Geospatial nodes fail to work with geo data that has a PROJJSON CRS definition

[koettert]

To reproduce read the attached geo file and use the resulting data table in a KNIME joiner node afterwards. The joiner will fail with a parsing exception. The reason is that the Java code validates the CRS and the validation only supports EPSG codes or WKT-CRS string.
tl_2020_us_state.parquet.gz

[koettert]

Current implementation that converts the Python object to Java (see here):
` geo_column = geopandas.GeoSeries(column)

    crs = None
    if hasattr(geo_column, "crs") and geo_column.crs:  # highest crs level
        crs = str(geo_column.crs)
    elif (
        hasattr(data_frame, "crs") and data_frame.crs
    ):  # else second-highest crs level
        crs = str(data_frame.crs)
    if not crs:
        raise ValueError(
            f"No coordinate reference system(CRS) found, neither onGeoDataFrame nor GeoSeries  with"
            f"name '{column_name}'. Please specify a CRS."
        )

`

Possible solution that always either converts to epsg or WKT but which might loose information:
`
geo_column = geopandas.GeoSeries(column)

    crs = None
    if hasattr(geo_column, "crs") and geo_column.crs:  # highest crs level
        crs = geo_column.crs
    elif (
        hasattr(data_frame, "crs") and data_frame.crs
    ):  # else second-highest crs level
        crs = data_frame.crs
    if not crs:
        raise ValueError(
            f"No coordinate reference system(CRS) found, neither onGeoDataFrame nor GeoSeries  with"
            f"name '{column_name}'. Please specify a CRS."
        )
    # make sure to either return an EPSG string or a WKT string
    crs_str = str(crs.to_epsg())
    if not crs_str:
        # fallback to WKT which shouldn't be required
        crs_str = str(crs.to_wkt())

`

Ideally we would identify that the CRS is a PORJJSON and only in this case convert to epsg or WKT but so far I couldn't find a reliable way to do it.

Another solution would be to do the conversion only in the IO nodes where this could happen maybe as part of the clean_dataframe method of the IO file

[koettert]

Seems to work with latest Geospatial Extension and KNIME AP 5.5.0

the problem only appears if we use row-based but not for columnar based data transfer

[koettert]

The PROJJSON seem to be the standard format to represent CRS systems. So whenever someone reads a GeoParquet file we will have that problem.