Sentinel 2 mosaic


maybe there was a mismatch between the coordinate system of the dataset and the boundary coordinates.
I often see extemely huge files being created when UTM coordinates are applied as boundaries for WGS84 projected files.

Just an idea…


I didn’t give boundary coordinates, maybe that was the problem. I managed to run the mosaic for a small area in the border of the granules just fine. I have subsetted the same area from older products and then the .data folder size was ~266 MB, with this mosaicked product it is ~683 MB. I don’t understand why there is double bands in the band listing :nerd:… what is in the _count bands?


I have recently explained what’s in the count bands:


I have the same problem to create a mosaic. I get the error of Product 'sourceProduct2' contains bands of different sizes. Currently it is not possible to use it for mosaicking.
I tried doing the mosaicing operation with the SNAP tutorial, but I cannot achieve to do it.

But when I do the mosaicing operation with one band (for example band 2 from each image) I have not any error and I will get the mosaic.


If I don’t get the multiple reader dialog when opening SENTINEL-2 data . How do I enable it?
Need it as preparation for mosaicing, getting the same resolution for all bands as indicated above.
Running W7 and W8.



The 10/20/60 m resolution readers are no longer available, but you will get the same result if you use the resampling operator after opening the multi-resolution product.


I have problem with mosaicing the sentinel 2 images. I tried resampling and subseting but it is not fixed. Do you find a way to fix mosaicing problem anybody?


I have a problem with mosaic 2 sentinel images . I subset them then make a mosaic. However, it said can not display source product, …" as below. what is a matter?


This is a known issue and it is actually already solved. The fix will be released soon.
Meanwhile you can try to set the mosaic boundary manually. This should help

See also:


Please take a look at this video it is already released now

Sentinel2 Mosaic

Hi, I followed the video to investigate the blending operation at native resolution for 1 band (8a) at native 20m resolution (tab 2 window below and final results) as the final result screenshot shows, the processing seems to have picked up some aspect of the cloud (mask?) with the rest of the pixel values set to NaN. I am using SNAP v5.0.8 on a 64-bit kubuntu OS.
If I do the same processing but not at 30m resolution instead of the native (20m) resolution then I do get a mosaic but the ‘blend’ option is ignored.

RE the video:
-it does not state the nature of the blend algorithm, is it distance weighted?
-why does the ‘expression’ field in the third control tab seems to have no function, unlike the one in the basic ‘Mosaic’ tool?


Hello Thomas.

To answer to your questions:

  • the blending is just a simple pairwise alpha blending (on two images, the first one has weight 1, the second one 0.5);
  • the ‘expression’, in the case of multi-size products, has not too much sense. Expressions were designed to work with pixel values. In the case of, let’s say, B8 and B8a, since a pixel in B8a corresponds to 4 pixels in B8, it doesn’t make sense to use these values in an expression (which value from B8 will it use?);
    If you do the mosaic at 30m, that means you resample the bands at 30m and the resulting mosaic is no more a multi-size one, but a single size. AFAIK, the blending, in this case, doesn’t work if there are NaN values (it’s a JAI - the image processing library used - limitation).
    For the result you’ve obtained for multi-size, could you please specify the products used?



Hi Cosmin,

I used S2 tiles S2A_MSIL1C_20170618T105621_N205_R094_T30SVF_20170618T110415 and S2A_MSIL1C_20170628T105651_N205_R094_T30SVF_20170618T110445.

With the blending, if the same blend is applied across the overlap zone, will this not lead to lines (radiometric jumps) along both overlap edges?

For the parameter expression I thought this would allow filtering using flag or classification information to be used (as with band maths). Otherwise, now are bad (invalid, snow, cloud etc.) pixels to be removed?