No module named '_gdal_array'

[dfguerrerom]

As it was reported in https://github.com/sepal-contrib/sepal_smfm_biota/issues/42, there is a problem related to gdal, I've been doing some test in my test-venv, and apparently, this is happening because the installation is done using a cached wheel, is there a way to rebuild the wheels or to clean the cache? so the venvs can use a clean one?

[dfguerrerom]

I'm just copying and adding some comments regarding GDAL installation that were set in the sepal_biota issue tracker.

The issue there was that gdal couldn't find "gdal_array" module, and the reason is explained below:

• biota package is using gdal.Open.ds.ReadAsArray()() method to open and return images as Numpy arrays (check https://github.com/OSGeo/gdal/tree/master/swig/python#numpy).
• the array support is only installed by python gdal binding if Numpy is installed before, the installation will check if it is installed, if not, it will skip the array support.
• there is no way to change the package installation order in this repo since SEPAL deals with this.
• pip doesn't ensure that the installation order in requirements.txt is the same order as the packages are listed.
• adding gdal[numpy] extra_require doesn't work in gdal==3.4.3. There is a patched version that does this https://github.com/OSGeo/gdal/issues/2158.

To give a prompt I've decided to modify our forked version of biota and use rasterio to open images and return the Numpy array, it's been done here https://github.com/dfguerrerom/biota/commit/64286f4f627082135d5445cddf257a16be082b9b.