Distinct scale_fill_gradient for each group in ggplot2

Question

I plot several densities of values corresponding to categories of a discrete variable. I can associate to each density a specific color or a gradient of color for all of them at the same time. Now I would like to add a specific gradient for each density with varying values.

Here is a reproducible example using ggridges:

data(iris)

library(ggplot2)
library(ggridges)
library(RColorBrewer)

cols <- brewer.pal(3, "BrBG")

# Plot with one color per group
ggplot(iris, aes(Sepal.Length, as.factor(Species))) +
  geom_density_ridges(aes(fill = as.factor(Species))) +
  scale_fill_manual("Sepal", values = cols)

# Plot with one gradient
ggplot(iris, aes(Sepal.Length, as.factor(Species))) +
  geom_density_ridges_gradient(aes(fill = ..x..)) +
  scale_fill_gradient2(low = "grey", high = cols[1], midpoint = 5)

I basically want to combine both plots. I am also interested in having a specific midpoint value for each density.

Answer 1

I came up with the workaround below as a curiosity, but I don't think it's really good practice, as far as data visualisation goes. Having a single varying gradient in a density chart is shaky enough; having multiple different ones won't be any better. Please don't use it.

Preparation:

ggplot(iris, aes(Sepal.Length, as.factor(Species))) +
  geom_density_ridges_gradient()
# plot normally & read off the joint bandwidth from the console message (0.181 in this case)

# split data based on the group variable, & define desired gradient colours / midpoints
# in the same sequential order.
split.data <- split(iris, iris$Species)
split.grad.low <- c("blue", "red", "yellow") # for illustration; please use prettier colours
split.grad.high <- cols
split.grad.midpt <- c(4.5, 6.5, 7) # for illustration; please use more sensible points

# create a separate plot for each group of data, specifying the joint bandwidth from the
# full chart.
split.plot <- lapply(seq_along(split.data),
                     function(i) ggplot(split.data[[i]], aes(Sepal.Length, Species)) +
                       geom_density_ridges_gradient(aes(fill = ..x..), 
                                                    bandwidth = 0.181) +
                       scale_fill_gradient2(low = split.grad.low[i], high = split.grad.high[i],
                                            midpoint = split.grad.midpt[i]))

Plot:

# Use layer_data() on each plot to get the calculated values for x / y / fill / etc,,
# & create two geom layers from each, one for the gradient fill & one for the ridgeline 
# on top. Add them to a new ggplot() object in reversed order, because we want the last
# group to be at the bottom, overlaid by the others where applicable.
ggplot() +
  lapply(rev(seq_along(split.data)),
         function(i) layer_data(split.plot[[i]]) %>%
           mutate(xmin = x, xmax = lead(x), ymin = ymin + i - 1, ymax = ymax + i - 1) %>%
           select(xmin, xmax, ymin, ymax, height, fill) %>%
           mutate(sequence = i) %>%
           na.omit() %>%
           {list(geom_rect(data = ., 
                           aes(xmin = xmin, xmax = xmax, ymin = ymin, ymax = ymax, fill = fill)),
                 geom_line(data = .,
                           aes(x = xmin, y = ymax)))}) +

  # Label the y-axis labels based on the original data's group variable
  scale_y_continuous(breaks = seq_along(split.data), labels = names(split.data)) +

  # Use scale_fill_identity, since all the fill values have already been calculated.
  scale_fill_identity() +
  labs(x = "Sepal Length", y = "Species")

Note that this method won't create a fill legend. If desired, it's possible to retrieve the fill legends from the respective plots in split.plot via get_legend & add them to the plot above via plot_grid (both functions from the cowplot package), but that's like adding frills to an already weird visualisation choice...