Colourblind-safe palettes for scientific figures

Around 1 in 12 men and 1 in 200 women have some form of colour vision deficiency. In a journal readership that is a large number of people, and several journals now check for this.

The combinations to avoid

Red and green together is the big one, affecting the most common forms of deficiency. Red and green lines on the same plot can be indistinguishable.

Green and brown, and light green with yellow, are also problematic.

Rainbow and jet colormaps are bad for a second reason as well: they are not perceptually uniform, so they create visual boundaries where the data are smooth and hide real transitions elsewhere. Journals increasingly discourage them regardless of colour vision.

What to use

Sequential data (magnitude, one direction): viridis, cividis, magma. Perceptually uniform, colourblind-safe, and they survive greyscale printing.

Diverging data (correlations, differences, anything with a meaningful midpoint): coolwarm, RdBu, or PuOr. Centre them on zero.

Categorical data: Okabe-Ito, an eight-colour palette designed specifically for colour vision deficiency. Paul Tol's qualitative palettes are also good.

Do not rely on colour alone

The most robust fix is redundant encoding. Give each series a different line style or marker shape as well as a different colour. Label lines directly instead of using a legend the reader must map back by colour.

Then your figure works in colour, in greyscale, and for every reader.

Check it

Convert your figure to greyscale. If you cannot tell the series apart, neither can a substantial part of your audience. Colourblindness simulators are also freely available online.

In Plotara

Viridis and cividis for sequential data, diverging scales by default for correlation data, and marker shape and line style available as independent encodings.