A war has been waged in recent years against the use of rainbow colormaps in graphical representations of scientific data (Hawkins, 2015 and here and here). Several papers have been published arguing against the use of these colormaps, most recently in Nature Communications (Crameri et al., 2020), a paper I’ll be discussing more thoroughly below. The argument against rainbow colormaps rests chiefly on two themes: 1) rainbow colormaps misrepresent data because they are not perceptually uniform; 2) rainbow colormaps are useless for those who are color deficient. But why then are rainbow colormaps so popular?
While the #endtherainbow campaign offers valid points in support of its aim, it’s also tinged with a vociferousness and disdain that takes away from its legitimacy, namely in brooking no arguments in favor of the rainbow. Rainbow colormaps in some corners of the scientific community have become “the enemy” (Moreland, 2009) with proponents saying “there’s plenty of research that suggests the rainbow makes it harder for most of us to understand scientific data”. Yet the same, few papers are continually referenced in support of this view, with little empirical data showing just how terrible these maps actually are. Within this area of debate on what constitutes the right and the good of data visualization, nomenclature and standard are still being discussed (Bujack et al., 2017).
De gustibus et coloribus non est disputandum. Color perception is, to some degree, a form of qualia, a state of subjective experience that cannot be shared, even if there are approximate objective references linked to internal phenomena. In my view, qualia are more of a quandary in philosophy. The behavioral sciences work hard towards breaking the 4th wall that represents the qualia problem, be it through clever experiments, modeling or technological advances. Still, today, we’re not yet able to fully share our experience of color perception with one another, which underlines the truth of color perception: it is subjective. Therefore any colormap can present misleading data. Color perception and its neural correlates are still an active area of research. Color in the brain was long thought to be processed through area V4 (McKeefry & Zeki, 1997; Lafer-Sousa et al. 2016; Brewer et al., 2005) in the visual cortex but a recent paper has found color categories tied to attentional, not visual, areas (Bird et al., 2014). There’s evidence that color categories are set in infancy (Skelton et al., 2017) and are invariant to culture (Lindsey et al., 2015; Jonauskaite et al., 2020).
Color is highly relevant in both science and art. It’s one of these things that straddles both worlds, which makes colors so valuable (and so much fun).
From a scientific standpoint, color corresponds most readily to the visible electro-magnetic spectrum and what stimulates cones in our eyes (what it means in the Standard Model remains a question mark for me). In art class though, we’re shown the color wheel, where the longest and shortest wavelengths meet (at least one source of distress regarding colormaps, because the electromagnetic spectrum doesn’t fold on itself.) Then again, as any pink-loving kid drawing a rainbow in kindergarten might tell you, there’s a place in our perceptual world for the color wheel.
Crameri and colleagues (2020) use the term “unscientific” in reference to rainbow colormaps 16 times, without adequately defining exactly what constitutes scientificity. Presumably, thelack of scientificity relates to how rainbow colormaps impede a universal understanding of data, because rainbows are not perceptually uniform. Therefore some colors may be weighted more than others. Indeed, yellow is perceived as being an uncertain color in a study that explicitly examines what constitutes an “uncertain” color (Tak & Toet, 2014). The authors argue that the figure in science is of primordial importance in communication, notably in that it conveys information in an accessible manner, however, some measure of accuracy is always sacrificed in the graven image, as these are representations of true phenomena. Therefore, while pains are taken in data visualization to minimize artistic license, the latter remains. And within that interstice of individual license is the scientist-artist’s individual notion of what needs to be conveyed to the reader.
The key feature of the rainbow with regards to science communication is that it is pertinent and salient to an average viewer. Color salience maps are an active area of research (Etchebehere & Fedorovskaya, 2017; Frey et al., 2008; Wool et al. 2015) but they often use measures such as gaze duration to assume salience, or rather, relevance (Jost et al., 2005). The thing with visual attention though is that it is an imperfect proxy for what is subjectively meaningful. And when we present a figure in science, our aim is to convey meaning. What is irksome in anti-rainbow arguments is that theoretic assumptions regarding conveyed meaning in colormaps – the main argument of the papers – are poorly validated. That is, the evidence that rainbow colormaps’ misrepresent data is sparse, as far as I know. Specifically, I’ve found few studies that test human understanding and interpretation of colormaps (Reda et al., 2020); but some who have yielded equivocal results (Dasgupta et al., 2018). An oft-referenced study shows a decrease in accuracy when using a 3D rainbow color map for a cardiac measure relative to a 2D perceptually accurate map, leaving one to wonder which feature (color or dimension) affected accuracy (Borkin et al., 2011).
One example used by Crameri et al., 2020, is that of a Mars rover misinterpreting a rainbow colormap of martian terrain, leading to it falling off a cliff. This scenario seems implausible. If the Mars Rover is self-driving, then it is not counting on the colormap, but on the numbers it represents. The scenario is also unlikely in the case of a human. One would expect an interplanetary rover driver to rely on numbers, and not a representation of her brain’s representation of a colormap.
Another argument posed against rainbow colormaps is that they are meaningless to people with color deficiencies. However, there is another class of people for whom color is highly pertinent, and that is synesthetics. I speak as a grapheme-color synaesthete (Hubbard et al., 2005). I do not study synaesthesia, and cannot speak for synaesthetes as a whole. But color is a very important, very meaningful feature of the world I inhabit (Nikolic et al., 2007) . Letters, numbers, spoken, thought, heard and read all have color, always, immutable hues from ever since I can remember. The consequence is that my visual attention to, say, black and white movies or a monotonic colormap is poor (Laeng, 2009) . Simply put, I will just skip over the figure to the caption or the text of the graph instead. These monotonic maps, to me, lack information. And in the end, these colormaps are meant to convey information, so where does that leave people like me? How many of us are even out there? Here is a quote from an oft referenced paper (Borland & Taylor, 2007). “Increasing luminance from black to white is a strong perceptual cue that indicates values mapped to darker shades of gray are lower in value than values mapped to lighter shades of gray. This mapping is natural and intuitive.” Is it (Valuch, 2021)?
ColorText
To quote an #endtherainbow proponent: “At this point I apologise to any readers who suffer from migraine. An obvious drawback of this approach is how garish it is.” Yes, obviously. And yet my world is so beautiful. Are we ready to make like Romans and agree that de gustibus et coloribus non est disputandum?
Which leads me to the following, highly uncomfortable point. A curious statistic distressingly cited in the Crameri et al. (2020) is the high proportion of Northern European males who suffer from color deficiency versus sub-Saharan African men who do (ca 0%). It is unclear why this fact is in the paper. Is it a dog whistle? An argument to cater to Northern European males because there are more such scientists? By contrast, synaesthesia has long been thought to be more widespread among females (although see Simner & Carmichael, 2015). The relegation of rainbow colormaps to the unscientific bin almost seems an extension of one group imposing a notion of what is adult (Brothers & Gaines, 1972); refined; Euro-centric; reasonable; tasteful; sophisticated. A constellation of adjectives linked by a conformity and uniformity that the rainbow transgresses.
The confusion that can occur with rainbow colormaps is nonetheless real, especially in some figures that are particularly sloppy, for instance in those where the two extremes of a colorbar yield the same color. It behooves the person making the figure to be responsible with their choice of colormap but also to include a colorbar and set appropriate intervals according to the message being conveyed in the image. If there is an unambiguous unscientific practice it is in omitting color-corresponding numerical measures in a colorbar or on the figure itself. Therefore, there is certainly room for improvement in how colormaps are used.
Likewise, antagonists of rainbow colormaps have raised an important concern in highlighting the impact of certain schemes on those readers with color deficiencies. The examples shown of how the latter perceive certain colormaps are remarkable and should give us pause before we make our figures. It would be a worthy compromise therefore to include alternative figures in scientific papers that are color-deficient friendly. The argument presenting synaesthesia in contrast to color deficiency is not meant to equivocate the two conditions. People are said to suffer from the latter, but not the former. Therefore, more care should be taken to accommodate color deficiency but perhaps not in the absolute terms promulgated by the #endtherainbow crowd.
The push to end rainbow colormaps rests on valid points that nonetheless fail to consider the most obvious consideration inherent in selecting visual tools to present data, and that is how it is received by the viewer. It is clear that misleading representations should be minimized but the easiest way to do so is to fall back on their numerical source. Further, while many colormaps offer little visual salience to color deficient individuals, the same may be argued for at least some synaesthetes and perhaps some people. What is effective therefore should probably be tested empirically and at a very large scale. For instance, when meteorological events, such as hurricanes, are presented, it may be that the infamous rainbow spurs viewers to action more than a muted spectrum does, a prospect less garish than unnecessary deaths.
*The code to write in color can be found here.
** Google has an improved rainbow colormap called Turbo
Bird, C. M., Berens, S. C., Horner, A. J., & Franklin, A. (2014). Categorical encoding of color in the brain. Proceedings of the National Academy of Sciences, 111(12), 4590–4595. https://doi.org/10.1073/pnas.1315275111
Borland, D., & Taylor Ii, R. M. (2007). Rainbow Color Map (Still) Considered Harmful. IEEE Computer Graphics and Applications, 27(2), 14–17. https://doi.org/10.1109/MCG.2007.323435
Brewer, A. A., Liu, J., Wade, A. R., & Wandell, B. A. (2005). Visual field maps and stimulus selectivity in human ventral occipital cortex. Nature Neuroscience, 8(8), 1102–1109. https://doi.org/10.1038/nn1507
Brothers, R., & Gaines, R. (1973). Perceptual Differences Between Hippies and College Students. The Journal of Social Psychology, 91(2), 325–335. https://doi.org/10.1080/00224545.1973.9923054
Bujack, R., Turton, T. L., Samsel, F., Ware, C., Rogers, D. H., & Ahrens, J. (2018). The Good, the Bad, and the Ugly: A Theoretical Framework for the Assessment of Continuous Colormaps. IEEE Transactions on Visualization and Computer Graphics, 24(1), 923–933. https://doi.org/10.1109/TVCG.2017.2743978
Clark, A. (1985). Qualia and the psychophysiological explanation of color perception. Synthese, 65(3), 377–405. https://doi.org/10.1007/BF00869276
Dasgupta, A., Poco, J., Rogowitz, B., Han, K., Bertini, E., & Silva, C. T. (2020). The Effect of Color Scales on Climate Scientists’ Objective and Subjective Performance in Spatial Data Analysis Tasks. IEEE Transactions on Visualization and Computer Graphics, 26(3), 1577–1591. https://doi.org/10.1109/TVCG.2018.2876539
Etchebehere, S., & Fedorovskaya, E. (2017). On the Role of Color in Visual Saliency. Electronic Imaging, 2017(14), 58–63. https://doi.org/10.2352/ISSN.2470-1173.2017.14.HVEI-119
Frey, H.-P., Honey, C., & König, P. (2008). What’s color got to do with it? The influence of color on visual attention in different categories. Journal of Vision, 8(14), 6–6. https://doi.org/10.1167/8.14.6
Hawkins, E. (2015). Scrap rainbow colour scales. Nature, 519(7543), 291–291. https://doi.org/10.1038/519291d
Hubbard, E. M., Arman, A. C., Ramachandran, V. S., & Boynton, G. M. (2005). Individual Differences among Grapheme-Color Synesthetes: Brain-Behavior Correlations. Neuron, 45(6), 975–985. https://doi.org/10.1016/j.neuron.2005.02.008
Jonauskaite, D., Abu-Akel, A., Dael, N., Oberfeld, D., Abdel-Khalek, A. M., Al-Rasheed, A. S., Antonietti, J.-P., Bogushevskaya, V., Chamseddine, A., Chkonia, E., Corona, V., Fonseca-Pedrero, E., Griber, Y. A., Grimshaw, G., Hasan, A. A., Havelka, J., Hirnstein, M., Karlsson, B. S. A., Laurent, E., … Mohr, C. (2020). Universal Patterns in Color-Emotion Associations Are Further Shaped by Linguistic and Geographic Proximity. Psychological Science, 31(10), 1245–1260. https://doi.org/10.1177/0956797620948810
Jost, T., Ouerhani, N., Wartburg, R. von, Müri, R., & Hügli, H. (2005a). Assessing the contribution of color in visual attention. Computer Vision and Image Understanding, 100(1), 107–123. https://doi.org/10.1016/j.cviu.2004.10.009
Lafer-Sousa, R., Conway, B. R., & Kanwisher, N. G. (2016). Color-Biased Regions of the Ventral Visual Pathway Lie between Face- and Place-Selective Regions in Humans, as in Macaques. Journal of Neuroscience, 36(5), 1682–1697. https://doi.org/10.1523/JNEUROSCI.3164-15.2016
Lindsey, D. T., Brown, A. M., Brainard, D. H., & Apicella, C. L. (2015). Hunter-Gatherer Color Naming Provides New Insight into the Evolution of Color Terms. Current Biology, 25(18), 2441–2446. https://doi.org/10.1016/j.cub.2015.08.006
Moreland, K. (n.d.). Diverging Color Maps for Scientific Visualization (Expanded). 20.
Nikolić, D., Lichti, P., & Singer, W. (2007). Color Opponency in Synaesthetic Experiences. Psychological Science, 18(6), 481–486. https://doi.org/10.1111/j.1467-9280.2007.01925.x
position and topography of the human colour centre as revealed by functional magnetic resonance imaging. | Brain | Oxford Academic. (n.d.). Retrieved May 5, 2021, from https://academic.oup.com/brain/article/120/12/2229/390973?login=true
Reda, K., & Szafir, D. A. (2021). Rainbows Revisited: Modeling Effective Colormap Design for Graphical Inference. IEEE Transactions on Visualization and Computer Graphics, 27(2), 1032–1042. https://doi.org/10.1109/TVCG.2020.3030439
Simner, J., & Carmichael, D. A. (2015). Is synaesthesia a dominantly female trait? Cognitive Neuroscience, 6(2–3), 68–76. https://doi.org/10.1080/17588928.2015.1019441
Skelton, A. E., Catchpole, G., Abbott, J. T., Bosten, J. M., & Franklin, A. (2017). Biological origins of color categorization. Proceedings of the National Academy of Sciences, 114(21), 5545–5550. https://doi.org/10.1073/pnas.1612881114
Tak, S., & Toet, A. (2014). Color and Uncertainty: It is not always Black and White. EuroVis – Short Papers, 5 pages. https://doi.org/10.2312/EUROVISSHORT.20141157
Valuch, C. (2021). Color Can Shorten Breakthrough Times in Continuous Flash Suppression through Increased Salience and Task Relevance. Vision, 5(1), 13. https://doi.org/10.3390/vision5010013
Wool, L. E., Komban, S. J., Kremkow, J., Jansen, M., Li, X., Alonso, J.-M., & Zaidi, Q. (2015). Salience of unique hues and implications for color theory. Journal of Vision, 15(2), 10–10. https://doi.org/10.1167/15.2.10