Updating spatial grid for spatial interaction model

[Robinlovelace]

Currently we're using this code to get a fixed grid of points 500 m apart in x and y directions:

npt/_targets.R

Lines 748 to 752 in f420b5f

	tar_target(grid, {
	grid = readRDS("./inputdata/grid_scot.Rds")
	grid = sf::st_transform(grid, "EPSG:4326")
	grid = grid[region_boundary_buffered, ]
	grid |> sf::st_transform("EPSG:27700")

One idea in https://github.com/nptscot/npt/blob/main/report-utility-methods.md#next-steps is:

Revisit the grid: should a more approach e.g. involving spatial clustering of points be used?

I know @Hussein-Mahfouz has looked at spatial cluster with DBSCAN. Could this be useful here?

Cc @joeytalbot and @wangzhao0217 as discussed today.

Not an issue just a Phase 3 idea so far hence putting in discussion not issue.

[Hussein-Mahfouz]

Is this related to the work on sampling subpoints for odjitter? Related atumie issue https://github.com/ITSLeeds/atumie/issues/975
Or is this a different problem?

[Robinlovelace]

Different but related.

Instead of using all possible points (POIs from OS data primarily), we're using a regular grid of points as the subpoints, with the weight related to the characteristics (currently just the number) of trip attractors in each cell. The idea is to not use a regular 500 m grid but a grid that becomes smaller towards places with high population/activity density and larger in more remote/rural areas.

[Hussein-Mahfouz]

Short answer: I'm not sure - we'd have to try to find out. Longer answer:

DBSCAN takes two parameters: epsilon (search radius) and minpts (minimum number of points required to form a cluster). These parameters are set globally, so you can't optimise for areas of different point density simultaneously. Take two scenarios

1. Radius is small (better for areas with high POI density)

   • In high density areas, it will cluster nearby points
   • In low density areas, the radius may be too small to cluster points

2. Radius is big (better for areas with low POI density)

   • In high density areas, you might find too many points in the same cluster. 
   • in low density areas, it might give you better results

It may be possible to cluster different areas separately (e.g. inner city alone, and outer city alone) with different parameters