Test break because difference in glacierareas map

[JoostBuitink]

Automated test break, because of a difference in the glacierareas map in the staticmaps. It appears that an change altered how a vector (or GeoDataFrame) is rasterized onto the model grid. Is potentially caused by an upstream change in rasterio.rasterize: but I couldn't find an clear change in their code.

See figure below, showing the results of rasterizing a glacier GeoDataFrame onto the model grid, the difference is in the red outline. Left shows the results created with rasterio 1.2.10 (and hydromt 0.6.0 and hydromt_wflow 0.2.1), right with rasterio 1.3.4 (and the master branches from hydromt and hydromt_wflow)

Note that the glacier geometry does indeed cover the topright pixel (glacier geometry in grey, and the green area is the pixel that is no longer considered to be a glacier with raster 1.3.4)

[hboisgon]

Thanks for checking @JoostBuitink. Seems gdal made a 3.6.0 (from 3.5.*) and rasterio 1.3.4 was released around the same time to support gdal 3.6+.
If it's a change in GDAL not really sure if we can do so much about it...
So just update our test example staticmaps to reflect the change?
@DirkEilander would you agree?

[DirkEilander]

The point is that with the new results seem wrong as we use the "ALL_TOUCHED=TRUE" options of gdal rasterize. Perhaps good to investigate if this is a (known) bug in the new version or that the argument is no long in use or ... If it's a (known) bug it might be better to ignore the error (or temporarily pin to an older gdal version) until it is fixed. Otherwise we'll need to change the map back once it is fixed.

[hboisgon]

I tried to look if the error was mentioned already, gdal already released two versions with fixes for 3.6 but could not find anything about rasterize and all_touched mentioned yet. Maybe we could raise the issue indeed?

[DirkEilander]

I also can't find anything. To raise the issue we first need to make sure that this behavior is reproducible with just rasterio (or just gdal rasterize) by creating a minimal reproducible example of the error. We could create this with a clip of the glaciers layer and a wflow map.

[hboisgon]

For now we update the glaciers map and hope this will be solved later

[JoostBuitink]

It appears in the latest version of rasterio (assuming the problem was here), this is fixed again. Tests in #123 were failing on the glacier maps, with the same issue as presented here (but then inversed). In this PR, I reverted the glacier map, as this is expected behavior with all_touched=True. This way, the tests can also be completed successfully