I’m not sure if a PCA is the best way to start flood mapping - do you have a source for this idea?
No Sir. I have heared about the IHS, PCA, Brovery transformation, wavelet based fusion. But I dont know what is the best method.
PCA is just a way to reduce the redundancy of a feature space (here a raster stack). It tells you where the images are alike (have the highest share of variation) and where they are different.
It is not necessarily linked to water body mapping, but there are studies which tested it:
Thank you for these papers. I refered this. I have finished PCA and HPF. After that they have made standardization of PCA components and inverse gray processing for High Pass Filtered microwave data. I have no idea about these processes.
In this flow chart can you please explain how to do that all processes?
(RGB to IHS and inverse IHS transform, standardization of PCA components, inverse gray processing, density separation, histogram matching and multiplication of the results?)
sorry, you have to ask more specific. I’m afraid I cannot tell you how to “do all that” 
SNAP is a image processing software, but you as a user must know what you want to do and why.
Ok sir… 
We published a new tutorial on the fusion of Sentinel-1 and Sentinel-2 today
4 Likes
sir ,I have a new ptoblem here,if i intend to fusion using sentinel-1 and sentinel-2 and to evaluate the accuracy of results.i got two methods to do that including create a stack and Collocate them,what the difference for the two ways?
Please use the search function for such things:
1 Like
Hi, would you mind share more information about how you implement Coregistration and to manage to show both Sentinel-1 and Sentinel-2 at the same time? Many thanks.
Appreciate your advice! I have tried with the tutorial, the collocate went successful
=D Big thumb for these very helpful tutorials!
Sir,
It’s coherent and corret to apply a classifier like RF on a PCA bands? You’ve said that for RF classification, it’s suficient stack the SAR and Optical bands.
Thank you.
the Random Forest classifier can handle input data of different units and scales, yes.
But the distribution and size of training areas is crucial.
1 Like
I Have tried to coregister Sentinel 1A and Sentinel 2B data to implement different fusion techniques (eg. PCA, IHS, HPF, Brovery Transform). The following processes have been applied to process the datasets.
Calibration ->Deburst ->Speckle filter -> Multilooking -> Terrain correction (Range doppler) for S1A data whereas the Sentinel 2B data were resampled to match the resolution of both the datasets.
In order to coregister the data sets, I have used collocation and stack tool, and then PCA was performed (result attached) in the stacked output. Due to the unavailability of specific tools for applying other fusion techniques such as IHS, BT, HPF in the SNAP platform, the collocated/stacked data were converted into IMG format and opened in ENVI.
However, upon opening the image in ENVI, only the bands of Sentinel 1 data (i.e., bands 2 and 3 in stacked output) contained any data, while the rest of the bands were blank.
Therefore, my query is regarding the coregistration of Sentinel 1 and 2 data, such that:
- Why am I unable to open the remaining bands of S2A data and
- How to verify if coregistration has taken place
The r g b values of some pixels have become Zero after converting the S1A data into IMG or Geotiff format.
Please assist me, why this issue is occurring??
What is the bit-depth of the S1 band in SNAP and the corresponding band in ENVI? If you need to reduce the number of bits scaling the values is necessary before export.
I suggest to directly read the img/hdr files from the data folder into ENVI instead of converting the entire scene to GeoTiff.
The reasons are explained here: Export of products from SNAP
1 Like
Thank you @ABraun for your suggestion. If I read the data directly from the data folder, I need to open data as individual bands.
Please assist me on my first query…
I need to precisely co-register data of S1A and S2B for applying fusion techniques. Therefore, I have implemented collocation/stack tools available in SNAP (as per the discussion here Image Fusion Using Sentinel 1 and Sentinel 2 ) and the output product saved as stacked of all the bands.
does this mean my data are co-registered ??
collocation is fine if you select the Sentinel-1 product as the reference and we assume that its terrain correction worked well.
However, it is not a coregistration at the sub-pixel level, but only a stacking of two well geocoded products.
You can check with an RGB image if the information of both products is well aligned.
1 Like