You can use VH/VV parameter. It has some correlation with NDVI.
Thanks for your response. yes, I have tried with the VV of Radarsat2 image and correlate the backscatter values with the NDVI, and the result of R2=0.75 . with the equation I got from the linear regression (y=0.1603x+2.3753), I tried to get the predicted radar value that equivalent with NDVI. So Y is the radar value that I looking for. So, I replace x with NDVI value, and calculate it. I apply for NDVI image (using bandmath) it give 0.6 to 0.8 values, but when I apply to Radarsat2 image it gives, weird values (-300 to - 800.00 ++).
Do you have any suggestion to get better result?
thanks.
How do you calculate correlation? it’s strangely high.
Wait when you use linear regression, x is NDVI and y is VV?
You can try to build dataset: VV | NDVI
And use tool like scikit-learn to build predictive model. There you can calculate performance of your model.
which software did you use to produce this compared ？
I just use simple linear regression. what do you mean it strangely high?
I use, y for NDVI and x for vv.
Sorry, for the late answer.
For the visual representation of average pixel values (these two plots), I write some python scripts and use for it https://matplotlib.org/ library.
I want to ask, how to combine 2 sentinel1 radar images that have the same date but the image is separate?
i’m sorry because I first used sentinel1 image.
By coregistration, found under the menu > radar > coregistration
Hi Andreas, you stated that:
you could also investigate the ratio of VV and VH. But be careful:
If the data is not logarithmised (Sigma0), the ratio is VH/VV
If the data is logarithmised (Sigma0db), the ratio is HV-VV
is this; VH-VV ???
1 Like
yes, that was a typo, sorry. VH/VV = VHdb-VVdb
1 Like
Hi Reznik…
You stated that “value of VV+VH it mean (5 * VH ^ 2 + 2 * VV ^ 2 - 6 VH * VV ) ^ 0.5”
Do you know any related paper, please?
Hi Andreas,
M y values of VV and VH looks different. I’m unable to find the relation between NDVI, VV, VH, VH/VV. My values look like below table, I’m not getting values between 0 to 1. can you please help me out and below values are from the coinciding date from S2 and S1.
VH VV VH/VV VH-VV NDVI
-15.44303158 -12.46460068 1.238951169 -2.978430901 0.609489051
-16.47754983 -12.50395196 1.317787359 -3.973597871 0.606344254
-15.75078747 -11.73454568 1.342257971 -4.016241792 0.605604018
-15.44239821 -11.76221284 1.312882059 -3.680185369 0.605376344
-17.11258532 -13.03223986 1.313096252 -4.080345454 0.605371335
-16.91891376 -12.94311332 1.307174969 -3.97580044 0.603872318
-16.50076777 -11.60078072 1.422384248 -4.899987041 0.603742433
-16.15313564 -12.16100724 1.328272841 -3.992128397 0.60359408
VV and VH should be different in most cases 
But your NDVI is nearly the same for all 8 rows. You need mor variation her to perform a linear regression.
Thanks for fast reply, but one thing I need to confirm, after performing VH/VV will the values range between 0 and 1 or they will be greater than one.
Edit: glemoine is right, it’s because of the dB scale.
1 Like
With VH/VV you are dividing dB values, which does not make sense. Work with the VH-VV difference (which is equivant to dividing VH and VV and than convert to dB).
For vegetation, VH will normally never be larger than VV, which is why you always have a negative VH-VV.
Do not expect to find a linear relation for all crops. Backscattering and reflectance are different physical processes, related to different vegetation parameters (structure vs chlorophyll).
2 Likes
Hi Glemoine,
I’m downloading data from google earth engine(GEE), in GEE the GRD scenes are processed to backscatter coefficient (σ°) in decibels (dB). so the data i’m pulling out will be in decibels. If i need to continue in decibles what would be the process for correlating NDVI with SAR?.
Regards,
Krishna
Use the S1_GRD_FLOAT collection instead, which is not in dB.