@ChrisFairless could you check if this resolves your issue? Instead of storing CRS objects, I resorted to converting them into WKT strings when passing them to the Impact constructor

Yes this looks good! Thanks for the fix.

I like how the WKT is recognised by rasterio.crs.CRS.from_user_input so we can still use this to check CRS equality.

Agreed. In my experience, WKT is the one that gives the least amount of warnings. 😁
I've added the crs to the attributes list so it's clear that it is a string.

• but then ... maybe it's not. What if it's a rasterio.crs.CRS?

rasterio.crs.CRS supports to_wkt, too: https://rasterio.readthedocs.io/en/latest/api/rasterio.crs.html#rasterio.crs.CRS.to_wkt

I'll patch that.

@emanuel-schmid Fixed handling of rasterio.crs.CRS here: a5ab8bd

Many thanks! To me it's fine for merging, I've added a changlog entry. Feel free to edit.

@emanuel-schmid How should we handle the erroneous pylint issue here: https://github.com/CLIMADA-project/climada_python/pull/713/files#diff-1c276908f04d1a0ca5b1692295a586ce54b0937673711203582d3d72099ecdd0R45 ?

🤷 I's just suppress it for this particular line by adding # pylint: disable=no-name-in-module at the end.
There are only 2 of this kind throughout climada_python so changing .pylintrc seems over-reacting.