Today I wanted to share with you a little project of mine from a few months ago, which may best be described by the question: What happens if you take the shoreline of a lake, cut it, and unfurl it?
The once-closed shoreline of the lake now becomes linear, providing a new perspective on a familiar feature. Warm up your scrolling finger, because here’s what happened when I linearized Lake Michigan:
Click for an even larger version. Take a while to browse around.
A drive around the lake becomes a reasonably straight line. Not only that, but the map is actually continuous — the roads running off the bottom of the map are the same as those coming in at the top. It provides a unique perspective on the way people arrange themselves around the lake.
I’ll get to the how of all this in a little bit, but first the big question: after seeing my map, most people ask me, “Why did you do this?” Actually, “What’s the point?” is usually how they put it.
Why Make the Map
Like many of my projects, this can be filed under “stuff I spent a huge amount of time making, and now don’t know what to do with.” Partially I did it to see if I could; I’m pleased with the technical achievement of having figured out how to construct the map. But the end result is not just an idle novelty, like the Penrose binning was.
I made this map because I wanted to show space referenced against a natural feature, rather than figuring locations based on the cardinal directions of north/south/etc. I think it’s a very human perspective, grounded in how we relate to the lake, rather than how it looks from space. Rob Roth just wandered by while I was writing this and said that this depicted “configural knowledge,” so there’s your search term if you want to read the academic side of this sort of thing.
As the idea took shape in my mind, it reminded me of a couple of other things that I’d heard about in the past. The first is the 1849 Petition of the Ojibwe Chiefs, sometimes attributed to Kechewaishke.
You can follow the link to read more about the interpretation, but the short version is that the map depicts the relationships of various natural features in northern Wisconsin. There is spatial information here, but it’s not presented against the grid to which we are accustomed. The vast expanse of Lake Superior is compressed into the thick horizontal blue line; its true size and shape is not relevant here. Instead, the map depicts understandings of relationships, rather than physical measurements.
While working on my map, I was also reminded of this Slate article on geocentric directional systems (thanks Alasdair Rae for reminding me where to find the story!). On Bali, you might find directions described not in terms of north/south/etc., but instead as to whether one is moving clockwise or counterclockwise around the island, or moving toward or away from the nearest major mountain. It’s orientation with reference to surroundings, rather than a superimposed grid.
I’m not sure if either of these were direct inspirations, but they were certainly in the back of my mind as I got to work, as examples of other uncommon ways of thinking about and depicting space.
Besides an opportunity to play around with a fun perspective, this is also a sentimental and personal project. I’ve lived almost the whole of my life on this map, in one part or another. Lake Michigan has been a dominant feature in my personal geography, and it anchors my understanding of my homeland. If I’d been raised in another context, out in the plains or along the ocean coasts, I don’t think I would ever have thought to do something like this. But, whether or not my upbringing figured in, this map makes perfect sense to me; it feels right. It also emphasizes just how far apart my current home (Madison, WI) seems from my former home (Kalamazoo, MI) when the only path between them involves driving around a vast expanse of water.
I’m certainly not the first to do something like this. Shortly after this post went up, I learned that Nick Martinelli sketched out a hand-drawn example of this linearizing idea earlier this year, in which he straightened out Oregon. I imagine there are other cool examples out there that I don’t know about.
Nick Martinelli’s linear Oregon sketch. Hand-drawn maps are just the best.
There are plenty of other linear maps out there that people have brought to my attention, though the comparison might sometimes be inexact. Works like Ogilby’s maps of British roads start with features that are already linear, rather than straightening out a closed shape.
Before I get into how the map was made, I have some random observations/thoughts on the design.
- The whole concept behind the map is weird, and the end product looks odd. Because of that, I wanted to, as far as possible, “naturalize” this distorted perspective — to make it feel like it’s perfectly normal to see the world this way. The warm colors are meant to feel more organic, and also to feel like something you’d see in any run-of-the-mill map. It’s not drawing attention to itself or shouting “hey look I made a weird distorted perspective!” Attractive, but understated.
- The brownish tones are a population density raster. I included it partly as a matter of visual interest, so things wouldn’t be so flat. I considered putting city boundaries instead, but showing the noisy distribution of humans in this way, rather than with hard political boundaries, has a more natural feeling to it. Again, trying to make the map feel more organic.
- I didn’t include a legend for the density. I feel like readers can figure out quickly that “browner = more people,” especially with the city labels being on there. I didn’t like the distraction of adding a legend, especially since the actual density numbers are irrelevant. I have a (possibly annoying) fondness for not really putting much explanation on my maps.
- I decided to go with Gill Sans for the city and highway labels, which I’ve not used much before. It feels a little older (because it is), and it’s a classic. Again, I think it fits the idea of an attractive, understated aesthetic. Using a typeface like this sells the idea that this is a normal and natural way to see the world.
- I used Sorts Mill Goudy for the lakes, because a website told me to. Here’s a real typography geek secret: we often look online to see what other people think, and then if their opinion makes sense we might adopt it. I did a search for “what serif to pair with Gill Sans,” and went with a suggestion that made sense to me.
- The map is formatted to print at the fabulously inconvenient size of 10″ × 60″
- One of the most difficult decisions was how to rotate the map. I actually started with a horizontal orientation, with the land at the bottom and the water at the top. To me, that seemed to embody the feeling of sitting on land and looking out toward the water. However, then Evan Applegate pointed out to me that if I ever put it on a website, people would dislike having to scroll horizontally. So, I rotated it 90º and relabeled it. This is me surrendering to the tyranny of modern digital devices.
- Where to center it was another concern. The map above starts with Chicago at the top, a familiar location. In print, I’d probably put Chicago near the center or just above. If someone’s looking over the whole object at once in person, I don’t think they’ll start at the top the way they’re forced to when they see the digital version in this post. And I want Chicago to be that familiar anchor that people see early on to begin to understand what’s happening.
- All this boils down to: digital displays really limit your ability to appreciate this map, and it would be great if we could all go back to printed maps.
How I Made the Map
So, let’s get into how I made this thing. I began by compiling data in ArcMap, pretty much as though I were making an ordinary map: roads, population density, state boundaries, and hydrography.
US National Map Small Scale, and the Atlas of Canada 1 million National Frameworks. I selected just the roads that were Interstates, US Routes, or those that were part of the Trans-Canada Highway or Canadian National Highway System.
Same sources. There are a lot of rivers in these data sets, and I so I thinned the network heavily. I actually took just the rivers that were in the Natural Earth 10 million scale data. The Natural Earth linework itself was too coarse, but it was a guide as to which of the more detailed 1 million vectors to use from my other sources.
As above, plus two more detailed sets — TIGER and Canadian Census Boundary Files. The more detailed sets had three purposes:
- I wanted more detail on the Lake Michigan coastline, since a lot of attention would be drawn there.
- The more generalized Lake Michigan polygon didn’t always line up well with the population data, so using the more detailed coastline helped fix gaps. As I went along, I manually generalized the linework here and there where it was too detailed to look good.
- The coarser, US 1 million scale data were fine for most smaller lakes, but sometimes the lakes were weirdly in the wrong place, so I occasionally had to use the TIGER data to manually correct the position of the more generalized data. I also eliminated all lakes under 0.5 sqmi, just to keep things clean and simple.
US National Map Small Scale.
For the US, I used Census data. Rather than join the Census data to shapefiles myself, I found some handy ones that already had selected demographics already in the attribute table. For Canada I grabbed some data from Statistics Canada.
As said, I wanted the population density layer to add some visual interest. I wanted it to be fairly finely textured, so I initially used Census blocks, the smallest geographic unit available. But there were way too many of them, and the files made Illustrator unhappy. But if I stepped up to larger units, like block groups or tracts, there was too little detail in rural areas. So, I used a mix: blocks in rural areas, and tracts in urban areas. I used the Census shapefiles for urbanized areas to make the distinction. Doing this drastically reduced my polygon count while keeping all the detail I wanted in rural areas. In urban areas, the downgrade in resolution was unnoticeable at my map scale.
I used my mixed census shapefile to make a sort-of-unclassed choropleth of population density. In reality, it’s got 50 classes, so it’s close enough to unclassed that you can’t tell (since I’m not aware of how to do a properly unclassed one in ArcMap). However, the classing is not linear. Instead, it shows the square root of the population density. I’m a big fan of giving unclassed maps a non-linear treatment (and drafted an unfinished blog post about it years ago). In this case, using the square root reveals small changes in low-density areas, details that would otherwise be washed out. In a linear choropleth, there would be major cities visible and not much else; the density of Chicago or Milwaukee would skew the color scheme so much that small villages and hamlets in the country would be unseen. Taking the square root of the data de-emphasizes the densest areas, helping fix this. Actually, I’ll tell you a secret: I showed the 2.25th root of density, not the square root, because I’m way too detail obsessive. That extra 0.25th of a root made a tiny difference that I liked and which you will never notice.
Notice all the towns that appear once freed from the tyranny of linear color schemes.
Finally, before applying the classing, I clipped the top 1% of the data. So, the very densest places all look the same, and the colors only change when you’re outside of that, in someplace that isn’t in the top 1%. This was, once again, to keep the very densest outliers from skewing everything.
Now comes the fun part: linearizing the map. First, I drew a simple polygon around the lakeshore in Illustrator. Then I bisected each angle in that polygon to create skewed quadrilaterals going both inland and into the lake.
I drew the lines such that each one going inland was the same length. This created a sort of rough buffer around the lake, so that my final map would end up including all places within a certain distance of the lakeshore. For the lines going into the lake, I made them only long enough to cover peninsulae, nearby islands, and any other bits of land that fell on the other side of my original simple polygon.
To actually make the linear map, I needed to take each of these skewed quadrilaterals, un-skew them into rectangles, and then stack them up. But, how does one straighten a quadrilateral in this way?
This is where I drew upon the expertise of Christopher Alfeld, a mathematician/programmer friend, because the answer is: lots of math. I once briefly understood why we were doing what we were doing, but that knowledge has since vanished into the ether. What I am left with are the equations which he determined that I needed to use:
x = x0 t s + x1 (1 − t)s + x2 t(1 − s) + x3 (1 − t)(1 − s)
y = y0 t s + y1 (1 − t)s + y2 t(1 − s) + y3 (1 − t)(1 − s)
Here, s and t are the new coordinates of our transformed quadrilateral. After asking Wolfram Alpha to solve for those variables, we get:
Fortunately, Wolfram Alpha outputs equations as images, so I didn’t have to re-type all this. Unfortunately, Wolfram Alpha outputs equations as images, so I did have to re-type all of this when I actually wrote the code to make the map.
To implement this, I wrote a Python script. It loads in an SVG, applies the transformation, and kicks out a new SVG. I didn’t find an existing Python SVG parser that I could immediately figure out, so the biggest part of my script was writing a simple one.
Obligatory code screenshot. I should share the real thing, but I think it needs some tweaks first.
Once all that was settled, I could make the map. All I needed to do was:
- Go to the untransformed map in Illustrator and select one of the skewed quadrilaterals;
- eliminate all the artwork outside the quad;
- save as an SVG;
- run the script;
- load the newly generated SVG into Illustrator;
- stack the unskewed quads together in a new Illustrator document; and
- style everything.
There were a few wrinkles along the way. I had to increase the point density of the linework (e.g., a straight line between two points becomes a straight line between 4 points), so that the shapes could warp a little more smoothly. Also the equation for whatever reason didn’t work right until I first rotated the skewed quad such that one line was horizontal.
Finally there was the labeling of cities. I decided not to do every one of them, but only the significant ones. The problem was how to decide what is “significant.”
Labeling every city above a certain population threshold wouldn’t work, as it would miss locally-important small towns while giving attention to every single suburb of Milwaukee or Chicago. Instead, I needed a way to measure regional significance.
I ended up with a process fairly similar to the one I developed for locally-enhanced hypsometric tints. Short version: I used my population density shapefiles to make a raster. For each pixel in the raster, I checked to see how many standard deviations above or below the regional mean its density was (the region being a 25km radius circle). Now I knew whether each pixel’s density was regionally significant, or whether it was merely about as dense as its surroundings. Then, for each city, I summed its pixels to get a score for how regionally relevant it was. Then I labeled the highest-scoring cities.
It worked out pretty nicely. A Chicago suburb of 10,000 people doesn’t make the cut, but a small town of 1,000 in the middle of nowhere gets labeled, as it’s a landmark for the area. It’s very much like the scale ranks that occur in Natural Earth’s populated place shapefile, but I have no idea how they did theirs.
As said, I’m not sure what to do with this right now, other than stick it here. A couple of people have asked about getting prints, and so I’ve also put it up on Zazzle, where you can find a vertical or a horizontal version.
I’ll be presenting on this work, and showing off a print version, at NACIS2015. I’d like to do all the Great Lakes, and put them onto one impractically sized poster that maybe someone would want. Such maps might also make nice wallpaper borders, I suppose, if I went back to a horizontal orientation. Mostly, though, I just want to get them out there and in front of people’s eyes, and provide them with a new perspective on a familiar feature.
If you made all the way to the end of this post, I have a bonus for you: I also did Lake Superior!
Click for an even larger version. Take a while to browse around.