating statistics from different acquisition dates (dry and wet season), polarisations (VV/VH) or even looking angles (ascending/descending).

Still, it is explained in these tutorials how you can find and make use of thresholds to discriminate classes with the mask manager

To find ideal thresholds, you can export the backscatter values and the classes with the pin manager and analyze them in more specialized software, for example the tree classifier in Orange.

1 Like
3

Concerning gamma nought, one should distinguish between the variant that only compensates for an ellipsoidal incident angle, versus an approach where the local terrain facets are projected into the gamma nought convention (perpendicular to slant range) and integrated throughout a whole DEM to give a “terrain-flattened” result. Beta nought is not compensated even for ellipsoidal incident angle - it reflects the “raw” measurement at the radar, and is extremely ill-suited to land-cover analysis.

2 Likes
4

can i use the combination of sigma and gamma bands for such studies? like sigma_vh,sigma_vv,gamma_vh,gamma_vv and diffrence bnds for both like sigmavv-vh and gamma vv-vh

5

I don’t expect any distinct relationships to landcover classes.

6

I apologize for asking about sigma and gamma nought when this was already discussed. But I am curious if my understanding is correct.

I have used sigma nought backscatter (S-1 (A/B) - IW –GRD –VV,VH) in my work (trained a ML model to identify cover crops). I have utilized sigma nought because the sites are relatively flat.

Question1 -
The sigma nought values are not normalised based on local incidence angle, hence same feature will have slightly different values in far and near range. Will these be a problem? Should I have to do terrain flattening (converting to gamma nought) after radiometric calibration to avoid the influence of incidence angle?

Question2
I have used Time-series data (Sep to Mar) which includes both sensors (S1A and S1B) and only ascending images. Gamma naught is preferred unless and until different sensors and viewing geometries are employed and the terrain is not flat.

So, is my assumption regarding the use of sigma nought correct or not? I value your response.

7

Please have a look at this article. It will help regarding the incidence angle correction:

Gamma Nought is a generic product. It is not good for places with no flat terrain, but it is the easiest-to-use variable because it can be downloaded from Sentinel Hub directly.

1 Like