How to remove/reduce soil moisture effects on Sentinel-1 data?

Microwave Toolbox
#1

Hi all, I am wondering if there is a way to remove (or at least reduce) soil moisture effects on Sentinel-1 radar backscatter.

Problem:
The two images are hardly comparable. the gamma0 backscatter values of the “dry” image range between 0 - 0.3, whereas the gamma0 backscatter values of the “wet” image range between 0 - 0.8.
The area of interest is covered by vegetation types: bare soil, grassland, shrubland and forest, and I am trying to classify these based on the gamma0 backscatter. Obviously a classification based on a 0 - 0.3 range gives a very different results if applied on a 0 - 0.8 range. Any idea/suggestion on how to compensate for this and make the two images comparable?

Data specifics:

  • Sentinel-1, GRD, IW, VV polarisation, Descending Pass, Ethiopia.
  • one image in December (dry season)
  • one image in August (wet season)

Data pre-processing:

  1. Radiometric - Calibration;
  2. Speckle filtering
  3. Radiometric - Terrain Flattening
  4. Geometric - Range-Doppler Terrain Correction.

Thanks a lot for your attention
simone

1 Like
#2

Hello Simone,

I don’t think it is possible directly as the radar is quite sensitive in the matter.

I would suggest trying to convert the values to dB (logarithmic) scale first. Should give you better starting point for classification in general.

Next try to apply some linear scale for dry or wet. Here I would go for scenario: more water the higher the scattering rate the lower values, therefore add (+5 to +10 [db]) to your rain season acquisitions.

Squaring the dB values could help as well.

Finally, if you would have the average over a larger time series you can target the deviated value from the average for the classification.

Good Luck

Petr

#3

dB conversion is a good step. The wetter scene should have, overall, higher values. So adding +5 to +10 dB is not going to help. Differences due to soil moisture should be largest for the bare soil class, assuming it is bare in both seasons, which may not be the case [in the rainy season]. For forest, there should not be much difference at all. For vegetated (non-forest) areas, the status of the vegetation in dry/wet season is a [more] important factor. Check if you can make a reasonable differentiation between bare and vegetated areas on the basis on the difference in backscattering. You could integrate some Landsat data to check what is really bare, vegetated and forest.

In Ethiopia, you only have VV unfortunately.

#4

Thank you both for your quick answers!
i tried to work around it by applying flexible thresholds (rather than fixed ones based on the dry image) based on the pixel values distribution through the histograms analysis of both the wet and dry images.
It seems to be working fine, with the assumption that the differences in soil moisture (and other calibration variables) between images acquired in different periods are somewhat spread across the entire scene. :grinning:
thanks a lot again!