Converting PNG to GTiff increases file size using Python

Question

I am running this script to convert PNG to GTiff and tag it with coordinates and a crs of my choice. It works just fine but after the conversion the geotiff file is 177 MB (the png is 1 mb or smt) and it blocks me from working with the file. I am new to programming and i dont know where this is coming from. Below shown is the code which worked for me:

import rasterio

dataset = rasterio.open("input_file_path", 'r')
bands = [1,2,3]
data = dataset.read(bands)
transform = rasterio.transform.from_origin(9.171524, 53.057311, data.shape[1], data.shape[2])
crs = {'init': 'epsg:4326'}

with rasterio.open("output_file_path", 'w', driver='GTiff',
               width=data.shape[2], height=data.shape[1],
               count=3, dtype=data.dtype, nodata=0,
               transform=transform, crs=crs) as dst:
dst.write(data, indexes=bands)

Answer 1

You will need to use a compress option in your dst rasterio.open() statement. The docs at https://rasterio.readthedocs.io/en/stable/topics/writing.html barely mention this, but compress='lzw' is a starting point.

rasterio uses GDAL under the hood, so ultimately you can use all of GDAL's options to tweak the compression, which includes deflate and packbits as well as LZW compression, and different predictor settings. See https://kokoalberti.com/articles/geotiff-compression-optimization-guide/ for a guide to the options (written from a GDAL command line point of view) which also points to an article regarding using jpeg compression if you're using satellite imagery (orthophotos), which might compress them better than the original png.

The best combination of settings will depend on our image characteristics (data type as well as what's actually in the image). I haven't tried it in rasterio specifically, but in my regular use of GDAL with QGIS, a combination of compress='zstd',predictor=2 works well with integer data, predictor=3 with floating point, and lossy algorithms like jpeg make a big difference for orthophotos or similar, however I'm not sure which versions of rasterio support ZSTD compression.

Editing to add: I note from Why does rasterio compression reduces image size with single band but not with multiple bands that others have had trouble with compressing rasterio multiband images, like you are doing. Another option is to write your huge file as you are doing now, and then run it through GDAL itself as a 2nd step to compress it. You can do that via command line using gdal_translate with compression options specified there, or as a separate step using Python. For the latter, follow the breadcrumbs starting at Calling options/arguments of gdal_translate in Python or How to call gdal_translate from Python code?