BUG: Why does a remote polygon with no touching edge get affected by difference?

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of geopandas.

• (optional) I have confirmed this bug exists on the master branch of geopandas.

---

A polygon’s coordinates located on the top right gets affected by a difference operation on the bottom left. I expect that the upper right polygon shouldn’t be affected.

import pandas as pd
import geopandas as gpd

df = pd.DataFrame({"coords": [
    'POLYGON ((-100.5378104 45.175, -100.54 45.1712889, -100.5439257 45.175, -100.545 45.1776854, -100.5453305 45.18, -100.545 45.1803472, -100.5427045 45.18, -100.54 45.1789127, -100.5378104 45.175))',
    'POLYGON ((-101.9053297 44.3, -101.91 44.2978065, -101.9114672 44.3, -101.91 44.3012996, -101.9053297 44.3))',
    'POLYGON ((-101.9034962 44.295, -101.905 44.2944638, -101.91 44.2940891, -101.9126333 44.295, -101.915 44.2982986, -101.9158031 44.3, -101.915 44.3038377, -101.9132229 44.305, -101.91 44.3052182, -101.9090659 44.305, -101.905 44.3029345, -101.9001957 44.3, -101.9034962 44.295))'
]})
gdf = gpd.GeoSeries.from_wkt(df["coords"], name="geometry").to_frame()
gdf["value"] = ["a", "b", "a"]

gdf.plot()

diff = gdf.loc[gdf["value"] == "a"].difference(gdf.loc[gdf["value"] == "b"].unary_union)

print(gdf.iloc[0]["geometry"])
print(diff.iloc[0])

The coordinates change a bit; is this expected?

# original
POLYGON ((-100.5378104 45.175, -100.54 45.1712889, -100.5439257 45.175, -100.545 45.1776854, -100.5453305 45.18, -100.545 45.1803472, -100.5427045 45.18, -100.54 45.1789127, -100.5378104 45.175))

# difference
POLYGON ((-100.54 45.1712889, -100.5439257 45.175, -100.545 45.1776854, -100.5453305 45.18, -100.545 45.1803472, -100.5427045 45.18, -100.54 45.1789127, -100.5378104 45.175, -100.54 45.1712889))

[brendan-ward]

The original and difference geometries are equal but start from a different vertex, and are completely equal once each has been normalized (start from consistent vertex). It is not unexpected that GEOS returns a geometry starting from a different vertex from an operation like difference.

>>> import pygeos as pg
>>> original = pg.Geometry('POLYGON ((-100.5378104 45.175, -100.54 45.1712889, -100.5439257 45.175, -100.545 45.1776854, -100.5453305 45.18, -100.545 45.1803472, -100.5427045 45.18, -100.54 45.1789127, -100.5378104 45.175))')
>>> difference = pg.Geometry('POLYGON ((-100.54 45.1712889, -100.5439257 45.175, -100.545 45.1776854, -100.5453305 45.18, -100.545 45.1803472, -100.5427045 45.18, -100.54 45.1789127, -100.5378104 45.175, -100.54 45.1712889))')
>>> pg.equals(original, difference)
True
>>> pg.equals_exact(original, difference)
False
>>> pg.equals_exact(pg.normalize(original), pg.normalize(difference))
True

However, the best approach is as @martinfleis suggests - first use a spatial index and only calculate the difference for those geometries that intersect; those that don’t intersect can be copied directly from the input (left side of difference operation). This will also be much better for performance, since difference can be an expensive spatial operation for many or complex geometries.

[martinfleis]

Thanks @brendan-ward!