centrality measures

[temospena]

Using sfnetworks, centrality and betweeness produces weired results

Maybe check with igraph, of use more road links from OSM

[szhorvat]

Note that closeness centrality is not meaningful for disconnected graphs.

I do not know your dataset, but the presence of a few vertices with very high closeness suggests that these exist in tiny connected components, thus they are "close" to all other vertices which are reachable from them. I'd check if the network is connected. If not, closeness is not a useful metric.

[temospena]

Thanks @szhorvat for the heads-up!

I added more OSM highway tags, did some network cleaning and topology validation in Grass, and the results really improved.
Well, at least are not almost all zero (although x10^-9)

We'll see if this is enough for now...
bcd22ec

[temospena]

I think that we should remove the unnecessary nodes from the OSM. This will prob affect the centrality measures, but provides a more realistic result.

See https://geoffboeing.com/2016/11/osmnx-python-street-networks/
https://github.com/anerv/NERDS_osmnx
and

Note that the simplification function only simplifies and merges edges that have identical values for important attributes such as geometry type, bicycle direction, and protection level.

Non-simplified network (left) and simplified network (right).

[temospena]

see here: https://osmnx.readthedocs.io/en/stable/user-reference.html#osmnx.simplification.simplify_graph

[temospena]

@Robinlovelace would you know if this is possible to do in R or with R/qgisprocess ?

https://github.com/gboeing/osmnx/blob/4f23b37b9a21c0fb513e58171d759bbd8b87ce60/osmnx/simplify.py#L207

My first thought would be to have a OSMX pluggin in QGIS and do from there (with qgisprocess connection).
But there isn't.
I thing that grass has not such option to simplify topology by removing unnecessary nodes.

Have you faced this issue before?
Thank you!

[Robinlovelace]

Hi @temospena interesting and important problem! Yes I think it can be done in R, including with sfnetworks but have never done it myself. One link that may be helpful (not in R) is this: https://github.com/anisotropi4/parenx Python package.

[temospena]

• include code from code/centrality.R regarding "Remove duplicate nodes" (uses sfnetworks) in code/centrality_qgis.R and assess the differences in centrality measures.

[temospena]

this was documented here
https://luukvdmeer.github.io/sfnetworks/articles/sfn02_preprocess_clean.html#smooth-pseudo-nodes

[temospena]

I managed to clean them with

net = as_sfnetwork(road_network_clean)
smoothed = convert(net, to_spatial_smooth)

nodes_smoothed = smoothed %>%
  activate(nodes) %>%
  as_tibble() |> 
  st_as_sf(coords = c('X', 'Y'))

After v.clean grass process:

Clean (smoothed):

[Robinlovelace]

Fantastic, great to see and thanks for sharing @temospena. Cc @juanfonsecaLS1 who's doing some network processing as a preparatory step on the pathway towards minor traffic estimation.

[juanfonsecaLS1]

This is very interesting work @temospena. I have been using dodgr for a similar analysis, the contracted dodgr graph simplifies the geometries and exclude redundant nodes.

Also, are you normalising the betweenness centrality? There is a useful publication on this here

[temospena]

Thanks @juanfonsecaLS1 ! This is indeed a very interesting paper.
Looking at repo here: https://github.com/christian-werner/spatial-centrality-pub

So, maybe I am using to many different packages to achieve the same purpose 🤔

What I am doing is:

• Get OSM highways
• Remove the edges from the disconnected groups
• Using grass7:v.clean (with break, snap, rmdangle, rmdupl, bpol)
• converting to sfnetwork and using spatial_smooth to remove the redundant nodes
• using grass7:v.net.centrality to estimate degree, closeness and betweness for each node (it provided better results than sfnetworks) - I would say that this uses betweeness centrality instead of spatial betweness centrality, but not sure
• get the mean of these values in a grid

What we are interested is in a mean of re-scaled centrality measures in a grid.
Here is the paper from @valenca13 that we are trying to reproduce and replicate for any city.

So, using dodgr would simplify the process?

[Robinlovelace]

This is awesome Rosa, loads of great detail. I discussed it with Juan yesterday, looking forward to seeing what he comes up with that's related to this stuff (to estimate motor traffic in our case).