When applying S-1 Thermal Noise Removal to a recent S1A_IW_GRDH image (e.g. S1A_IW_GRDH_1SDV_20221214T045155_20221214T045220_046323_058C41_4751) there is a gap between subswaths IW1 and IW2 in the noise removal pattern that results in a bright strip in the image after noise removal.
In the following screenshot you see (1) the VH channel of the untouched GRD product, (2) the VH channel after applying orbit file, removing border noise, removing thermal noise, calibration and conversion to decibel, (3) the NEP pattern of the removed noise.
Any explanation as to why this is happening?
This doesn’t seem to be a problem local to my machine, since the mosaics in EO browser also contain these strips: S1A IW on 2022-12-14 in EO browser.
A bit more investigation shows that this seems to affect scenes between 2022-03-23 and 2023-03-30 when IPF versions 3.51 and 3.52 where used.
The changelog of IPF 3.51 includes
Correction of the misalignment between the elevation antenna pattern and the annotated thermal noise vector
and the fix in IPF 3.61 seems to be listed as
Avoid missing data in range denoising vectors annotations for TOPS GRD products on long datatakes
Is this the cause? Can we “repair” the denoising vectors annotations for GRD products processed by IPF 3.5x?
Any feedback on this is appreciated.
The issue is indeed related specifically to products processed with IPF 003.51 and IPF 003.52 but is in fact related to a change of the processing environment. This was mostly solved starting with IPF 003.61 (*)
Please notice that 0 values in the denoising vectors have to be considered as missing values, and not as valid values (with a value of 0). Those missing values shall then not be considered while performing the denoising. This applies both here, and in the borders of the products.
There is no direct and easy way to “repair” those denoising vectors or to infer the missing values.
Those specific missing values are present on some GRD products near the subswath transitions. The denoising values are supposed to be continuous within a given subswath. Then, an extrapolation of the previous/next series of valid values could be close to the valid ones. However, there is no guarantee for this. Furthermore such extrapolation shall be done only up to the real subswath transition that is varying within the product (refer to the swath_merge annotations in the products).
There is currently no plan to reprocess the products processed with IPF 003.5x and having missing values in denoising vector with a more recent version of the processor.
(*) the issue was observed on a very limited number of products processed with IPF 003.61 on a very specific configuration (around 5 products only)