Perceptually-Uniform, Colorblind-Friendly Colormaps

Perceptually-Uniform, Colorblind-Friendly Colormaps#

Avoid rainbow/jet#

  • Not perceptually uniform: equal numeric steps don’t look like equal visual steps; non-monotonic lightness introduces false “edges” that can mislead interpretation.

  • Accessibility issues: rainbow palettes map distinct hues that many viewers with color-vision deficiencies (approx. 1 in 12 men) cannot reliably differentiate.

  • Context switching: abrupt hue jumps make reading gradients and subtle trends harder, especially when printed or viewed on projectors.

What to use instead#

Choose colormaps designed for uniform perceived change (usually monotonic lightness) and CVD robustness:

  • Sequential (low > high): e.g., Matplotlib’s viridis, magma, plasma, inferno; cmocean’s domain-aware sequences (e.g., cmo.thermal for temperature, cmo.haline for salinity).

  • Diverging (midpoint emphasis): use when values deviate around a meaningful center (0, climatology, etc.). Examples: Matplotlib’s seismic-style but perceptually tuned options like coolwarm (still imperfect) or cmocean’s balance, delta, curl, which are engineered for symmetry and lightness control.

  • Cyclic (wrap-around variables): for phase/aspect (0°≡360°). Use cyclic maps such as cmocean’s phase.

  • Categorical (discrete classes): use distinct, desaturated palettes with good lightness separation; avoid “rainbow” categories for quantitative data.

Picking the right map for your data#

  • Monotonic data: sequential.

  • Signed anomalies around a reference: diverging with a clearly defined, perceptually central midpoint.

  • Angles/orientations: cyclic so the endpoints match.

  • Dynamic range: ensure the lightness ramp spans the range where your audience needs discrimination (you can trim/clip the colormap range if needed).

  • Background: pick a map whose lightness contrasts with the figure background (dark maps on dark backgrounds obscure low values).

Quick recipes (Matplotlib + cmocean)#

Install cmocean (Matplotlib port):

pip install cmocean

Set a global default and plot:

import matplotlib.pyplot as plt
import numpy as np
import cmocean

# Set a perceptually-uniform default
plt.rcParams["image.cmap"] = "viridis"

# Example data
x = np.linspace(-3, 3, 400)
y = np.linspace(-3, 3, 400)
X, Y = np.meshgrid(x, y)
Z = np.hypot(X, Y)

# Sequential (distance field)
plt.imshow(Z, origin="lower", cmap=cmocean.cm.thermal)
plt.colorbar(label="Temperature-like quantity")
plt.title("Sequential, perceptually-uniform")
plt.show()

# Diverging (positive/negative anomaly)
Z_anom = np.sin(X) * np.cos(Y)
plt.imshow(Z_anom, origin="lower", cmap=cmocean.cm.balance, vmin=-1, vmax=1)
plt.colorbar(label="Anomaly")
plt.title("Diverging around 0")
plt.show()

# Cyclic (phase)
Z_phase = np.angle(np.exp(1j*(X)))
plt.imshow(Z_phase, origin="lower", cmap=cmocean.cm.phase)
plt.colorbar(label="Phase [rad]")
plt.title("Cyclic for wrap-around variables")
plt.show()

More cmocean info: matplotlib.org/cmocean

Make figures colorblind-safe (Linux + GNOME)#

A handy way to simulate common color-vision deficiencies directly on your screen is the GNOME Shell extension Colorblind Filters. It applies real-time filters so you can preview how your plots look under Deuteranopia, Protanopia, Tritanopia, etc.

  • Extension: G-dH/gnome-colorblind-filters

  • Workflow: generate your plots > toggle the relevant filter > adjust colormap/range/line styles until the figure remains readable.

Best-practice checklist#

  • Use perceptually uniform maps with monotonic lightness for quantitative gradients.

  • Match map type to data type (sequential/diverging/cyclic).

  • Ensure contrast and legible colorbar ticks/labels; set meaningful limits (vmin/vmax, centered diverging scales).

  • Verify accessibility with CVD simulation; don’t rely on color alone—add contours, annotations, or varying line styles where helpful.

  • Be consistent across panels and publications; document the colormap choice in captions (e.g., “cmocean ‘balance’”).

Contents