Next steps on network simplification

[Robinlovelace]

We now have a prototype simplified network, as described here: nptscot/networkmerge#58

I'm opening this issue here because ultimate the code that builds the simple network will live in this repo, tests in .R, .Rmd, or .qmd files can go in the code/tests folder.

Next steps that come to mind:

• Try installing R 'build tools' to get dev versions of stplanr and rsgeo
• Re-join for a small random sample across Scotland, using the new crs = "EPSG:27700" argument of rnet_merge() (this may depend on upstream changes in stplanr, see bug in geo_buffer ropensci/stplanr#538 and Bug in line_segment() when using certain values on projected data with rsgeo implementation ropensci/stplanr#546 )
• Try rnet_merge() on projected data, with x and y in EPSG 27700 CRS
• Try doing simplification only on certain segments, e.g. those on which cycling under the Go Dutch scenario is greater than some threshold (e.g. 500) and segments within a threshold buffer around such segments, or by identifying dual carriageways or other road types in need of simplification. This could involve
  1. Filter parts of the input route network based on OSM data 'rnet_x' to generate a minimal dataset
  2. Remove the filtered dataset from the original to avoid duplicating them
  3. Subset the rnet_y object (OS Open Roads) to only those within a buffer surround the rnet_x and merge on the attributes. Use the following syntax to get only OS roads that are within the buffer, e.g. with the following syntax:

rnet_x_selected_buffer = st_buffer(st_union(rnet_x_selected), dist = 50)
rnet_y_selected = rnet_y[rnet_x_selected_buffer, , op = st_within]

5. rbind() the resulting merged network back with the original rnet_x (the version that excludes the lines to simplify)
6. 🚀 pmtiles

• Check to see if merge is better in places far from the centre of Scotland (hypothesis: yes)
• Share the updated GeoJSON for tiling
• Think about how to keep segments that are not in the open roads dataset but which are in OSM, thoughts?

One idea on the last point: only do simplification on segments that have high cycling potential (e.g. more than 100) under the Go Dutch scenario or which are contained by a buffer around those high cycling segments.

And points from @mem48 in nptscot/networkmerge#8

• Outputs need to be transformed to 4326 current output is in the British National Grid
• The simplified network is still based on old data with the wrong columns, please use https://github.com/nptscot/outputdata/releases/download/v2023-09-09/combined_network_tile.zip
• Please use appropriate coordinate precision to reduce file sizes

In terms of re-building the national network, I'm thinking about how to speed-up the process. As a bonus, you could take a look at and try to fix this: ropensci/stplanr#522

All this will help to fix #64 have a simplified network up soon!

[Robinlovelace]

Good conversation with Zhao and I this morning, we talked about testing on a small network first and ways to only do the simplification on a subset of 'rnet_y'. Next steps: @wangzhao0217 to iterate on this, report progress here and update code in #292. Stretch aim: merge that PR based on results for a small subset of the network based on the latest version of {stplanr} by end of the day.

[Robinlovelace]

Looks like there's a new release here: https://github.com/nptscot/npt/releases/tag/simplified_network_merge

Can you point us to the code that generated this dataset @wangzhao0217 ? We should get #292 merged.

@mem48 can you take a look at get get this onto our test website if good?

[wangzhao0217]

Looks like there's a new release here: https://github.com/nptscot/npt/releases/tag/simplified_network_merge

Can you point us to the code that generated this dataset @wangzhao0217 ? We should get #292 merged.

@mem48 can you take a look at get get this onto our test website if good?

Hi @Robinlovelace @mem48 ,
Please find the latest results here https://github.com/nptscot/npt/releases/download/simplified_network_merge/rnet_simplifed_20_15_20231015_130303.pmtiles
It doesn't resolve all the issues, but it addresses 3 out of 4.

[Robinlovelace]

Good stuff Zhao. It's moving in the right direction and better than the tiles currently hosted on our test website!

[Robinlovelace]

This is looking great to me:

Source: https://nptscot.github.io/#12.06/55.95122/-3.17407

Credit: @mem48 here: nptscot/nptscot.github.io@6cfb364

This is good enough for now. @wangzhao0217 can you open a new issue to document any remaining issues? Whether or not we prioritise fixing these additional issues is TBC.