Rasterio applies a random error to all pixels when saving a raster

Question

When I read a raster and then save it with rasterio the pixels are sligthly modified. Code example:

import rasterio

inputFile = './band3.jp2'
with rasterio.open(inputFile) as b3:
    v3 = b3.read(1)
    print('Before: ',v3)
    profile3 = b3.profile

##Saving raster
output='./test.jp2'
with rasterio.open(output, 'w', **profile3) as dst:
    dst.write(v3, 1)

##Reading the raster I just saved
with rasterio.open(output) as b3:
    v3 = b3.read(1)
    print('After: ',v3)

This outputs:

Before:  
         [[  0   0   0 ... 665 632 661]
         [  0   0   0 ... 728 687 704]
         [  0   0   0 ... 703 702 700]
         ...
         [  0   0   0 ... 613 605 603]
         [  0   0   0 ... 610 600 605]
         [  0   0   0 ... 608 606 607]]

Warning 6: driver JP2OpenJPEG does not support creation option TILED #I don't think this is important

After:  
        [[ 11   0   0 ... 663 629 655]
        [  0   0   0 ... 717 680 687]
        [  0   0   0 ... 689 693 687]
        ...
        [  1   0   0 ... 604 597 587]
        [  3   0   0 ... 604 595 593]
        [  0   0   0 ... 594 594 589]]

I'm not sure if it is important because I didn't modify it but the profile looks like:

{'driver': 'JP2OpenJPEG', 'dtype': 'uint16', 'nodata': None, 'width': 10980, 'height': 10980, 'count': 1, 'crs': CRS.from_epsg(32615), 'transform': Affine(10.0, 0.0, 600000.0,
   0.0, -10.0, 1600020.0), 'blockxsize': 1024, 'blockysize': 1024, 'tiled': True}

Answer 1

As @mikewatt commented changing the driver to 'GTiff' works:

profile3['driver']='GTiff'

Answer 2

JPEG2000 can be lossy or lossless. The GDAL JP2OpenJPEG driver is using the lossy method by default unless a lossless compression is explicitly requested with the creation option "REVERSIBLE=YES". See the driver documentation https://gdal.org/drivers/raster/jp2openjpeg.html

REVERSIBLE=YES/NO : YES means use of reversible 5x3 integer-only filter, NO use of the irreversible DWT 9-7. Defaults to NO (unless the dataset is made of a single band with color table, in which case reversible filter is used).