Colorbar Confusion

A press release from Purdue University describes the effect of greenhouse gas emissions on “heat stress,” using the diagram above to illustrate the difference in effect between accelerated emissions (top) and decelerated emissions (bottom). A description from the web page:

“This image illustrates heat stress in the 21st century for two greenhouse gas emissions scenarios. The top panel shows the expected intensification of the severity of extreme hot days given accelerating increases in greenhouse gas concentrations. The bottom panel shows the expected decrease in intensification associated with decelerated increases in greenhouse gas concentrations.”

(I apologize for the nearly illegible size… The Purdue website offers up the diagram in the teeny-tiny size above, or print quality, which I assumed would be excessive.)

There are a couple of things I find odd (and counterintuitive and frankly counterproductive) about the diagram…

Firstly, the color spectrum used in this false-color representation of the data feels wrong to me, since it ranges from cool blues through warm oranges and reds and thence to… The beginnings of a cool violet? Particularly since we’re talking about temperature (well, sort of) here, and most people have grown accustomed to weather maps colored by temperature. Stopping at red gives you plenty of color resolution. (And maybe next time, you can choose something other than the garish rainbow colors?)

A more egregious error permeates the diagram, however. Perhaps we can simply call this the “apples and oranges&rduo; issue: two images, side-by-side, offered up for comparison, need to share enough to allow for easy comparison. I last blogged about this in relation to an NCAR visualization of Hurricane Katrina, but the idea is simple enough: don’t ask the viewer to do unnecessary work in interpreting your imagery, because unnecessary work leads to unnecessary risk of miscommunication. In the case of the two images above, the color bars are flipped for no apparent reason, so increasing values get warmer (in hue) on the top and cooler (in hue) on the bottom. Why? Also, the scale of the two color bars changes, running from 3 to 8 on top and from –3 to 0 on the bottom. Why? Why? Why?

(Well, okay, I can acknowledge one drawback in this particular case. Since the two datasets do not overlap, coming up with a single colorbar would be a little tricky; indeed, you’d almost need to insert an intermediate model showing, say, no change in greenhouse admissions, which would presumably result in values in between. But the issue of inverting the colorbar still stands: “red on top bad, red on bottom goooood” simply leads to confusion.)

I find behavior of this sort annoying when watching a scientist presenting data in a talk, but as part of a press release, it just saddens me. My fear is that the folks in the university press offices don’t even try to fix these problems… Perhaps because they don’t care, but perhaps because they don’t even think the data should be easily understood.

Hmmm. Maybe it’s time for a Tufte-like “Graphics 101” for science types? I looked for such a thing just now, but I didn’t find anything. Anyone reading know of such a thing?


I know I already posted one hurricane-related entry this week, but I have another. This one is related to a press release from the National Center for Atmospheric Research (NCAR) about a new technique “that provides a detailed 3-D view of an approaching hurricane every six minutes.” I was curious what a 3-D view of an approaching hurricane might look like, so I followed the links, and…

I got the above. Hmmm.

To be fair, the page makes no claim for the above to be any kind of 3-D view, but it does supposedly offer a “side-by-side” comparison of radar data (on the left) and “NCAR’s ARW experimental forecast” (on the right). An animation shows the evolution of the hurricane, and as the caption duly notes, “The radar vantage point is stationary, on the Gulf Coast, while the ARW viewpoint follows the hurricane itself.” And therein lies my cavil (I’m trying to find synonyms for “gripe”).

The presentation of the images should facilitate side-by-side comparison; instead, the camparison seems hampered by the graphical choices. The change in background color strikes me as mildly annoying, but the field of view of the two images is also slightly different, and the manner in which the left-hand image obscures the state lines makes comparison even more difficult. It’s rather hard to tell how well the model replicates the observed behavior of the hurricane.

The animation only exacerbates the problems because the simulation follows the eye of the storm whereas the Doppler radar remains stationary (as noted in the caption). C’mon, folks, this is data! You can plot it however you want! Why not present it in a way that allows us to get a real feel for how well the computer model matches reality?

How to do it right, in brief: make the background of the two (observed data and computed data) as similar as possible, in terms of scale and markings (e.g., state and county lines), then plot the same quantities using the same color bar (which, as far as I can tell, is what they did in the above example). Would that be so hard?