How to…

Drop duplicate geometry in all situations

Using the standard Pandas drop_duplicates() function on a geometry column can lead to some duplicate geometries not being dropped, in certain circumstances. When used on a geometry columnm, the Pandas function compares the WKB of each geometry object. This is sensitive to the orders of various components of the geometry - for example, a line with co-ordinates in the order left-to-right should be equal to a line with the same co-ordinates in the order right-to-left, but the WKB representations will be different. The same applies for the order of rings of polygons and parts in multipart geometries.

To deal with this problem, use the normalize() method first to order the co-ordinates in a canonincal form, and then use the standard drop_duplicates() method:

gdf["geometry"] = gdf.normalize()
gdf.drop_duplicates()

The effect of the normalize() method can be seen in the following example:

>>> geopandas.GeoSeries([
...     shapely.LineString([(0, 0), (1, 0), (2, 0)]),
...     shapely.LineString([(2, 0), (1, 0), (0, 0)]),
... ]).normalize().to_wkt()
0    LINESTRING (0 0, 1 0, 2 0)
1    LINESTRING (0 0, 1 0, 2 0)
dtype: object