Hi,
My intension is to use S1 data to do multi-temporal analysis of agricultural sites. I would like to work just with intensity, so with GRD Data. I would like to assure about to procedure and kindly ask if you could give me advices or tell me where I am wrong…
The first thing radiometric terrain correction then-
Calibration, to get sigma 0 values. (or Beta to process terrain flattening?)
Then speckle filtering,
Range Doppler Terrain Correction
Terrain flattening (I am wondering if this step is necessary because all the agricultural size will be almost flat)
I would like to process these steps on multiple scenes (time series). Should I geocode them before or after. And is it necessary to coregister if I will use scenes from the same orbit? Do I need to geocode if there is Terrain Correction?
Then I have a polygons of my agricultural fields in shp, I understood that I could clip the raster with land-sea mask module. And after I will see the statistics of individual fields. I also read somewhere that for statistical purposes it is better to work with gama 0 values… where is the difference?
Is in SNAP something like multi temporal features generator as it is in SARscape?
Terrain Flattening compensates radiometric distortions due to the local incidence angle. If you are not having much topography you can skip this, otherwise you would first calibrate to Beta0 and then directly apply Terrain Flattening.
Incorporating the local incidence angle instead of the angle between the signal and the ‘flat earth’ surface would give you Gamma0 instead of Sigma0, but as you said, this is only necessary if you want to reduce topographic influence.
What may help would be the conversion to db to get an evenly distribution value range.
If all your data is from the same track you can terrain correct and stack the images. If you have different looking directions, angles and so on, coregistering would be more suitable.
During Terrain correction I have several options for DEM to use.
If I am right, there should be available SRTM 1 sec (30x30) in Europe.
In my area I can also get DEM grid (10x10) and also high quality DEM up to (2x2).
I would like to ask how precise the DEM has to be? Does it make significant improvement when precise DEM is used? Because that precise DEM will really slow the process if am not wrong.
As I mentioned, my intension is to use intensity images for landcover analysis and classification in agriculture area (without rough topography). But could it affect the results somehow?
In this step also resampling methods should be chose. I could not find any descriptions about these methods, advantages or disadvantages and differences in use.
Could you please give me a hint where can I find further info?
SRTM 1 (30m) is fine in Europe with some exceptions (e.g. Alps) but if you have a DEM of higher resolution give it a try. Choose external DEM and make sure it is projected to WGS84 (lat/lon). As the TC process doesn’t take so long you can also compare if there is any improvement. In general: The more detailled the topographic information the better can geometric errors be handled.
Bilinear resampling is a good compromise between processing time and complex generation of the new rasters. It just determines how your DEM is adjusted to the resolution of the SAR product.
These slides are quite interesting regarding resampling, at least the last few slides.
And while I am thinking about future steps I am wondering about co-registration. As you said it is better than just create stack when there are different angles. In my case I will work with more than 15 images. I tried to co-register two of them at first and it already took a long time. Is there any difference if I will co-register images individually when I will choose the same master every time? I assume that to process with all images will not be done…