Category: PostGIS

Creating OpenStreetMap History Visualisations

I created a couple of OSM visualisations for my talk at the OSGeo Ireland conference.

See: History of OpenStreetMap in Ireland

These are pretty easy to make, but take a fair bit of time. I did mine for Ireland, but should work with any part of the world.

Required software:

  • PostgreSQL with PostGIS
  • Python
  • QGIS
  • osmium-tools

This is the trickiest part, installing osmium-tools: here.

Data:

An OSM full history export. The best source for these is GEOFABRIK.

For Ireland:

http://download.geofabrik.de/europe/ireland-and-northern-ireland.html

Due to GDPR, you will have to log in with an OSM id to download the full history extracts. User ID’s are personal data.

Process:

The workflow is pretty simple. Osmium-tools provides pretty easy API access to the history files, where you can provide a data, and it will extract what OSM was like at that date. We simply need to loop through the desired dates we want to extract, and pipe the results into a workflow that loads the data into PostgreSQL. The final step is simply rendering in QGIS using the time manager plugin.

Python Script:

Github GIST:

https://gist.github.com/HeikkiVesanto/f01ea54cca499a6a144d18cf8909c940

The tables in the database will be:

  • lines
  • multilinestrings
  • multipolygons
  • other_relations
  • points

Each feature will be tagged with the date it is associated with.

Visualisation:

To visualise the data in QGIS we use simply use the excellent time manager plugin, filtering on the load_date field and with a monthly interval.

Result:

Multi Ring Buffer – Buffer the Buffer or Incrementally Increasing Distance?

Does it matter, and who cares?

Multi-ring buffers can be useful for simple distance calculations as seen in:
X Percent of the Population of Scotland Lives Within Y Miles of Glasgow
And:
X Percent of the Population of Scotland Lives Within Y Miles of Edinburgh

For these I simply created multiple buffers using the QGIS buffer tool. This works for small samples, but was quite frustrating. I had initially hoped to do the whole analysis in SQLite, which worked pretty well initally, but struggled on the larger buffers. It took too long to run the queries, and did not allow for visualisation. I think using PostGIS would however be pretty feasible.

But creating a multi-ring buffer plugin for QGIS also seemed like a good learning experience. Which got me thinking, does it matter if you create increasingly large buffers around the original feature, or if you buffered the resulting buffer sequentially. My hypothesis was that there would be pretty significant differences due to the rounding of corners.

I asked on StackExchange but the conversation did not really take off:
http://gis.stackexchange.com/questions/140413/multi-ring-buffer-methodology

My question is not about the overlapping-ness of the buffers, since I think multi-ring buffers should be “doughnuts” anyway. But rather if smoothing will occur. The only answer was to try it myself.

Buffer styles:
Buffer the resulting buffer sequentially: Sequential
Buffer the original feature with increasing buffer distance: Central
[table caption=”Speed – In seconds”]
Features, Rings,Central, Sequential
1, 5, 0.59, 0.56
55, 5, 8.06, 6.38
1, 200, 60.83, 31.76
3, 200, 62.89, 40.89
55, 200, 628.38, 586.67
1, 2000, 203.84, 67.00
[/table]

No matter how you do it the sequential style is quicker, but that may be down to my code.

Rendering

Interestingly, although understandably, the sequential style results in a lot more vertices in the outer rings. For comparison, for a 500 ring buffer the outermost ring had the following vertice counts:
[table]
Style, Vertices
Central,488
Sequential,30918
[/table]

We can see this with editing turned on.
Central:
Central_editing
Sequential:
Sequential_editing

We can also see a smoother profile in the sequential buffer. However the difference is not major, and hard to discern with the naked eye.

So we have at most about around a 10m discrepancy, with 500 50m rings, so around 25000m of distance from the original feature.
Screenshot[34]
This impacts rendering time dramatically, an example with our 500 rings:

Central:

Sequential:

So quicker to create but slower to draw. So which one is better, quicker calculation, or quicker rendering? Or should we not do 200+ ring buffers?

Hard to say. In version 0.2 of the Multi Ring Buffer Plugin. There is an option for either in the advanced tab.

Plugin: https://plugins.qgis.org/plugins/Multi_Ring_Buffer/
Please report any issues through GitHub: https://github.com/HeikkiVesanto/QGIS_Multi_Ring_Buffer/issues