Is it time scientists abandoned rainbow?

Posted on

If you are a climate scientist or marine scientists or almost any type of scientist, then you are almost certainly familiar with the famous colour scheme called rainbow. It is one I have used in a recent paper. I reproduce a figure below.

It is also a colour scheme I use throughout my recently finished thesis. (Let’s just say you need more than two hands to count the number of times it appears. Though, thankfully I have now decided to stop using projected maps, which quite simply don’t work in R)

So, I am probably inclined to be resistant to calls this week in Nature to stop using it. Hawkins et al. have a rather provocative piece of correspondence in this week’s Nature, called “Graphics: Scrap rainbow colour scales“.

It’s a short piece, but here’s an excerpt:

It is time to clamp down on the use of misleading rainbow colour scales that are increasingly pervading the literature and the media (see Accurate graphics are key to clear communication of scientific results to other researchers and the public — an issue that is becoming ever more important.

Aside from the challenge they pose for colour-blind readers (S. C. Allred et al. Nature 510, 340; 2014), spectral-type colour palettes can introduce false perceptual thresholds in the data (or hide genuine ones); they may also mask fine detail in the data (D. Borland and R. M. Taylor Comput. Graph. Appl. 27, 1417; 2007). These palettes have no unique perceptual ordering, so they can de-emphasize data extremes by placing the most prominent colour near the middle of the scale.

I am inclined to half agree and half disagree. Let’s begin with what I think is the key problem with their argument. The use of rainbow has now become a convention in many areas of science. In some fields, such as what I am currently working on, it is more or less universal. And I would argue that anything that departs from rainbow is actually more difficult for researchers to understand than rainbow itself.

The other problem with the arguments of Hawkins et al. is that they are essentially calling for scientists to spend a huge amount of time inventing ad hoc colour scales. For example, read their first piece “An open letter to the climate science community.” Under the heading “What are the possible solutions?” they provide some helpful guidance, but nothing particularly universal. And that’s what is needed a relatively universal colour scheme, or 3 or 4 colour schemes that will become conventions within science.

Until then, I am skeptical that discarding rainbow is a good idea. Why replace a slightly flawed graphical convention with utter chaos? After all, consistency is at the route of good style.


6 thoughts on “Is it time scientists abandoned rainbow?

    ob said:
    March 19, 2015 at 11:50 am

    Was there a graphical convention on contour lines? Or on greyscales? I don’t think there is or was a graphical convention on the rainbow. It is just the way of least resistance in most visualisation environments, i.e. the default. And Matlab already changed the default. For R changing the color scale means at the most a few lines of code – load package, define color palette, plot.


      Robert Wilson said:
      March 19, 2015 at 12:10 pm

      In many areas of science there is a de facto convention. The critics of rainbow need to come up with a new convention, instead of simply railing against the new one.

      For example, if I read a marine science paper and see a map, it is practically 99% certain it will use a rainbow scale. This mean that I can understand the map quickly. That’s the advantage of having graphical conventions. It saves the reader time. As far as I can tell, anything other than rainbow is probably going to be more difficult for the reader to understand.


        Arne Melsom said:
        March 19, 2015 at 1:12 pm

        I find using color scales based on two base colors to be useful in many applications, particularly when displaying anomalies or trends. And in such contexts my experience, or my very subjective view if you like, is that the rainbow is not well suited.

        My preference is for blue-red color scales, and in the cases I mention, blue and red naturally associates with negative and positive changes, respectively. A short report where you find a couple of examples of what I have in mind, is available from


        Robert Wilson said:
        March 19, 2015 at 1:29 pm


        For these cases I agree. That’s the way I do it in my thesis for positive and negative changes.

        Whether you should go beyond blue to red is debatable. Adding more colours in can provide more detail, but then it reduces clarity.

        But when all of the data is positive I’m not a fan of blue to red. One of the problems is that people may assume automatically that it is going from positive to negative, in particular if that’s the way it’s done in the rest of a paper/thesis/report. So, I would always aim for consistency.


    duffer70 said:
    March 19, 2015 at 12:31 pm

    A chap named Matteo Niccoli ( has put a fair bit of effort into critiquing rainbow more constructively, coming up with what he calls perceptual palettes that try to get the best of both brightness variation and multi-hue worlds. is a good place to start. I’ve started to incorporate some of his stuff (e.g. ‘cubeYF’) in my work.


    Chris Weijenborg said:
    June 8, 2015 at 7:04 pm

    I have not read the paper yet, but it is well known for a very long time that the rainbow palette is misleading and that it creates artificial data boundaries. Off course keeping conveniences is generally a good idea, but if a color palette leads to misinterpretation of your data, it should be replaced with a better one. A good post on this, which was written a few years ago by Robert Simmon from NASA earth observatory:
    There are alternatives, most notable R color brewer, which is also mentioned in the above series of blog posts. If you insist of on using a spectral palette, there is also a spectral palette in the “HCL” color space (see, and in the Colorbrewer package for R.


Comments are closed.