Fill region between two loess-smoothed lines in R with ggplot

Question

I'd like to know how to fill the area between to loess-smoothed lines in ggplot.

The following data frame is used for the pictures:

    x         y      ymin     ymax grp     ydiff
1   1  3.285614  3.285614 10.14177 min 6.8561586
2   1 10.141773  3.285614 10.14177 max 6.8561586
3   2  5.061879  5.061879 11.24462 min 6.1827368
4   2 11.244615  5.061879 11.24462 max 6.1827368
5   3  8.614408  8.614408 13.45030 min 4.8358931
6   3 13.450301  8.614408 13.45030 max 4.8358931
7   4  6.838143  6.838143 12.34746 min 5.5093150
8   4 12.347458  6.838143 12.34746 max 5.5093150
9   5 10.390673 10.390673 14.55314 min 4.1624713
10  5 14.553144 10.390673 14.55314 max 4.1624713
11  6 12.166937 12.166937 15.65599 min 3.4890495
12  6 15.655987 12.166937 15.65599 max 3.4890495
13  7 13.943202 13.943202 16.75883 min 2.8156277
14  7 16.758830 13.943202 16.75883 max 2.8156277
15  8  5.950011  5.950011 11.79604 min 5.8460259
16  8 11.796037  5.950011 11.79604 max 5.8460259
17  9 17.495731 17.495731 18.96452 min 1.4687841
18  9 18.964515 17.495731 18.96452 max 1.4687841
19 10 15.719466 15.719466 17.86167 min 2.1422059
20 10 17.861672 15.719466 17.86167 max 2.1422059
21 11 19.271996 19.271996 20.06736 min 0.7953623
22 11 20.067358 19.271996 20.06736 max 0.7953623

Following source produces a figure with normal lines (not smoothed):

ggplot(intdf) + 
    geom_point(aes(x=x, y=y, colour=grp)) +
    geom_ribbon(aes(x=x, ymin=ymin, ymax=ymax), fill="grey", alpha=.4) +
    geom_line(aes(x=x, y=y, colour=grp))

where x and y are continuous numeric values. ymin and ymax each contain the y-values from the green and red line at position x.

Now I'd like to smoothen the lines. I simply do this with following code:

ggplot(intdf) + 
    stat_smooth(aes(x=x, y=ymin, colour="min"), method="loess", se=FALSE) +
    stat_smooth(aes(x=x, y=ymax, colour="max"), method="loess", se=FALSE)

which gives following plot:

But I did not manage to fill the area between these two lines. I tried to fit a loess-model and use the predicted values, but I guess I completely used the wrong predictors.

Who could tell me how to fill the region between the smoothed lines?

Thanks in advance Daniel

Answer 1

A possible solution where the loess smoothed data is grabbed from the plot object and used for the geom_ribbon:

# create plot object with loess regression lines
g1 <- ggplot(df) + 
  stat_smooth(aes(x = x, y = ymin, colour = "min"), method = "loess", se = FALSE) +
  stat_smooth(aes(x = x, y = ymax, colour = "max"), method = "loess", se = FALSE)
g1

# build plot object for rendering 
gg1 <- ggplot_build(g1)

# extract data for the loess lines from the 'data' slot
df2 <- data.frame(x = gg1$data[[1]]$x,
                  ymin = gg1$data[[1]]$y,
                  ymax = gg1$data[[2]]$y) 

# use the loess data to add the 'ribbon' to plot 
g1 +
  geom_ribbon(data = df2, aes(x = x, ymin = ymin, ymax = ymax),
              fill = "grey", alpha = 0.4)