Inherent in the idea of remixing or remaking someone's existing, original design--in any milieu--is the conceit that you believe that you have something to add to the discussion. Maybe you can improve the visual appeal; maybe you can clarify the message; maybe you can add another layer of meaning somehow; or maybe you can shepherd the original vision across communities to a new audience.
What to do, then, when faced with a remix challenge like "Global Temperature Change," 2017's 51st Makeover Monday project, based on NOAA weather data collected by Ken Black (@3danim8), hosted by EXASOL, and brilliantly visualized by Antti Lipponen (@anttilip)?
(You can download the video from Lipponen's Flickr page.)
This already takes tons of data and excellently displays how the monthly temperatures within each continent and country vary from month to month and year to year, sometimes higher than "normal" and sometimes lower...until the last 30 years, when they are consistently higher.
In my opinion, this visualization is already close to ideal for telling the story. If you care about a specific country, you can follow that; if you want to track the exact temperature variation, you can follow that; if you just want the data from a specific year, you can get that as well. It does a good job of providing detail while also giving an overview. Moreover, it doesn't take a lot of explanation for viewers to get the point of what the viz is saying.
Given all of this, I had no real insight into how I could add onto this viz meaningfully. I didn't want to recreate it, remix it, or echo it in any way. But, climate change being arguably the most universal and immutable threat to our continued existence on earth, I was not going to refuse the call; of course I would use this massive data set to produce my own related design.
So what direction should I take?
The existing visualization is quite explicit and specific. It's almost pure data. These are temperature anomalies. These are the countries. This is the temperature difference from the baseline in this specific year. The background is neutral. The colors encode temperature. It would not be out of place in a major media website.
In order to produce a different take on the data, I decided to go in the other direction: as close to pure art as possible, using the data as the basis for a visual aesthetic and effect that took a position--that is, is not in any sense neutral.
In my original concept, I wanted the design to include as few words as possible--even no words. Or, possibly, to have words that were not obviously legible, but that supported the overall feel and vibe of the design.
I thought that I could use color, brightness, and position of marks--based on some qualities of the data--to drive the viewer's emotional reaction to the global temperature changes over time.
I didn't think I would even have to be particularly explicit about what any individual mark meant, since this wasn't meant to be an analytic or exploratory product. My goal was to create something that immediately gets a viewer's attention; then conveys a dangerous trend over time; and ultimately reinforces the seriousness of the climate change issue, and the intensity of the peril in which we all find ourselves.
I did make some compromises along the way: there are more words than I originally thought I'd have; I used a smaller (monthly-aggregated) subset of the data than I wanted to; I tested but discarded some interactivity (filters/highlighters by country); and my original intentions to either use animation or a grid/matrix layout for marks were both rendered sub-optimal by my approach to the data. Additionally, my reliance on a few table calculations DESTROYED my development speed ... maybe an unintended benefit, since it made me more thoughtful and intentional with my design decisions along the way. (Fortunately, the final version that is out in the world contains a table-calc-free data extract that loads and renders quickly.)
A general explanation of the viz is:
There were about 13,000 different weather stations taking temperature readings around the world at some point in time between 1855 and 2017. They weren't all in continuous operation; most of them started up in the last few decades; and more than half of them are in the United States. However, aggregated to the monthly level, we can see what the average high temperatures and low temperatures were at each station.
I decided to identify the specific month and year that each station had its highest average high temp, and its lowest average low temp. My suspicion (borne out by data) was that lots more stations were going to have their record high months in recent years, and that not many were going to have their record low months recently. I did not care about actual temperatures too much, or change over time, or latitude of station; just, when did you hit the high point and when did you hit the low point?
There's a built-in alarm to the notion that SO MANY stations are recording record high months, and that many of those stations recorded their record low months decades, if not a full century, ago.
Visually, by plotting the marks on a left-to-right timeline and a bottom-to-top temperature range--showing only the months of record highs and lows--the viewer's eye is drawn forward and up. Lots of the marks are crammed into the top right of our scatterplot. Lots of the lines connecting low-month-to-high-month also run from bottom left to top right. It's a visual sweep upwards and forward, into a corner; a corner we're painting ourselves into.
The Boring Technical Section
I'll mention a bunch of tech things quickly here but it isn't meant to be a how-to article. Most of the complex calculation stuff isn't in the final data extract anyway. If you don't really care about the specific technical stuff that went into creating this viz, feel free to scroll down until you see a picture of my cat giving you side-eye.
I used table calculations to find the month and year for each station where the maximum temperature was the same as the window-max temperature, or where the minimum temperature was the same as the window-min temperature. I combined these into one field called "Maximum or Minimum," which became my Y axis. For any month/year that wasn't an extreme month, "Maximum or Minimum" carried a Null value. (I also created a helper field for later, called "Show It?," which was a value of -1 if that month/year was the minimum month, +1 if it was the max month, and 0 if it wasn't a special month.
Now, there are plenty of times when the max or the min was matched in different month/years. I didn't correct for any of this (like, I didn't take ONLY the earliest month/year). I wasn't too concerned by that because, you know, record highs are record highs, whether they are ties or outright records.
There are also some calculations that make sure that I'm only pulling stations that recorded something in March 2017 (I wanted active stations), and a few calculations that make the tooltips read smoothly (you know, switches that make sure that the tooltip says "highest" or "lowest" appropriately; switches that add the state as well as the country to the tooltip if there's non-null State data).
Finally, I calculated how many months of readings each station was active for, and sized their marks correspondingly in the display. (This is also shown in text in the tooltip.) I figured that some of the recent record low readings would come from stations that only came on line in the last few years, so I wanted to make sure they were smaller to accommodate for this.
I created my own ordered sequential color scale called "Meteorology," based on common color ramps used on weather maps on television and in full-color newspapers. I added it to my Preferences.tps file. (Go here for instructions if you don't know how to do this.) It's here if you want it:
<color-palette custom='true' name='Meteorology' type='ordered-sequential'> <color>#bc88b1</color> <color>#c147df</color> <color>#774be7</color> <color>#4480e8</color> <color>#2ee7df</color> <color>#32ed8d</color> <color>#83f043</color> <color>#e6ed46</color> <color>#eb9828</color> <color>#e55d15</color> <color>#f35237</color> <color>#db4438</color> </color-palette>
Text and Images
Since Tableau Public has such a small range of supported fonts, I tend to use the standard issue Tableau family, particularly in tooltips and axes. However, this was intended to be an art project, and should be able to stand alone outside of Tableau (for instance, if the screenshot is shared on other platforms, it shouldn't necessarily scream "THIS IS A TABLEAU DASHBOARD" at first glance).
For this reason I wanted to use different typefaces. And to include different typefaces, you need to use a third-party design application, be it Photoshop or PowerPoint or GIMP or whatever you prefer, so that you can create images that are pulled into Tableau.
However, I had a secondary requirement. I didn't want my text, minimal as I expected it to be, to be the dominant element of the viz, and I didn't want it to interfere with any interactivity. Since we do not have the ability to make worksheets transparent, any image-based text that we include must be placed on top of our charts; and we don't have the ability to clip the borders of our images, so they are always going to be big old rectangles that cover up elements on the worksheets. (I don't necessarily mean visually; obviously we're going to use transparent PNGs. But even transparent PNGs interfere with a viewer's ability to hover over or click on marks.)
To get around this, I used the old standby: background images. In Photoshop, I created one big image that included all the text elements I wanted, and then set that as the background image to my "map;" that is, the scatterplot that is the main worksheet (actually, the only worksheet) of my design. Here are the layers that went into creating the final background image.
And the final background image:
As you can see, there is a lot more text on there than I originally planned. I included the time axis labels as part of the image, because (again) I wanted the typeface consistency and the color gradient that the photo overlay (a Pixabay photo, BTW) provided. And as much as I wanted to just leave the large text, which was my original plan, I felt obliged to include the additional text boxes, for the sake of explaining some of the design decisions. I'm still not convinced the design doesn't suffer for having them, but at some point you have to just call the project completed.
Revising the Reviz
The original "completed" design looked like this:
I liked it and was mostly satisfied with it, but two things bothered me:
I thought the reference lines for 100 degrees and 32 degrees cluttered the design too much. They didn't serve enough of a purpose to be worth keeping. If I were going for precision and accuracy in mark placement, they would be worth keeping because they'd be good visual anchors for an analyst. But this is an art project. The color ramp tells you, more or less, how cold or hot the extreme temperatures are. The tooltips can show you exactly. So, I chose to remove them.
The bigger problem: I could tell from looking at this chart that more record-high months happened recently than record-low months. But, I soon realized, the story is more that the record-low months FOR EACH STATION probably happened longer ago than the record-high months...at least, for the significant majority of stations. From this chart, there's no way to see that. I wanted to link each station's extremes together, and hopefully those lines would draw the eye up and to the corner, emphasizing the continuing rise in record high temperatures.
Fixing these two things (and giving credit to the actual original viz designer, Antti Lipponen, instead of Ken Black, who gathered the data), and swapping out the giant and slow calculation-heavy data extract for a lighter, flatter extract-of-an-extract, finally got me to a point where I considered the project finished.
Art vs. Data Visualization: Truth, Validity, Value
I started out this post talking about, in not so many words, the hubris required to reviz, makeover, or remix a data visualization that is already outstanding on its merits. This is why I did not set out to do so.
What I did do, instead, was change my frame of reference.
Antti Lipponen used temperature anomaly information to create a magnificent data visualization. It will appeal to the rational mind. It is rooted in objective facts and rewards the viewer with a powerful, if expected, payoff at the end of a 35 second video.
But what if people don't care to hear this message? What if they see the headline "Temperature Anomalies" and a big "play" button, and say, "Ugh, not THIS nonsense again, global warming is a myth, temperature varies all the time." Then 80% of them go on to the rest of the Internet; 20% of them accidentally press play, and the first 20 seconds of the video SUPPORTS THEIR ARGUMENT. "See?" say the half-informed, "it's fluctuating all over the place. Just like I said it does." Then they jump in their Hummers and drive away.
I did not try to convince people rationally. I did not ask them to click a button. I did not write a clear, easy-to-read headline on a neutral background. I took the artist's approach. I used the medium at hand--data--to create a mood and an aesthetic that provokes an immediate reaction--even before you've read a WORD on the page, you see the dominant theme. BLACK. Serious. Intense. You see a color ramp: familiar; visually pleasing; dramatic. You see a flow of marks: rising; converging; gaining in brightness and intensity.
Then, or at the same time, you grab onto a few words, set out from the rest but still subtler than the marks: THESE ARE THEIR MONTHLY EXTREMES. Whose extremes? What is extreme? I know from the design something serious is going on, something intense; now I know it's extreme. What am I looking at?
Now I've drawn in an audience that might fully disagree with what I'm talking about. And they might not stay with the design for very long once they realize what the topic is. But they've spent more time with my design--an oblique, aesthetically-oriented presentation that aims to elicit an emotional response bang out of the gate--than they have with the rational, accurate, irrefutable presentation that Antti Lipponen so expertly created.
Is Antti's design more truthful than mine? They carry the same truth: extreme high temperatures are more common now. Climate change is happening. His is more accurate and more precise, for sure, but they both support the alarming truth.
Is Antti's design more valid than mine? The data is honestly represented, in slightly different ways, in both of our works. An analyst could accomplish more with his than with mine. It's more properly a "data visualization," whatever your definition of that term may be. But to use data as the basis for art...is that an invalid act? Does it diminish the work as a whole if you make visual appeal, aesthetics, and emotional stimulation your primary missions, rather than creating...ahem..."a visual display of the most important information needed to achieve one or more objectives; consolidated and arranged on a single screen so the information can be monitored at a glance"?
Is Antti's design more valuable than mine? To this, I have a specific and confident answer: no. It serves a different purpose, to be sure. But art, in the hands of those of us who are data-inclined, is an extremely powerful and valuable tool. It lets us get our message in through the side window when the front door is locked tight. It motivates and provokes; it challenges and illuminates. It transmits serious ideas--even fact-based ideas--in the guise of entertainment.
This week I had a point to prove: that data visualization can be art. That art can be based upon data. That to think of yourself as a "data visualizer" or a "dashboard designer" is to rob yourself of the power you have as an artist.
You are, all of you are, artists.
Your medium is data, your role is truth-teller, and your artistry is what gives you the power not only to convince the rational, willing audience of the objective truth, but also the unconvinced, the opposed, the reluctant, and the ignorant.
The dealings with artists, for instance, require great prudence; they are acquainted with all classes of society, and for that very reason dangerous. - Leopold I