Tuesday, 3 December 2019

Land doesn't vote, but it does matter

A short post today on the well-work topic of election mapping and, specifically, how to represent the results of elections in a way that reflects the true proportions of who voted for who. This isn't an attempt at a solution or a definitive answer, just some ideas and maps to provoke further thinking on the topic. Let's start by looking at two election maps (below). The one on the left colours in all constituencies according to the party of the MP at the end of the most recent UK parliamentary session. The one on the right is coloured in the same way but only where there are buildings. People like to use these kinds of maps with accompanying comments like 'land doesn't vote' or 'because sheep don't vote' (although I would like to see some polling on how they'd vote if they were so enfranchised). Scroll to the bottom to see the method for this in QGIS.

Standard map vs 'dasymetric' map
Some obvious stuff first... There's a lot of blue on the map to the left above. There is also quite a bit of yellow, and not much red. So, when we look at the map on the left it might give the impression of Tory dominance, when in fact they are not as far ahead as the colour share would suggest and  things are nowhere near as bad for Labour as it might appear from the colour share. Reasons include population density, urban voting patterns, etc, etc. But, but, but, when we get to this point, I always think "yes, true, but the vast majority of people know that". But then again maybe the vast majority of people have more exciting lives and don't think about this kind of stuff at all and end up being inadvertently misled my maps. Perhaps. SO, ENTER THE HEXAGON!!! Yes, it's our old friend, the hex map - example below from a new book. This definitely helps shift the story of the conventional election map from 'wow, the Tories are crushing the competition' to 'it looks quite close'. I quite like this approach, but with only a small quite and probably because of my own biases.

Hex for the win?

Yes, this hex map is better in many respects, but I'm not a huge fan of using these on their own and wouldn't really mind not seeing them again. But I do agree it provides a more useful representation of the vote share and as a compromise it's a fairly good one if we're interested in displaying vote share. I just think in this case a bar chart might be better if we're most interested in vote share, given how hard it is to get the hexagons arranged in a way that matches the underlying geography and what it does to my brain. That's why I've been experimenting recently with the dasymetric map style you see above. Talking of which, here are a few zoomed-in versions, with place labels, and a bit more discussion below.

What colour is your town?

This is a bit more informative I think

But I wouldn't use this approach on its own

Although I do think there is a place for this style

The colour balance is a bit more satisfying here

The last thing to say in this bit is that I also like the buildings-only maps with labels as it allows us to say a bit more about specific places. Let's imagine for a moment that there was a human being out there who didn't know the precise boundaries of the Berwickshire, Roxburgh and Selkirk constituency. I know it's hard to believe but I'm told such people do exist. They know it's a blue constituency and on the first map above they can actually see Selkirk (to the bottom right of the first labelled map) in blue, rather than a giant block of blue covering the entire area. 

Of course, this overlooks the fact that not everyone in the constituency voted for the Conservatives but that's another issue, and in a first-past-the-post electoral system I think shading by winning party is quite logical. Anyway, my point here is really that we can focus in on individual towns and how they voted without being distracted by giant wodges of colour that strictly speaking we do not need. I think the North of England map above does a good job of this - so you can just look at the east coast  of England and see that Bridlington or Scarborough are represented by the Conservatives and that Blackpool has a red wedge between two blue blobs on the west coast.

If you did want to colour areas by the strength of vote for one party then you could use what I call a 'staunchmap' approach, like the example below that I've shared previously. The deeper the colour the higher the vote share - e.g. see Merseyside and its deep Labour reds.

Who will win in 2019? I predict nobody will win.

So, what's my final verdict on how best to map elections? I think a mix of approaches is best, and because most of what we view is on multi-media platforms I think some use of animation might be useful, as in the gif I made already and shared on Twitter. In print or static format this could just be a simple side-by-side version, like the image at the top of the page. Actually, having just written this, I think broadcasters and print media should do both - perhaps standard map + hex + dasymetric or standard map + one other. Not that it matters that much but if we've got the data and tools seems like it would be a good idea to use them here and, maybe, pass on a little more information with this slightly different encoding.


Ooh, fancy - but it does add a bit of value I think


("land doesn't vote, but it does matter" is my usual thought on this, so I'm adding it here, and that's why I'm using a technique that does at least show the land at some point, before switching to a view that more closely reflects the numbers we're probably interested in. Plus, we can't have the Highlands zapped from the map, or at least not permanently.)


Data and method: a cumbersome way to do this would be to use a buildings layer to clip the constituencies but that would take ages. I did this in QGIS using a simple blend mode but it's possible with other tools as well. The data I used are from the spreadsheet I put together and shared on Github previously. You'll also find a constituencies file there to to join it to. You can grab the buildings layer from this page and then you've got all you need to replicate what you see here. Just take a look at the screenshots below to see how I did it. Just make sure your map canvas background is black (via Project > Properties > General).

This is the layer order you need

The constituency colours come from the attribute table

The buildings are all white, with a thin outline


Friday, 8 November 2019

Amazing 3D rendering with Aerialod - a tutorial

One of the more exciting developments in recent years for those of us into geospatial things is the arrival at the end of October 2019 of Aerialod by ephtracy. What am I talking about? I'm talking about being able to create the kind of images you see below in only a few minutes using free software and open data. Scroll past the images for some tips on how to do this, and note that this write up was completed on 8 November 2019 and refers to v0.0.1.

The Cuillin ridge in Skye

Part of the City of Belfast in Northern Ireland

This is indeed a little bit of London

A little circular bit of London

This is also London, for Millwall fans

The Cuillins in Skye, featuring Sgùrr Alasdair

This is Sheffield, around the area where I work

Okay, so how do you get up and running? First of all, go to the Aerialod website and download the package you need. It's Windows 64- or 32-bit only for now and you just download and unzip and then run the .exe to launch Aerialod.

You do get some sample raster data in the zipped download (in the 'map' folder) but if you also download the 'Sample Maps' archive next to the software download button you'll get a central Manchester Lidar png and a Mars (yes, the planet) png. This is a nice reminder that Aerialod is able to handle different formats, including .asc, .png. and .tif for example. I haven't tried any other formats though I think you'll be okay with .jpg too.


They don't look like much here, but wait and see!

When you launch Aerialod you'll see something pretty much like the image below - and it will have that blocky sample layer in there. This is useful for playing around with so you can get to grips with navigation etc. Just note that when you zoom or move around Aerilod may briefly look pixelly/fuzzy as it re-renders, so don't worry about that. It sharpens up perfectly once it's done, although with more complex layers it takes a bit longer.

You may be a bit bewildered at first, but it won't last long

Before I forget, be sure to look in the config folder and open the hotkey.txt file, which I've shown below. That's really useful. But I find mouse navigation easier, so read on. Also note that the second section of hotkey actions below combine a left mouse button click with keyboard actions too.


Hit D and be amazedx

If you are finding the interface too small and can't actually see the icons easily (e.g. if you have a 4k monitor or something like that) then you can use CTRL + or - to scale the UI but you could also just edit this bit of the config.txt file (in the config folder) so instead of 1.0 it says 2.0, like below:

view :
{
// 0.5 ~ 3.0
ui_scale : '1.0'
}


Okay, we're all set now so here are the basics of moving things around:


  • Scroll wheel/middle mouse button - you can scroll forwards and backwards to zoom in and out and with the button pressed down you can position the layer wherever you want. 
  • Right mouse button - tilt/pan/rotate etc. Just have a play and you'll see what I mean.
  • And of course with keys, as above, W to zoom in S to zoom out, D to rotate clockwise and A to rotate anti-clockwise.

You loaded the software, figured out navigation with the sample data but now want to render some real world stuff. See below for how to do that.

There are tons of sources for this, including things like NASA's 30 metre SRTM but really it's going to look best with high-resolution DSM or DTM data and for this Lidar is ideal. On the Aerialod page they link to two potential sources of this - the UK's Defra Lidar page where you can download a variety of 25cm, 50cm, 1m and 2m Lidar data for England or get NASA's HiRISE data for elsewhere in the universe.

To get data into Aerialod, the easiest method in my opinion is just to drag and drop a raster file straight in, so that's what I just did with the Manchester.png layer you can see in the screenshot below. To be clear, what you see below is just the result of me dragging and dropping a png file into Aerialod. I haven't done anything else yet.

That was easy!

Before I say more about settings, a word on getting data into Aerialod in other ways. In the top right of the window you'll see a save icon. That will save the raster layer (not the 3D render) to a location of your choice. The next button (the open folder icon) will let you open a new single layer instead of dragging and dropping into the window. The next button (the closed folder icon) is a bit different, but it effectively allows you to stitch together all layers in a single folder, and it's amazing. Just click one item in the folder and it will add them all, as you can see below. The final button (blank page) just starts a new Aerialod blank view.

Stitch multiple files!

Okay, see below for where I've used the open folder button to add a single layer. This is for a single tile (of about 7MB) near the University of Sheffield. In the image below this, I've used the closed folder icon to select the first item and then all layers are added. You don't actually have to select the first item though, just any of the files in the same folder. This is really handy as when you download Lidar data it's more than likely going to be comprised of lots of little chunks as individual files.

Single layer from within a folder

All layers from within a folder

If you get too greedy and try to add a gazillion zigabytes of data, Aerialod may crash. I know, I've tried. Anyway, for the rest of this tutorial I'm mainly going to use the Manchester.png sample file provided on the Aerialod download page - and what you see below is the result of me just dragging and dropping it into the viewer. This covers a good chunk of Manchester city centre and also a bit of Salford and Trafford and to the left of the image below you might just spot Old Trafford - both the football and cricket ground.

Just another sunny day in Manchester


Okay, so the rest of this will cover the main things you need to know but I can't cover everything because a) that would take too long and b) I don't actually know enough to tell you everything and I'm kind of learning as I type here. So the two annotated screenshots are my main contribution for now. Look at the images closely though and you should learn enough to produce great visualisations in not much time at all.

Click on the next two images for a little explainer of what does what and then try it yourself.

This covers the basics, click to expand

Some more tips - including on Focus


Offset - actually quite useful for flood viz
A full screenshot so you can see some of the settings here

That's basically it for now. All that's left to say is don't forget to check the #Arielod hashtag on twitter as well as the @ephtracy Twitter feed. Last of all, don't forget that you can drop all kinds of things into Aerialod, not just terrain models and suchlike. I've experimented with adding in photos of people, which often looks sort of cool but also weird, and I've tried all sorts of other stuff but I'll end with one of the MiniScale raster relief maps of Great Britain which are part of their open data offering. It's not intended to be used this way but I think it looks quite interesting.

This actually turned out alright


Saturday, 2 November 2019

A deprivation by constituency chart

Yesterday I decided to update a little chart I made after the 2017 General Election. It was inspired by a histogram that Owen Boswarva made and the idea was very simple: put England's 533 constituencies into 10 columns, with the most deprived on the left and least deprived on the right, and then colour it by party. The image below is the result. UPDATE: I have now done a full-UK version of this - see below. Also includes an animation and the individual frames which show one party at a time. Read the notes on the UK chart for more information.

Link to slightly higher resolution version

Full size version

Easier to decipher as a gif

Labour

Conservative

SNP

Liberal Democrats

DUP

Sinn Féin

Plaid Cymru

Green

Independent (at 2017 General Election)

OORDEEER!


I did this out of curiosity the last time and then after speaking to my colleague Philip Brown about data, elections and suchlike I decided to update and try to improve the older chart, which was informative but a bit of an assault on the eyeballs. I then saw that the House of Commons Library team re-ran their analysis aggregating the English Indices of Deprivation for the 533 English constituencies, so that was all I needed. And, by the way, the House of Commons Library team are in my opinion doing some of the best data curation, manipulation and analysis out there - really great team. 

Of course, the results are hardly surprising but I didn't expect the sorting to be quite so stark. Naturally I tweeted the graphic and lots of people also found it interesting. So, I've done a slightly different version below which has the constituency names in the boxes - but you'll have to click and zoom to read those.

Full-size version here

Zoomed-in extract of the image above


It's not possible to do a full UK-wide one using a single dataset from the same time point, because the deprivation indices for each country of the UK are slightly different and cover different time period but it would be nice to have been able to - and at least in Scotland the colours would be quite spread across the deprivation spectrum. UPDATE: as you can see above, this last statement is true and I've also done the full-UK chart.

I also did one more version of the graphic, this time in a very long single column vertical one, the original of which is here, with a lower resolution one below. The tends to work great on a mobile but needs a bit of zooming in and out in most browsers. I've added the 'required swing %' figure to this one, showing what percent vote shift it would take for a constituency to change hands.

Click to see full size


Looks like this when full width on screen


There wasn't any great agenda or rhyme or reason behind this, I just wanted to see what it looked like and in particular how much the colours would be grouped and where the obvious anomalies were.


Given what I do for a living I'm duty-bound to point out the following obvious but important things:


  • Correlation is not the same as causation (yawn! but true).
  • I don't believe voting Labour makes you poor, or that voting Conservative makes you rich, as some people online seem to have implied - or even that 'the poor' vote Labour or 'the rich' vote Conservative - and I certainly wasn't trying to demonstrate either of these but understand that's sometimes how people see things.
  • There is a reasonably high amount of variation within most constituencies in relation to levels of deprivation, although at the top and bottom of the scale a lot less than you might imagine - I'm currently working on a project all about this kind of thing.
  • Colours - they are the html values for the party colours from the respective party websites.
  • The aggregation method used to derive the constituency rankings could be done a number of different ways, and therefore can produce slightly different results based on how you do it but the order of places would always be broadly the same.

That's all for now.

Monday, 2 September 2019

Globe projections and insets in QGIS

Update, January 2022: this tutorial may still work (although you will also notice a WKT option in the projection settings now, and the proj method is not recommended any more) but what I'd really recommend is that you use the Globe Builder Plugin (just go to the Plugins menu in QGIS and search for it). 

Proper QGIS boffins may remember Hamish Campbell's excellent post on the topic of 'Azimuth Orthographic Projections with QGIS' from 2014, in which he described a method for creating maps with an azimuthal orthographic projection - or, what non-boffins might call a 'globe view'. The good news is that in QGIS 3.4 you can do this very easily using the built in The_World_From_Space projection and you can easily create new ones with the view of the world you want. See below for an example and then read on for the method. The great thing is that you can also have a globe view and a normal, flat view in the same Print Layout as QGIS now supports multiple projections in the same layout.

The graticules are from Natural Earth

Okay, how to find this 'world from space' projection? The easiest way is to click the little CRS button at the bottom right of the QGIS window, or go via Project > Properties > CRS. Then you just search for it in the CRS search box, as you can see in the screenshot below - where you'll also see a couple of modified CRS versions I created. Just in case you don't know, CRS stands for coordinate reference system and they all have an EPSG code so that's why you'll see 'EPSG' and some numbers. If you're a proper carto boffin you'll know your 4326 from your 3347.

You can see the properties of the projection here

Okay, so that's how you set it. How to create a view centred on where you want? For that, you first need to copy a bit of the text in the above box, which I've pasted below so you can try it yourself.

First, copy this text: 

+proj=ortho +lat_0=42.5333333333 +lon_0=-0.53333333339999 +x_0=0 +y_0=0 +a=6370997 +b=6370997 +units=m +no_defs

Then replace the 42.5333333333 wih the latitude you want and the -0.53333333339999 with the longitude you want. You don't need so many decimal places! So, for example, I did this for New Zealand and this gives me:

+proj=ortho +lat_0=-40.5333333333 +lon_0=-157.3333333339999 +x_0=0 +y_0=0 +a=6370997 +b=6370997 +units=m +no_defs

You can see by looking at the numbers above that I have centred the projection on 40 degrees south and 157 degrees east. In order to create this new projection you just need to go to Settings > Custom Projections > click the green plus symbol and then at the bottom of the window give your project a name and then paste in your new projection information into the box. See below for what this looks like for my NZ projection.

Click OK to activate it

Once you've created a new projection you can just go back to Project > Properties > CRS and search for the new projection and use it. This is what the NZ one looks like, below.

This is not a map without New Zealand

I mentioned above that you can have a globe view like the above in a Print Layout with a different, flat projection in the main map. This is very easy. See below for an example where I have the globe view projection in the top left and a different projection with the same layers in the bottom of the image - in this case using the Equal Earth projection.


One Print Layout, two different projections

This is where you set the CRS for each map

I've put this together quickly so it's not very polished but see below for an example of the kind of thing you could do using this method - I quite like the globe locator inset, and it's also used by a good few news organisations, including the BBC. This used to be quite a lot of hassle in QGIS but now it's very easy and, I think, quite effective.

I've used Layer Effects to add a drop shadow to the globe

The only thing I should add is that in order to get the sea showing in the background here - rather than having the land floating on a blank canvas, I took the 1 degree graticule and converted it to a polygon layer via Vector > Geometry Tools > Lines to Polygons...

One last example below, with lots of cities, just for fun.

Cities are from Natural Earth as well

That's all for now. I hope you find this useful.