secondary axis is not working properly with geom_col

Question

I am trying to use help_secondary of ggh4x with geom_col like

library(tidyverse)
library(ggh4x)
library(scales) 

# Run the secondary axis helper
sec <- help_secondary(df, primary = c(Tmax, Tmin, RH), 
                      secondary = Rainfall)

ggplot(df, aes(x = Date)) +
  geom_col(aes(y = sec$proj(Rainfall)), colour = "blue") +
  geom_line(aes(y = Tmax), colour = "red") +
  geom_line(aes(y = Tmin), colour = "green") +
  geom_line(aes(y = RH), colour = "black") +
  scale_y_continuous(sec.axis = sec) +
  ylab("Max and Min temperature, RH")

As you can see from the plot, the secondary y-axis starts at -500. The 0 rainfall value is shown as -500 which ideally should be left out in the barplot.

Another thing, how can I have the legend?

Data

df = structure(list(Date = structure(c(18628, 18659, 18687, 18718, 
18748, 18779, 18809, 18840, 18871, 18901, 18932, 18962, 18993, 
19024, 19052, 19083, 19113, 19144, 19174, 19205, 19236, 19266, 
19297, 19327, 19358, 19389, 19417, 19448, 19478, 19509, 19539
), class = "Date"), Tmax = c(34.3774193548387, 34.8428571428571, 
35.7387096774194, 35.44, 34.1161290322581, 30.9333333333333, 
29.2193548387097, 29.9225806451613, 29.78, 32.4096774193548, 
32.97, 32.4129032258065, 32.2225806451613, 33.9428571428571, 
35.6, 35.2133333333333, 33.658064516129, 31.0566666666667, 29.4516129032258, 
29.6548387096774, 30.2066666666667, 32.4870967741935, 34.2566666666667, 
34.558064516129, 33.8161290322581, 36.3821428571429, 35.9483870967742, 
35.2266666666667, 35.9709677419355, 33.04, 28.6322580645161), 
    Tmin = c(21.5483870967742, 20.4714285714286, 24.1483870967742, 
    24.41, 25.2354838709677, 23.3966666666667, 23.2129032258064, 
    23.7935483870968, 24.6633333333333, 24.0225806451613, 23.3466666666667, 
    20.8516129032258, 19.3806451612903, 19.5, 23.558064516129, 
    24.9533333333333, 25.8741935483871, 23.92, 23.3709677419355, 
    22.8677419354839, 22.5433333333333, 21.7161290322581, 21.2133333333333, 
    21.0258064516129, 19.3612903225807, 20.6142857142857, 21.641935483871, 
    24.5343333333333, 26.0903225806452, 25.4443333333333, 23.7645161290323
    ), RH = c(62.1129032258064, 57.5892857142858, 70.4193548387097, 
    68.25, 74.8387096774194, 85.3333333333333, 88.5161290322581, 
    85.8870967741936, 86.4333333333334, 79.1129032258064, 74.6, 
    69.1935483870968, 62.6290322580645, 65.5892857142858, 66.3548387096774, 
    70.7166666666667, 75.7741935483871, 85.6833333333334, 88.3387096774194, 
    86.8064516129033, 84.1166666666667, 74.3870967741936, 64.4, 
    63.7903225806452, 60.1290322580645, 58.7321428571429, 58.4677419354839, 
    69.3, 65.9193548387097, 79.45, 91.6451612903226), Rainfall = c(9.1, 
    2, 0, 27, 422.6, 903.9, 1345.9, 343.9, 481, 239.9, 166.4, 
    105.9, 0, 0, 0.3, 51.2, 99.6, 700.6, 1098.3, 347.6, 276.8, 
    71.3, 0, 2.2, 0, 0, 0, 0, 0, 597, 1763.4)), row.names = c(NA, 
31L), class = "data.frame")

Answer 1

The issue is that by default help_secondary will rescale the based on the range of the variables mapped on the primary and the secondary axis, i.e. the primary minimum is mapped on the secondary minimum.

However, as you are using a geom_col the range or the limits of the y scale are extended to include 0. This is not and can't be accounted for by help_secondary.

For your case, where the minimum of the secondary variable is zero, you could fix that by switching to method="max" in help_secondary.

To get a legend, you have to map on aesthetics, i.e. move color= inside aes() and set your colors via scale_color_manual.

library(ggplot2)
library(ggh4x)
library(scales)

sec <- help_secondary(df,
  primary = c(Tmax, Tmin, RH),
  secondary = Rainfall,
  method = "max"
)

ggplot(df, aes(x = Date)) +
  geom_col(aes(y = sec$proj(Rainfall), color = "rain")) +
  geom_line(aes(y = Tmax, color = "tmax")) +
  geom_line(aes(y = Tmin, color = "tmin")) +
  geom_line(aes(y = RH, color = "rh")) +
  scale_color_manual(
    values = c(
      tmin = "green", tmax = "red",
      rh = "black", rain = "blue"
    )
  ) +
  scale_y_continuous(sec.axis = sec) +
  ylab("Max and Min temperature, RH")

EDIT To fix the legend, map on the fill aes to get filled bars, apply the color scale both on the color and fill aes by adding aesthetics = c("fill", "color"), set the same name for both scales (so that both legends get merged) and tweak the legend via the override.aes argument of guide_legend, i.e. remove the legend fill for the lines:

ggplot(df, aes(x = Date)) +
  geom_col(aes(y = sec$proj(Rainfall), fill = "rain")) +
  geom_line(aes(y = Tmax, color = "tmax")) +
  geom_line(aes(y = Tmin, color = "tmin")) +
  geom_line(aes(y = RH, color = "rh")) +
  scale_color_manual(
    values = c(
      tmin = "green", tmax = "red",
      rh = "black", rain = "blue"
    ),
    aesthetics = c("fill", "color")
  ) +
  scale_y_continuous(sec.axis = sec) +
  labs(y = "Max and Min temperature, RH", color = NULL, fill = NULL) +
  guides(color = guide_legend(
    override.aes = list(fill = c("blue", rep("transparent", 3)))
  ))

Answer 2

Here's one way:

ggplot(df, aes(x = Date)) +
  geom_col(aes(y = Rainfall * 0.1), color = "blue") +
  geom_line(aes(y = Tmax), color = "red") +
  geom_line(aes(y = Tmin), color = "green") +
  geom_line(aes(y = RH), color = "black") +
  scale_y_continuous(
    name = "Max and Min temperature, RH",
    sec.axis = sec_axis(~ . / 0.1, name = "Rainfall"))

Update: I had a play around with help_secondary. It has a bunch of methods of projecting the secondary axis data:

"range" Causes the ranges of primary and secondary data to overlap completely.

What appears to be the default

"max" Causes the maxima of primary and secondary data to coincide.

Gets the right answer

"fit" Uses the coefficients of lm(primary ~ secondary) to make the axes fit.

Throws an error in this case

"ccf" Uses the lag at which maximum cross-correlation occurs to then align the data by truncation. The aligned data is then passed to the "fit" method.

Also throws an error

"sortfit" Sorts the both primary and secondary independently before passing these on to the "fit" method.

Throws an error too.

So changing sec to be sec <- help_secondary(df, primary = c(Tmax, Tmin, RH), secondary = Rainfall, method = "max"), then the code should work as expected. See the documentation for more information.