Backscattering for Soil Moisture

Hello EveryOne…

I want to use SAR Sentinel 1 Backscattering for soil moisture? but I don’t know how to get backscattering data from Sentinel 1A? Please share your knowledge.

  1. I have download data S1A-SLC-IW
  2. Having Snap

But I don’t know how to convert my data to backscattering?


Bacscatter can be retrieved by calibrating to Sigma0

Please have a look at these steps:

Downloading GRD data could be sufficient for your purpose.


Thank you Sir,

Which one is best product to be used for soil moisture measuring and why?

I would say GRD is sufficient. You probably need multi-temporal data so file size can be an important aspect. GRD is 500-900 MB, SLC is up to 7 GB

1 Like

and Sir which equation plus algorithm would you prefer?

if there only was an equation… :slight_smile:

Please have a look here:

Hello forum how are you doing? am a little bit confused I hope I will get your suggestions the problem is I used two techniques to estimate soil moisture content which are the empirical multiple regression and the change detection however, my result appeared very far from previous studies for instance the R2 for observed and estimated soil moisture content using empirical method became 0.013 the backscatter coefficient has much dependency with local incidence angle rather than the soil moisture content in my study area

@Anna2: It’s a bit unclear what kind of response you expect, because I don’t see a question in your post. Based on the information you provided, we cannot know where the deviations between your methods (and yours and other studies) come from. Objectively, possible reasons are

  • the studies you used for comparisons are based on different data or (spatial and temporal) scales
  • your field observations did not allow to create a robust regression
  • errors during data processing (resampling, offsets, image dates)
  • the two methods you compare produce different target variables
  • the spatial resolution of the satellite data is too coarse
  • vegetation or roughness introduces errors

We really cannot know. You have worked with the data for the past years, so you are the person who knows best where potential errors could have been introduced.

This article discusses many of the mentioned error sources: Advances in soil moisture retrieval from synthetic aperture radar and hydrological applications

am sorry here is my question I predicted soil moisture content using multiple linear regression but I got low performance I mean the backscatter coefficient showed less dependency with the soil moisture content which are collected from the field

Sampling Date R2 σο & SMC θi & SMC σο & θi
1 October 14 ,2019 0.19429 0.14455 0.8247
2 October 26 ,2019 0.00018 0.00008 0.8023
3 November 7 ,2019 0.17 0.1889 0.8298
4 December 25 ,2019 0.08 0.09 0.9132

that indicates that there are other factors contributing to the backscatter intensity, for example roughness or vegetation. How many samples are used to create the R²?

Would it make sense to create a regression over all four acquisitions at once?

i had total 60 samples 15 for each date i used 40 samples out of the total to make the regression and the rest 20 samples were used for validation

so how many were used to validate each date?

five points for each dates

this is not sufficient for a robust validation. If I were in your situation, I’d use 60 points for training and then use the 10-fold cross validation to test its accuracy. Or the leave-one-out method.

Orange is a fantastic software to model/predict/validate tabular data. You read the data as csv or table, define the target variable and then apply different predictors followed by a validation.

you are right but my study area is a micro watershed with an area of 1.8 square kilometers that’s why I took 15 samples per day

nothing wrong with that, as long as you extract these from the images of the same date. I just don’t think it is feasible to compute an R² from only 5 points. There are better ways of calculating the performance of a regression.

Maybe the images are acquired from different look directions as well?

In the end you have 60 value pairs, right? Sigma and moisture (from different dates)

yeah. Additionally, I want your suggestion also for another thing which is when I collect soil measurement I took it by considering the elevation difference of the watershed which means the samples were taken from the lower, middle, and upstream of the watershed separately so when I apply the empirical technique I got different performances in each stream, the middle stream showed better performance relatively and the middle stream is relatively flat and bare land so what do you think, please let me have your assumptions thank you so much for your time.

how is this possible within 1.8 km²?

am sorry i don’t understand?

you said that your samples are within 1.8 square kilometers. How can these contain differnt parts of an entire watershed?