Color scales in R

Author

Andrew Siefert

A little color theory

Three components of color (HCL space):

Code

library(colorspace)

swatchplot(
  "Hue (type of color)" = hex(polarLUV(H = seq(0, 300, length.out = 10), C = 50, L = 70)),
  "Chroma (colorfulness)" = hex(polarLUV(H = 0, C = seq(0, 100, length.out = 10), L = 65)),
  "Luminance (brightness)"  = hex(polarLUV(H = 260, C = 25, L = seq(90, 10, length.out = 10)))
)

Captures human perceptual axes well
Perceptually uniform – a unit step anywhere in HCL space produces a constant perceptual change in color

colorspace package lets you create and explore palettes based on trajectories in HCL space.

Hue

Type of color (red, blue, green, etc.)
Not perceived as ordered
Values go from 0 to 360 (color wheel)

Code

library(colorspace)

hue <- polarLUV(H = seq(0, 300, 75), C = 60, L = 65)
hue

      L  C   H
[1,] 65 60   0
[2,] 65 60  75
[3,] 65 60 150
[4,] 65 60 225
[5,] 65 60 300

Code

hex(hue)

[1] "#DF8396" "#AF9E3F" "#28B37F" "#2FAAD0" "#C787D3"

Code

swatchplot(hex(hue))
text(x = seq(0.1, 1, 0.2), y = 0.05, labels = seq(0, 300, 75))

Chroma

Purity or colorfulness of the color
Perceived as ordered
Ranges from 0 (gray) to maximum that varies with hue and luminescence

Code

chroma <- polarLUV(H = 0, C = seq(0, 100, 25), L = 65) 
swatchplot(hex(chroma))
text(x = seq(0.1, 1, 0.2), y = 0.05, labels = seq(0, 100, 25))

Luminance

Brightness
Perceived as ordered
Ranges from 0 (black) to 100 (white)

Code

lum <- polarLUV(H = 260, C = 25, L = seq(90, 10, -20))
swatchplot(hex(lum))
text(x = seq(0.1, 1, 0.2), y = 0.05, labels = seq(90, 10, -20))

Type of palettes

Qualitative: categorical information
Sequential: ordered/numerical information (high to low)
Diverging: ordered/numeric information around a central neutral value

Qualitative palettes

Distinguishes categories by a sequence of hues while keeping chroma and luminance constant
Best to use equidistant sequence of hues spanning the color wheel
Use lighter colors (moderate chroma, high luminance) for shading areas (e.g., bar plots, maps)
Use more flashy colors (high chroma) for points and lines

Qualitative palettes in colorspace:

Code

hcl_palettes(type = "qualitative", plot = T)

Create a palette using qualitative_hcl():

Code

my_qual <- qualitative_hcl(6, palette = "Set3")
my_qual

[1] "#FFB3B5" "#DAC584" "#95D69A" "#61D8D6" "#AAC8FC" "#F1B2EE"

Code

demoplot(my_qual)

Code

# plot palette in HCL space
hclplot(my_qual)

Sequential palettes

Codes numeric values by sequence of increasing or decreasing luminance
Without chroma (c = 0), corresponds to grayscale
Use larger range of luminance if you are plotting more values
Can also vary chroma and/or hue to better distinguish colors

Code

hcl_palettes("sequential (single-hue)", n = 7, plot = TRUE, nrow = 6)

Code

hcl_palettes("sequential (multi-hue)", n = 7, plot = TRUE)

Code

seq9 <- sequential_hcl(10, palette = "Greens 3")
demoplot(seq9, "heatmap")

Code

hclplot(seq9)

Code

specplot(seq9, type = "o")

Diverging palettes

Combination of two sequential palettes with different hues in the left and right “arms”
Neutral central value has zero chroma
Chroma and luminance trajectories are mirrored in the two arms

Code

hcl_palettes("diverging", n = 7, plot = TRUE, nrow = 10)

Code

div12 <- diverging_hcl(12, palette = "Blue-Red")
demoplot(div12)

Code

hclplot(div12)

Code

specplot(div12, type = "o")

Using colorspace with ggplot2

Scales are named via the scheme:

scale_<aesthetic>_<datatype>_<colorscale>(palette = <Palette>)

<aesthetic> is the name of the aesthetic (fill, color, etc.)
<datatype> is the type of variable plotted (discrete or continuous)
<colorscale> sets the type of color scale used (qualitative, sequential, or diverging)
<Palette> is the name of the palette (e.g., "Dark 3", "Blue-Red")

Code

library(ggplot2)

ggplot(iris, aes(x = Sepal.Length, fill = Species)) + 
  geom_density(alpha = 0.6) + 
  scale_fill_discrete_qualitative(palette = "Dark 3")

Color blindness

colorblindr package simulates how a color palette will look to people with common forms of color blindness.

Make a figure using default ggplot color scale:

Code

df <- data.frame(x = c("a", "b", "c", "d"), y = c(3, 4, 1, 2))
bars <- ggplot(df, aes(x, y, fill = x)) + 
  geom_bar(stat = "identity") + 
  labs(x = NULL, y = NULL) +
  theme(legend.position = "none")
bars

Look at color-vision-deficiency simulations:

Code

library(colorblindr)

cvd_grid(bars)

A discrete color scale that works better for people with color blindness is the Okabe Ito palette.

Code

palette_OkabeIto

[1] "#E69F00" "#56B4E9" "#009E73" "#F0E442" "#0072B2" "#D55E00" "#CC79A7"
[8] "#999999"

Code

swatchplot(palette_OkabeIto)

Code

bars2 <- bars + scale_fill_OkabeIto()
bars2

Code

cvd_grid(bars2)

Some useful palettes

viridis

viridis color scales are designed to have good perceptual properties, work well for people with colorblindness, and print well in gray scale.

Code

library(viridis)

viridis(10)

 [1] "#440154FF" "#482878FF" "#3E4A89FF" "#31688EFF" "#26828EFF" "#1F9E89FF"
 [7] "#35B779FF" "#6DCD59FF" "#B4DE2CFF" "#FDE725FF"

Code

swatchplot(
  "viridis" = viridis(10),
  "magma" = viridis(10, option = "magma"),
  "inferno" = viridis(10, option = "inferno"), 
  "cividis" = viridis(10, option = "cividis"), 
  "rocket" = viridis(10, option = "rocket"),
  "mako" = viridis(10, option = "mako")
)

The viridis scales are available in ggplot2 using using scale_fill_viridis_d() and scale_fill_viridis_c().

Discrete viridis scale:

Code

bars + scale_fill_viridis_d()

Continuous viridis scale:

Code

erupt <- ggplot(faithfuld, aes(waiting, eruptions, fill = density)) +
  geom_raster()

erupt

Code

erupt + scale_fill_viridis_c(option = "magma")

ColorBrewer

ColorBrewer scales are designed to work well for maps, but are also good for other types of data. They are available in the RColorBrewer package:

Code

library(RColorBrewer)

display.brewer.all()

Most palettes are colorblind friendly.

Code

display.brewer.all(colorblindFriendly = T)

You can construct a palette using brewer.pal():

Code

brewer.pal(9, name = "PuBu")

[1] "#FFF7FB" "#ECE7F2" "#D0D1E6" "#A6BDDB" "#74A9CF" "#3690C0" "#0570B0"
[8] "#045A8D" "#023858"

Code

display.brewer.pal(10, name = "PuBu")

ColorBrewer scales are also built into ggplot2:

scale_color_brewer() and scale_fill_brewer() for discrete scales
scale_color_distiller() and scale_fill_distiller() for continuous scales

Code

bars + scale_fill_brewer(palette = "Set2")

Code

erupt + scale_fill_distiller(palette = "RdPu")

scico

scico includes 39 palettes designed for scientific data visualization. They have good perceptual properties and are colorblind safe.

Code

library(scico)

scico_palette_show()

Code

scico(20, palette = "lapaz") |> swatchplot()

Code

erupt + scale_fill_scico(palette = "davos")

Paletteer

The paletteer package provides a unified interface to color scales from many packages.

Code

library(paletteer)

paletteer_d("nationalparkcolors::BryceCanyon", 5)

<colors>
#E39B38FF #C7D8C8FF #B6BDCCFF #BFC4C5FF #9B593FFF

Code

paletteer_c("scico::devon", 10)

<colors>
#2C194CFF #283467FF #265085FF #3568ACFF #6181D0FF #979AE6FF #B9B2F1FF #D0CCF4FF #E7E4FAFF #FFFFFFFF

Code

bars + scale_fill_paletteer_d("wesanderson::FantasticFox1")

Manual scales

You can set colors manually in ggplot2 using scale_color_manual() and scale_fill_manual().

Code

bars + scale_fill_manual(values = c("forestgreen", "goldenrod", "firebrick", "cornflowerblue"))

Here are Erika’s cheat sheets for ColorBrewer scales:

Code

bars + scale_fill_manual(values = c("#7FC97F", "#BEAED4", "#FDC086", "#FFFF99"))