Why Rainbow Colour Scales Mislead — a Python Follow-up to Tony Ladson’s Post

Fourth in an occasional series drawing on Tony Ladson’s blog — his explanation and the underlying argument are his (and, one step further back, Ed Hawkins’); I’m providing a Python-side follow-up for readers who’d rather see it measured than just told. Source: Rainbow colour scales in hydrologic maps and charts, drawing on Ed Hawkins’ Scrap rainbow colour scales (Nature, 2015).

“Don’t use jet/rainbow colourmaps” is good advice that usually arrives as a style preference. It isn’t one — it’s a measurable property of the colourmap, independent of whatever data you plot with it.

Measuring it

A well-behaved colourmap should have perceptual lightness change monotonically across its range — consistently light-to-dark or dark-to-light. If lightness goes up, then down, then up again as the underlying value increases smoothly, the eye reads false boundaries at each light/dark transition, whether or not the data has a boundary there.

import numpy as np
import matplotlib.pyplot as plt
from colorspacious import cspace_convert  # pip install colorspacious

def colormap_lightness(cmap_name, n=256):
    cmap = plt.get_cmap(cmap_name, n)
    rgb = cmap(np.linspace(0, 1, n))[:, :3]
    lab = cspace_convert(rgb, 'sRGB1', 'CIELab')
    return lab[:, 0]  # L* channel

for name in ['jet', 'viridis', 'cividis']:
    L = colormap_lightness(name)
    direction_changes = np.sum(np.diff(np.sign(np.diff(L))) != 0)
    print(f'{name:10s} direction changes = {direction_changes}')

jet        direction changes = 5
viridis    direction changes = 0
cividis    direction changes = 0

Perceptual lightness (CIE L*) across each colourmap. jet rises, falls, and rises again — genuinely non-monotonic. viridis and cividis are both smooth and monotonic (and nearly identical to each other).

This is the same library (colorspacious) used to justify viridis’s design in the first place — not a new claim, just independently re-measured.

What that looks like on real-shaped data

The clearest version of this doesn’t need real data at all — a single smooth Gaussian peak, with zero genuine edges anywhere, makes the point on its own:

Same smooth field, two colourmaps. jet (left) shows what reads as a distinct yellow-green ring and a separate dark red plateau at the centre. Neither exists in the data — Z is one continuous gradient in every direction. viridis (right) shows the field as what it actually is.

Nothing about the underlying field changed between those two panels. Only the colourmap did.

The other half of the lesson

Ed Hawkins’ warming stripes make a different, complementary point that’s worth keeping separate from “pick a better colourmap”: sometimes the right fix isn’t a better colourmap at all, it’s asking whether the reader needs axes, gridlines and a legend, or whether colour alone — no other chart furniture — communicates the pattern more directly. That’s a genuinely different design decision, and I’d rather point you to Hawkins’ own explanation and the real visualisation than build an approximation of it here with data I couldn’t verify.

---

Companion notebook: notebooks/11_rainbow_colour_scales/

References:

• Ladson, A.R. (2016). Rainbow colour scales in hydrologic maps and charts.
• Hawkins, E. (2015). Scrap rainbow colour scales. Nature 519, 291.
• Hawkins, E. #ShowYourStripes.