Hi!

I would like to get a handle for the underlying GDAL dataset behind a RasterIO dataset.
Is this possible somehow?

 

The reason I want to do this is that, in some cases, GDAL operations are significantly faster. I have a case, where reading a subset of channels in a window with the just-released GDAL 3.3.0 is approximately 25% faster than RasterIO.

from osgeo import gdal, gdal_array

rio_ds = rasterio.open(path)

gdal_ds = gdal.Open(path)

bands = [1,2,3]

resampling = rasterio.enums.Resampling.bilinear

win = ds.window(*bbox)

xoff, yoff, xsize, ysize = map(lambda x: int(round(x)), win.col_off, win.row_off, win.width, win.height)

rio_img = rio_ds.read(window=win, indexes=bands, out_shape=out_shape, resampling=resampling)

gdal_img = gdal_array.DatasetReadAsArray(gdal_ds, xoff=xoff, yoff=yoff, win_xsize=xsize, win_ysize=ysize, buf_xsize=out_shape[1], buf_ysize=out_shape[0], resample_alg=resampling, band_list=bands)

(Note that band_list is new in GDAL 3.3.0)

Best, Søren