Axis labels on two lines with nested x variables (year below months)

The code below provides two potential options for adding year labels.

Option 1a: Faceting

You could use faceting to mark the years. For example:

library(ggplot2)
library(lubridate)

ggplot(df, aes(Date, value)) +
  geom_line() +
  scale_x_date(date_labels="%b", date_breaks="month", expand=c(0,0)) +
  facet_grid(~ year(Date), space="free_x", scales="free_x", switch="x") +
  theme_bw() +
  theme(strip.placement = "outside",
        strip.background = element_rect(fill=NA,colour="grey50"),
        panel.spacing=unit(0,"cm"))

Note that with this approach, if there are missing dates at the beginning or end of a year (by "missing", I mean rows for those dates are not even present in the data) then the x-axis will start/end at the first/last date in the data for that year, rather than go from Jan-1 to Dec-31. In that case, you'd need to add in rows for the missing dates and either NA for value or interpolate value. In addition, with this method there is no space or line between December 31 of one year and January 1 of the next year, so there's a discontinuity across each year.

Option 1b: Faceting + centered month labels

To address @AF7's comment. You can center the month labels by adding some spaces before each label. But you have to choose the number of spaces manually, depending on the physical size of the plot when you print it to a device. (There's probably a way to center the labels programmatically based on the internal grob measurements, but I'm not sure how to do it.) I've also removed the minor vertical gridlines and lightened the line between years.

ggplot(df, aes(Date, value)) +
  geom_line() +
  scale_x_date(date_labels=paste(c(rep(" ",11), "%b"), collapse=""), 
               date_breaks="month", expand=c(0,0)) +
  facet_grid(~ year(Date), space="free_x", scales="free_x", switch="x") +
  theme_bw() +
  theme(strip.placement = "outside",
        strip.background = element_blank(),
        panel.grid.minor.x = element_blank(),
        panel.border = element_rect(colour="grey70"),
        panel.spacing=unit(0,"cm"))

Option 2a: Edit the x-axis label grob

Here's a more complex and finicky method (though it could likely be automated by someone who understands the structure and unit spacings of grid graphics better than I do) that avoids the pitfalls of the faceting method described above:

library(grid)

# Fake data with an extra year added for illustration
set.seed(2)
df = data.frame(Date=seq(as.Date("1718-03-01"),as.Date("1721-09-20"), by="1 day"))
df$value = cumsum(rnorm(nrow(df)))

# The plot we'll start with
p = ggplot(df, aes(Date, value)) +
  geom_vline(xintercept=as.numeric(df$Date[yday(df$Date)==1]), colour="grey60") +
  geom_line() +
  scale_x_date(date_labels="%b", date_breaks="month", expand=c(0,0)) +
  theme_bw() +
  theme(panel.grid.minor.x = element_blank()) +
  labs(x="")

Now we want to add the year values below and in between June and July of each year. The code below does that by modifying the x-axis label grob and is adapted from this SO answer by @SandyMuspratt.

# Get the grob
g <- ggplotGrob(p)

# Get the y axis
index <- which(g$layout$name == "axis-b")  # Which grob
xaxis <- g$grobs[[index]]   

# Get the ticks (labels and marks)
ticks <- xaxis$children[[2]]

# Get the labels
ticksB <- ticks$grobs[[2]]

# Edit x-axis label grob
# Find every index of Jun in the x-axis labels and add a newline and
# then a year label
junes = which(ticksB$children[[1]]$label == "Jun")
ticksB$children[[1]]$label[junes] = paste0(ticksB$children[[1]]$label[junes],
                                           "\n      ", unique(year(df$Date))) 

# Put the edited labels back into the plot
ticks$grobs[[2]] <- ticksB
xaxis$children[[2]] <- ticks
g$grobs[[index]] <- xaxis

# Draw the plot
grid.newpage()
grid.draw(g)

Option 2b: Edit the x-axis label grob and center the month labels

Below is the only change that needs to be made to Option 2a to center the month labels, but, once again, the number of spaces needs to be tweaked manually.

# Make the edit
# Center the month labels between ticks
ticksB$children[[1]]$label = paste0(paste(rep(" ",7),collapse=""), ticksB$children[[1]]$label)

# Find every index of Jun in the x-axis labels and a year label
junes = grep("Jun", ticksB$children[[1]]$label)
ticksB$children[[1]]$label[junes] = paste0(ticksB$children[[1]]$label[junes], "\n      ", unique(year(df$Date))) 

I came upon this question and thought maybe I can add a solution. We can display both month and year in every year's first displayed month by using a simple condition. You can play with the date_breaks to remove January from the labels, and this will still work. I'm using month() and year() from lubridate.

library(tidyverse)
library(lubridate)

df %>% 
   ggplot(aes(Date, value)) +
   geom_line() +
   scale_x_date(date_breaks = "2 months", 
                labels = function(x) if_else(is.na(lag(x)) | !year(lag(x)) == year(x), 
                                             paste(month(x, label = TRUE), "\n", year(x)), 
                                             paste(month(x, label = TRUE))))

If you want to try to hack together a sub-label, you could convert it to a grob. I edited this from the original post to create a function that adds the sublabels and returns a gtable object. Note that the sublabs input must be the same length as your x-axis breaks:

library(grid)
library(gtable)
library(gridExtra)

add_sublabs <- function(plot, sublabs){

  gg <- ggplotGrob(plot)

  axis_num <- which(gg$layout[,"name"] == "axis-b")

  xbreaks <- gg[["grobs"]][[axis_num]][["children"]][[2]][["grobs"]][[2]][["children"]][[1]]$x
  if(length(xbreaks) != length(sublabs)) stop("Sub-labels must be the same length as the x-axis breaks")

  to_breaks <- c(as.numeric(xbreaks),1)[which(!duplicated(sublabs, fromLast = TRUE))+1]
  sublabs_x <- diff(c(0,to_breaks))
  sublabs_labels <- sublabs[!duplicated(sublabs, fromLast = TRUE)]

  tg <- tableGrob(matrix(sublabs_labels, nrow = 1))
  tg$widths = unit(sublabs_x, attr(xbreaks,"unit"))

  pos <- gg$layout[axis_num,c("t","l")]

  gg2 <- gtable_add_rows(gg, heights = sum(tg$heights)+unit(4,"mm"), pos = pos$t)
  gg3 <- gtable_add_grob(gg2, tg, t = pos$t+1, l = pos$l)

  return(gg3)
}


#Plot and sublabels
p <- ggplot(data = df, aes(x = Date, y = value)) + geom_line() +
  scale_x_date(date_breaks = "2 month", date_minor_breaks = "1 month", labels = date_format("%b")) +
  xlab(NULL)
sublabs <- c(rep("1719",2),rep("1720",6))

#Draw
grid.draw(add_sublabs(p, sublabs))

One way to avoid the complexities would be to change the required output so that January is replaced by the year.

The lab function returns the labels given the breaks. Unexpectedly, ggplot will pass NAs to it so in the first line of the function body we replace those with some date -- it does not matter which date since such values are not subsequently used by ggplot. Finally we format the date as a year or abbreviated month depending on whether the month is January (which corresponds to the POSIXlt component mon equalling 0) or not.

library(ggplot2)
library(scales)

lab <- function(b) {
  b[is.na(b)] <- Sys.Date()
  format(b, ifelse(as.POSIXlt(b)$mon == 0, "%Y", "%b"))
}

ggplot(df, aes(Date, value)) + 
   geom_line() +
   scale_x_date(date_breaks = "month", labels = lab)

Note: I have added Issue 2182 to the ggplot2 github issues list regarding the NAs that are passed to the label function. If subsequent versions of ggplot2 no longer pass the NAs then the first line of the body of lab could be omitted .

Update: fixed.