MultiLineString in routed triangulation

[marianamirandapessoa]

In the case of Turin, there is one routed edge of the triangulation that is represented by a MultiLineString.

Steps to verify the existance of a MultiLineString:

1. Add a checkpoint in the middle of growbikenet.growbikenet.growbiknet()

def growbikenet():

    (...)

    grown_bikenet_edges = create_gdf_with_geoms(grown_bikenet_edges_abstract, edges)
    progress_bar.update(1)

    grown_bikenet_edges_checkpoint = grown_bikenet_edges.copy()

    (...)

    return edges_ranked, grown_bikenet_edges_checkpoint

2. Run growbikenet() for Turin

from growbikenet import growbikenet

edges_ranked, grown_bikenet_edges_checkpoint  = growbikenet(
    city_name = "Turin",
    export_data = False
)

3. Check the geometry types of grown_bikenet_edges_checkpoint

>>> grown_bikenet_edges_checkpoint['geometry'].geom_type.value_counts()
LineString         319
MultiLineString      1
Name: count, dtype: int64

Supposedly, all routed edges of the triangulation should be represented by a LineString geometry, as pointed out in growbikenet.functions.create_gdf_with_geoms()

def create_gdf_with_geoms(df, edges):
    
    (...)

    # Merge multilinestring into linestring where possible (should be possible everywhere)
    gdf["geometry"] = gdf.line_merge()
    return gdf

For Turin, there seems to be an edge in the triangulation where the routed version has a loop, and therefore, the geometry is kept as a MultiLineString, with 3 LineStrings.

This bug was found in the context of building the function growbikenet.functions.add_point_data_to_net()

[marianamirandapessoa]

The Problem

The MultiLineString represents a case where the shortest path between the two seed point nodes (1693872406 and 837235170) has an apparent visual loop, but in the real network it represents a highway and a local road that have been built at different height levels.

The problem was found to be in the use of the function shapely.union_all(). In this case, the function inserts a point where the LineStrings visually intersect and separates them, ultimately changing the geometries of the real network.

Below is a simplified example of the process.

Taking this example, the merging of the geometries' intended result should be LineString([A,B,C,D,E,F,G]), but instead it is MultiLineString(LineString([A,H]), LineString([H,B,C,D,E,H]), LineString([H,F,G])

Because the 3 LineStrings share an endpoint (the created point H), the function geopandas.GeoSeries.line_merge(), which is used afterwards, doesn't dissolve the MultiLineString into one continuous LineString.

Proposed Solution

Replace shapely.union_all() with joining everything into a MultiLineString.

This way:

• No new point is added
• The geopandas.GeoSeries.line_merge() is able to simplify the MultiLineStrings.

Old	New
def create_gdf_with_geoms(df, edges): (...) df["geometry"] = df.path_edges .apply(lambda x: edges.loc[x].geometry.union_all()) (...) return gdf	def create_gdf_with_geoms(df, edges): (...) df["geometry"] = df.path_edges .apply(lambda x: MultiLineString(list(edges.loc[x].geometry))) (...) return gdf

[mszell]

Thank you for investigating so thoroughly!
The solution looks great, please feel free to implement it in a new branch, see if the tests work, and open a pull request.