FutureWarning: Assigning CRS to a GeoDataFrame without a geometry about dask-geopandas HOT 4 OPEN

Do you have a reproducible example? Or can you show the stacktrace if you turn the warning into an error?

I don't see anything with this simple example (using master of all):

import geopandas
import dask.dataframe as dd

df = geopandas.read_file(geopandas.datasets.get_path("naturalearth_lowres")).to_crs("EPSG:3857")
ddf = dask_geopandas.from_geopandas(df, npartitions=4)

ddf['area'] = ddf.geometry.area

from dask-geopandas.

I missed one important line, sorry. It happens with renamed geometry column.

import geopandas
import dask_geopandas

df = geopandas.read_file(geopandas.datasets.get_path("naturalearth_lowres")).to_crs("EPSG:3857")
df = df.rename_geometry('geom')
ddf = dask_geopandas.from_geopandas(df, npartitions=4)

ddf['area'] = ddf.geometry.area

Thinking about it now, we may have to fix that in GeoPandas, because you can reproduce it without dask as

df = geopandas.read_file(geopandas.datasets.get_path("naturalearth_lowres"))
df = df.rename_geometry('geom')
geopandas.GeoDataFrame(df, crs=4326)

With geometry column geometry this is either silent if you pass the same CRS or warns about CRS mismatch if you pass different one. With a different name, it does not see existing geometry in the first place. Even geopandas.GeoDataFrame(df).geometry breaks if you rename the column before.

from dask-geopandas.

Ah, yes, that seems indeed something to solve in geopandas (or at least the pure geopandas snippet should also not warn I think?)

from dask-geopandas.

Yeah, in the creation of GeoDataFrame from another GeoDataFrame we should capture that and pass metadata (geometry column, maybe crs for now) to a new one. I'll move this issue to geopandas repo.

from dask-geopandas.